You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by tq...@apache.org on 2023/01/21 02:15:21 UTC
[tvm-site] branch asf-site updated: deploying docs (apache/tvm@d907de3b358e3b0052e0c39b71e65f4a509e6459)
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/tvm-site.git
The following commit(s) were added to refs/heads/asf-site by this push:
new 7b485e7b24 deploying docs (apache/tvm@d907de3b358e3b0052e0c39b71e65f4a509e6459)
7b485e7b24 is described below
commit 7b485e7b244647a47368b8d2aa114e531a7ae93a
Author: tvm-bot <95...@users.noreply.github.com>
AuthorDate: Sat Jan 21 02:15:15 2023 +0000
deploying docs (apache/tvm@d907de3b358e3b0052e0c39b71e65f4a509e6459)
---
docs/_images/sphx_glr_micro_train_001.png | Bin 302810 -> 322817 bytes
docs/_images/sphx_glr_micro_train_thumb.png | Bin 22617 -> 23407 bytes
.../how_to/compile_models/from_darknet.rst.txt | 2 +-
.../how_to/compile_models/from_keras.rst.txt | 2 +-
.../how_to/compile_models/from_mxnet.rst.txt | 2 +-
.../how_to/compile_models/from_oneflow.rst.txt | 2 +-
.../how_to/compile_models/from_pytorch.rst.txt | 2 +-
.../how_to/compile_models/from_tensorflow.rst.txt | 2 +-
.../compile_models/sg_execution_times.rst.txt | 22 +-
.../deploy_models/deploy_model_on_adreno.rst.txt | 4 +-
.../deploy_models/deploy_model_on_android.rst.txt | 2 +-
.../deploy_object_detection_pytorch.rst.txt | 4 +-
.../deploy_models/deploy_prequantized.rst.txt | 6 +-
.../deploy_prequantized_tflite.rst.txt | 4 +-
.../how_to/deploy_models/deploy_quantized.rst.txt | 2 +-
.../deploy_models/deploy_ssd_gluoncv.rst.txt | 4 +-
.../deploy_models/sg_execution_times.rst.txt | 20 +-
.../extend_tvm/bring_your_own_datatypes.rst.txt | 2 +-
.../how_to/extend_tvm/sg_execution_times.rst.txt | 8 +-
.../how_to/extend_tvm/use_pass_instrument.rst.txt | 16 +-
.../optimize_operators/opt_conv_cuda.rst.txt | 2 +-
.../optimize_operators/opt_conv_tensorcore.rst.txt | 2 +-
.../how_to/optimize_operators/opt_gemm.rst.txt | 16 +-
.../optimize_operators/sg_execution_times.rst.txt | 8 +-
.../sg_execution_times.rst.txt | 14 +-
.../tune_conv2d_layer_cuda.rst.txt | 1992 +++++++++++---------
.../tune_network_cuda.rst.txt | 4 +-
.../tune_network_x86.rst.txt | 4 +-
.../tune_sparse_x86.rst.txt | 85 +-
.../tune_with_autotvm/sg_execution_times.rst.txt | 8 +-
.../tune_with_autotvm/tune_conv2d_cuda.rst.txt | 1057 ++++++++++-
.../work_with_microtvm/micro_autotune.rst.txt | 16 +-
.../work_with_microtvm/micro_pytorch.rst.txt | 4 +-
.../how_to/work_with_microtvm/micro_train.rst.txt | 18 +-
.../work_with_microtvm/sg_execution_times.rst.txt | 12 +-
.../work_with_relay/sg_execution_times.rst.txt | 10 +-
.../how_to/work_with_schedules/intrin_math.rst.txt | 2 +-
.../work_with_schedules/sg_execution_times.rst.txt | 18 +-
.../how_to/work_with_schedules/tensorize.rst.txt | 2 +-
.../tutorials/autotvm/sg_execution_times.rst.txt | 4 +-
.../frontend/deploy_classification.rst.txt | 2 +-
.../tutorials/frontend/deploy_detection.rst.txt | 2 +-
.../tutorials/frontend/sg_execution_times.rst.txt | 6 +-
.../tutorials/optimize/sg_execution_times.rst.txt | 6 +-
.../topic/vta/tutorials/sg_execution_times.rst.txt | 6 +-
.../tutorial/auto_scheduler_matmul_x86.rst.txt | 6 +-
docs/_sources/tutorial/autotvm_matmul_x86.rst.txt | 20 +-
docs/_sources/tutorial/autotvm_relay_x86.rst.txt | 58 +-
.../tutorial/cross_compilation_and_rpc.rst.txt | 2 +-
docs/_sources/tutorial/intro_topi.rst.txt | 2 +-
docs/_sources/tutorial/sg_execution_times.rst.txt | 22 +-
.../tutorial/tensor_expr_get_started.rst.txt | 45 +-
docs/commit_hash | 2 +-
docs/how_to/compile_models/from_darknet.html | 2 +-
docs/how_to/compile_models/from_keras.html | 2 +-
docs/how_to/compile_models/from_mxnet.html | 2 +-
docs/how_to/compile_models/from_oneflow.html | 12 +-
docs/how_to/compile_models/from_pytorch.html | 8 +-
docs/how_to/compile_models/from_tensorflow.html | 2 +-
docs/how_to/compile_models/sg_execution_times.html | 26 +-
.../deploy_models/deploy_model_on_adreno.html | 4 +-
.../deploy_models/deploy_model_on_android.html | 2 +-
.../deploy_object_detection_pytorch.html | 41 +-
docs/how_to/deploy_models/deploy_prequantized.html | 8 +-
.../deploy_models/deploy_prequantized_tflite.html | 4 +-
docs/how_to/deploy_models/deploy_quantized.html | 2 +-
docs/how_to/deploy_models/deploy_ssd_gluoncv.html | 35 +-
docs/how_to/deploy_models/sg_execution_times.html | 20 +-
.../extend_tvm/bring_your_own_datatypes.html | 2 +-
docs/how_to/extend_tvm/sg_execution_times.html | 8 +-
docs/how_to/extend_tvm/use_pass_instrument.html | 16 +-
docs/how_to/optimize_operators/opt_conv_cuda.html | 2 +-
.../optimize_operators/opt_conv_tensorcore.html | 2 +-
docs/how_to/optimize_operators/opt_gemm.html | 16 +-
.../optimize_operators/sg_execution_times.html | 8 +-
.../sg_execution_times.html | 14 +-
.../tune_conv2d_layer_cuda.html | 1987 ++++++++++---------
.../tune_with_autoscheduler/tune_network_cuda.html | 4 +-
.../tune_with_autoscheduler/tune_network_x86.html | 4 +-
.../tune_with_autoscheduler/tune_sparse_x86.html | 85 +-
.../tune_with_autotvm/sg_execution_times.html | 8 +-
.../how_to/tune_with_autotvm/tune_conv2d_cuda.html | 1057 ++++++++++-
docs/how_to/work_with_microtvm/micro_autotune.html | 16 +-
docs/how_to/work_with_microtvm/micro_pytorch.html | 4 +-
docs/how_to/work_with_microtvm/micro_train.html | 16 +-
.../work_with_microtvm/sg_execution_times.html | 12 +-
.../how_to/work_with_relay/sg_execution_times.html | 10 +-
docs/how_to/work_with_schedules/intrin_math.html | 2 +-
.../work_with_schedules/sg_execution_times.html | 18 +-
docs/how_to/work_with_schedules/tensorize.html | 2 +-
docs/reference/api/python/auto_scheduler.html | 4 +-
.../api/typedoc/classes/bytestreamreader.html | 12 +-
.../api/typedoc/classes/cachedcallstack.html | 34 +-
docs/reference/api/typedoc/classes/dldatatype.html | 12 +-
docs/reference/api/typedoc/classes/dldevice.html | 10 +-
.../reference/api/typedoc/classes/environment.html | 12 +-
docs/reference/api/typedoc/classes/ffilibrary.html | 20 +-
.../api/typedoc/classes/graphexecutor.html | 16 +-
docs/reference/api/typedoc/classes/instance.html | 40 +-
docs/reference/api/typedoc/classes/memory.html | 34 +-
docs/reference/api/typedoc/classes/module.html | 10 +-
docs/reference/api/typedoc/classes/ndarray.html | 22 +-
.../api/typedoc/classes/packedfunccell.html | 6 +-
docs/reference/api/typedoc/classes/rpcserver.html | 14 +-
docs/reference/api/typedoc/classes/scalar.html | 6 +-
.../api/typedoc/classes/webgpucontext.html | 12 +-
docs/reference/api/typedoc/enums/argtypecode.html | 30 +-
.../api/typedoc/enums/aynccallbackcode.html | 4 +-
.../api/typedoc/enums/dldatatypecode.html | 8 +-
.../api/typedoc/enums/rpcserverstate.html | 12 +-
docs/reference/api/typedoc/enums/sizeof.html | 18 +-
docs/reference/api/typedoc/index.html | 112 +-
.../api/typedoc/interfaces/disposable.html | 2 +-
.../api/typedoc/interfaces/functioninfo.html | 6 +-
.../api/typedoc/interfaces/libraryprovider.html | 4 +-
docs/searchindex.js | 2 +-
.../vta/tutorials/autotvm/sg_execution_times.html | 4 +-
.../tutorials/frontend/deploy_classification.html | 2 +-
.../vta/tutorials/frontend/deploy_detection.html | 2 +-
.../vta/tutorials/frontend/sg_execution_times.html | 6 +-
.../vta/tutorials/optimize/sg_execution_times.html | 6 +-
docs/topic/vta/tutorials/sg_execution_times.html | 6 +-
docs/tutorial/auto_scheduler_matmul_x86.html | 6 +-
docs/tutorial/autotvm_matmul_x86.html | 20 +-
docs/tutorial/autotvm_relay_x86.html | 279 ++-
docs/tutorial/cross_compilation_and_rpc.html | 2 +-
docs/tutorial/intro_topi.html | 2 +-
docs/tutorial/sg_execution_times.html | 26 +-
docs/tutorial/tensor_expr_get_started.html | 41 +-
129 files changed, 5062 insertions(+), 2828 deletions(-)
diff --git a/docs/_images/sphx_glr_micro_train_001.png b/docs/_images/sphx_glr_micro_train_001.png
index 1a78ef007a..9d7b73ba75 100644
Binary files a/docs/_images/sphx_glr_micro_train_001.png and b/docs/_images/sphx_glr_micro_train_001.png differ
diff --git a/docs/_images/sphx_glr_micro_train_thumb.png b/docs/_images/sphx_glr_micro_train_thumb.png
index 8e911753f3..9ae7d8cdc5 100644
Binary files a/docs/_images/sphx_glr_micro_train_thumb.png and b/docs/_images/sphx_glr_micro_train_thumb.png differ
diff --git a/docs/_sources/how_to/compile_models/from_darknet.rst.txt b/docs/_sources/how_to/compile_models/from_darknet.rst.txt
index 804a15dca6..3335735edd 100644
--- a/docs/_sources/how_to/compile_models/from_darknet.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_darknet.rst.txt
@@ -318,7 +318,7 @@ The process is no different from other examples.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 16.080 seconds)
+ **Total running time of the script:** ( 1 minutes 16.875 seconds)
.. _sphx_glr_download_how_to_compile_models_from_darknet.py:
diff --git a/docs/_sources/how_to/compile_models/from_keras.rst.txt b/docs/_sources/how_to/compile_models/from_keras.rst.txt
index 55db197d76..ef1b2f5575 100644
--- a/docs/_sources/how_to/compile_models/from_keras.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_keras.rst.txt
@@ -232,7 +232,7 @@ Look up prediction top 1 index in 1000 class synset.
.. code-block:: none
Relay top-1 id: 285, class name: Egyptian cat
-
1/1 [==============================] - ETA: 0s
1/1 [==============================] - 1s 964ms/step
+
1/1 [==============================] - ETA: 0s
1/1 [==============================] - 1s 947ms/step
Keras top-1 id: 285, class name: Egyptian cat
diff --git a/docs/_sources/how_to/compile_models/from_mxnet.rst.txt b/docs/_sources/how_to/compile_models/from_mxnet.rst.txt
index 6551c32cf5..6754ed68ff 100644
--- a/docs/_sources/how_to/compile_models/from_mxnet.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_mxnet.rst.txt
@@ -116,7 +116,7 @@ In this section, we download a pretrained imagenet model and classify an image.
.. code-block:: none
- Downloading /workspace/.mxnet/models/resnet18_v1-a0666292.zipadb64d9b-d983-4e24-9005-220557c91021 from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
+ Downloading /workspace/.mxnet/models/resnet18_v1-a0666292.zip7772c4e3-b5bb-4af5-84d2-2e2de09354bb from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
x (1, 3, 224, 224)
diff --git a/docs/_sources/how_to/compile_models/from_oneflow.rst.txt b/docs/_sources/how_to/compile_models/from_oneflow.rst.txt
index 7c3a6c19b2..e1b6b62dfa 100644
--- a/docs/_sources/how_to/compile_models/from_oneflow.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_oneflow.rst.txt
@@ -121,7 +121,7 @@ Load a pretrained OneFlow model and save model
.. code-block:: none
Downloading: "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/ResNet/resnet18.zip" to /workspace/.oneflow/flowvision_cache/resnet18.zip
-
0%| | 0.00/41.5M [00:00<?, ?B/s]
19%|#9 | 7.99M/41.5M [00:00<00:00, 60.9MB/s]
39%|###8 | 16.0M/41.5M [00:00<00:00, 63.5MB/s]
58%|#####7 | 24.0M/41.5M [00:00<00:00, 56.6MB/s]
77%|#######7 | 32.0M/41.5M [00:00<00:00, 62.0MB/s]
96%|#########6| 40.0M/41.5M [00:00<00:00, 61.8MB/s]
100%|##########| 41.5M/41.5M [00:00<00:00, 62.9MB/s]
+
0%| | 0.00/41.5M [00:00<?, ?B/s]
19%|#9 | 7.99M/41.5M [00:00<00:00, 82.3MB/s]
39%|###8 | 16.0M/41.5M [00:00<00:00, 57.7MB/s]
58%|#####7 | 24.0M/41.5M [00:00<00:00, 55.3MB/s]
77%|#######7 | 32.0M/41.5M [00:00<00:00, 62.1MB/s]
96%|#########6| 40.0M/41.5M [00:00<00:00, 61.6MB/s]
100%|##########| 41.5M/41.5M [00:00<00:00, 62.6MB/s]
diff --git a/docs/_sources/how_to/compile_models/from_pytorch.rst.txt b/docs/_sources/how_to/compile_models/from_pytorch.rst.txt
index 8ab4a09743..0ef08d96d2 100644
--- a/docs/_sources/how_to/compile_models/from_pytorch.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_pytorch.rst.txt
@@ -101,7 +101,7 @@ Load a pretrained PyTorch model
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and will be removed in 0.15. The current behavior is equivalent to passing `weights=ResNet18_Weights.IMAGENET1K_V1`. You can also use `weights=ResNet18_Weights.DEFAULT` to get the most up-to-date weights.
warnings.warn(msg)
Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to /workspace/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth
-
0%| | 0.00/44.7M [00:00<?, ?B/s]
27%|##7 | 12.1M/44.7M [00:00<00:00, 127MB/s]
54%|#####4 | 24.2M/44.7M [00:00<00:00, 108MB/s]
78%|#######7 | 34.7M/44.7M [00:00<00:00, 105MB/s]
100%|##########| 44.7M/44.7M [00:00<00:00, 106MB/s]
+
0%| | 0.00/44.7M [00:00<?, ?B/s]
27%|##7 | 12.3M/44.7M [00:00<00:00, 129MB/s]
55%|#####4 | 24.5M/44.7M [00:00<00:00, 110MB/s]
90%|########9 | 40.0M/44.7M [00:00<00:00, 119MB/s]
100%|##########| 44.7M/44.7M [00:00<00:00, 120MB/s]
diff --git a/docs/_sources/how_to/compile_models/from_tensorflow.rst.txt b/docs/_sources/how_to/compile_models/from_tensorflow.rst.txt
index 12bb45ff81..8cca4cc8d5 100644
--- a/docs/_sources/how_to/compile_models/from_tensorflow.rst.txt
+++ b/docs/_sources/how_to/compile_models/from_tensorflow.rst.txt
@@ -424,7 +424,7 @@ Run the corresponding model on tensorflow
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 19.756 seconds)
+ **Total running time of the script:** ( 1 minutes 20.526 seconds)
.. _sphx_glr_download_how_to_compile_models_from_tensorflow.py:
diff --git a/docs/_sources/how_to/compile_models/sg_execution_times.rst.txt b/docs/_sources/how_to/compile_models/sg_execution_times.rst.txt
index f71137f4ea..78afe5aa8a 100644
--- a/docs/_sources/how_to/compile_models/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/compile_models/sg_execution_times.rst.txt
@@ -5,26 +5,26 @@
Computation times
=================
-**06:17.311** total execution time for **how_to_compile_models** files:
+**06:17.675** total execution time for **how_to_compile_models** files:
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_tensorflow.py` (``from_tensorflow.py``) | 01:19.756 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_tensorflow.py` (``from_tensorflow.py``) | 01:20.526 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_darknet.py` (``from_darknet.py``) | 01:16.080 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_darknet.py` (``from_darknet.py``) | 01:16.875 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_paddle.py` (``from_paddle.py``) | 00:51.867 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_paddle.py` (``from_paddle.py``) | 00:52.063 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_oneflow.py` (``from_oneflow.py``) | 00:34.823 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_oneflow.py` (``from_oneflow.py``) | 00:35.509 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_coreml.py` (``from_coreml.py``) | 00:30.953 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_mxnet.py` (``from_mxnet.py``) | 00:30.426 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_mxnet.py` (``from_mxnet.py``) | 00:30.871 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_coreml.py` (``from_coreml.py``) | 00:30.080 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_tflite.py` (``from_tflite.py``) | 00:26.285 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_tflite.py` (``from_tflite.py``) | 00:26.201 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_pytorch.py` (``from_pytorch.py``) | 00:24.216 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_pytorch.py` (``from_pytorch.py``) | 00:23.765 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_keras.py` (``from_keras.py``) | 00:19.821 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_keras.py` (``from_keras.py``) | 00:19.631 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_onnx.py` (``from_onnx.py``) | 00:02.639 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_onnx.py` (``from_onnx.py``) | 00:02.598 | 0.0 MB |
+-----------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/deploy_models/deploy_model_on_adreno.rst.txt b/docs/_sources/how_to/deploy_models/deploy_model_on_adreno.rst.txt
index 3497df3b09..ba182633df 100644
--- a/docs/_sources/how_to/deploy_models/deploy_model_on_adreno.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_model_on_adreno.rst.txt
@@ -727,7 +727,7 @@ well as provides information about the model's performance
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 3485.6176 3484.7953 3490.6544 3482.3343 2.6236
+ 3340.4418 3339.6930 3345.6862 3337.8607 2.4028
@@ -736,7 +736,7 @@ well as provides information about the model's performance
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 5.197 seconds)
+ **Total running time of the script:** ( 1 minutes 3.250 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_model_on_adreno.py:
diff --git a/docs/_sources/how_to/deploy_models/deploy_model_on_android.rst.txt b/docs/_sources/how_to/deploy_models/deploy_model_on_android.rst.txt
index 2464742812..95b6e1a61d 100644
--- a/docs/_sources/how_to/deploy_models/deploy_model_on_android.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_model_on_android.rst.txt
@@ -437,7 +437,7 @@ Execute on TVM
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 16.3794 16.3462 16.5128 16.2863 0.0881
+ 16.3099 16.1822 17.2296 15.9700 0.4217
diff --git a/docs/_sources/how_to/deploy_models/deploy_object_detection_pytorch.rst.txt b/docs/_sources/how_to/deploy_models/deploy_object_detection_pytorch.rst.txt
index 889922ba9b..df2c392220 100644
--- a/docs/_sources/how_to/deploy_models/deploy_object_detection_pytorch.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_object_detection_pytorch.rst.txt
@@ -130,7 +130,7 @@ Load pre-trained maskrcnn from torchvision and do tracing
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and will be removed in 0.15. The current behavior is equivalent to passing `weights=MaskRCNN_ResNet50_FPN_Weights.COCO_V1`. You can also use `weights=MaskRCNN_ResNet50_FPN_Weights.DEFAULT` to get the most up-to-date weights.
warnings.warn(msg)
Downloading: "https://download.pytorch.org/models/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth" to /workspace/.cache/torch/hub/checkpoints/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth
-
0%| | 0.00/170M [00:00<?, ?B/s]
7%|7 | 12.4M/170M [00:00<00:01, 130MB/s]
15%|#4 | 24.9M/170M [00:00<00:01, 108MB/s]
21%|## | 35.4M/170M [00:00<00:01, 107MB/s]
27%|##6 | 45.6M/170M [00:00<00:01, 104MB/s]
33%|###2 | 55.7M/170M [00:00<00:01, 103MB/s]
39%|###8 | 65.5M/170M [00:00<00:01, 102MB/s]
44%|####4 | 75.2M/170M [00:00<00:00, 101MB/s]
50%|####9 | 84.9M/170M [00:00<00:00, 101MB/s]
56%|#####5 | 94.5M/170M [00:00<00:00, 101MB/s]
61%|######1 | 104M/170M [00:01<00:00, 99.1MB/s]
67%|######7 | 114M/170M [00:01<00:00, 101MB/s]
73%|#######2 | 124M/170M [00:01<00:00, 101MB/s]
79%|#######8 | 133M/170M [00:01<00:00, 100MB/s]
84%|########4 | 143M/170M [00:01<00:00, 100MB/s]
90%|########9 | 152M/170M [00:01<00:00, 99.3MB/s]
96%|#########5| 162M/170M [00:01<00:00, 100MB/s]
100%|##########| 170M/170M [00:01<00:00, 102MB/s]
+
0%| | 0.00/170M [00:00<?, ?B/s]
5%|4 | 7.99M/170M [00:00<00:03, 44.4MB/s]
7%|7 | 12.2M/170M [00:00<00:04, 38.3MB/s]
9%|9 | 16.0M/170M [00:00<00:04, 37.6MB/s]
13%|#2 | 21.4M/170M [00:00<00:03, 43.7MB/s]
17%|#6 | 28.8M/170M [00:00<00:02, 54.3MB/s]
20%|## | 34.2M/170M [00:00<00:03, 38.8MB/s]
24%|##4 | 41.3M/170M [00:00<00:02, 47.3MB/s]
30%|### | 51.1M/170M [00:01<00:02, 61.5MB/s]
36%|###6 | 61.3M/170M [00:01<00:01, 73.4MB/s]
41%|#### | 69.0M/170M [00:01<00:02, 52.5MB/s]
44%|####4 | 75.3M/170M [00:01<00:01, 53.1MB/s]
48%|####7 | 81.3M/170M [00:01<00:01, 51.1MB/s]
52%|#####2 | 89.1M/170M [00:01<00:01, 58.1MB/s]
60%|###### | 103M/170M [00:01<00:00, 78.8MB/s]
65%|######5 | 111M/170M [00:02<00:00, 78.1MB/s]
71%|####### | 120M/170M [00:02<00:00, 73.3MB/s]
76%|#######6 | 129M/170M [00:02<00:00, 79.5MB/s]
81%|######## | 137M/170M [00:02<00:00, 67.7MB/s]
85%|########5 | 144M/170M [00:02<00:00, 62.3MB/s]
90%|########9 | 153M/170M [00:02<00:00, 68.4MB/s]
94%|#########4| 160M/170M [00:02<00:00, 67.5MB/s]
100%|##########| 170M/170M [00:02<00:00, 62.1MB/s]
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torch/nn/functional.py:3897: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
for i in range(dim)
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torchvision/models/detection/anchor_utils.py:124: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
@@ -299,7 +299,7 @@ Get boxes with score larger than 0.9
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 3 minutes 34.987 seconds)
+ **Total running time of the script:** ( 3 minutes 30.093 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_object_detection_pytorch.py:
diff --git a/docs/_sources/how_to/deploy_models/deploy_prequantized.rst.txt b/docs/_sources/how_to/deploy_models/deploy_prequantized.rst.txt
index 03670a5cac..5245e80f2a 100644
--- a/docs/_sources/how_to/deploy_models/deploy_prequantized.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_prequantized.rst.txt
@@ -227,7 +227,7 @@ training. Other models require a full post training calibration.
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and will be removed in 0.15. The current behavior is equivalent to passing `weights=MobileNet_V2_Weights.IMAGENET1K_V1`. You can also use `weights=MobileNet_V2_Weights.DEFAULT` to get the most up-to-date weights.
warnings.warn(msg)
Downloading: "https://download.pytorch.org/models/mobilenet_v2-b0353104.pth" to /workspace/.cache/torch/hub/checkpoints/mobilenet_v2-b0353104.pth
-
0%| | 0.00/13.6M [00:00<?, ?B/s]
59%|#####8 | 7.99M/13.6M [00:00<00:00, 80.0MB/s]
100%|##########| 13.6M/13.6M [00:00<00:00, 101MB/s]
+
0%| | 0.00/13.6M [00:00<?, ?B/s]
59%|#####8 | 7.99M/13.6M [00:00<00:00, 58.8MB/s]
100%|##########| 13.6M/13.6M [00:00<00:00, 48.1MB/s]
@@ -409,7 +409,7 @@ Here we give an example of how to measure performance of TVM compiled models.
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 90.4587 90.3497 92.1262 90.1692 0.2914
+ 90.4858 90.3190 95.8622 90.0693 0.7481
@@ -458,7 +458,7 @@ TODO
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 14.706 seconds)
+ **Total running time of the script:** ( 1 minutes 13.807 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_prequantized.py:
diff --git a/docs/_sources/how_to/deploy_models/deploy_prequantized_tflite.rst.txt b/docs/_sources/how_to/deploy_models/deploy_prequantized_tflite.rst.txt
index 32f9072f96..4a9165907a 100644
--- a/docs/_sources/how_to/deploy_models/deploy_prequantized_tflite.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_prequantized_tflite.rst.txt
@@ -423,7 +423,7 @@ Here we give an example of how to measure performance of TVM compiled models.
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 123.0591 122.7188 147.8364 121.6682 2.5565
+ 120.8559 120.8256 123.1634 119.9825 0.4285
@@ -460,7 +460,7 @@ Here we give an example of how to measure performance of TVM compiled models.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 2 minutes 29.779 seconds)
+ **Total running time of the script:** ( 2 minutes 28.294 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_prequantized_tflite.py:
diff --git a/docs/_sources/how_to/deploy_models/deploy_quantized.rst.txt b/docs/_sources/how_to/deploy_models/deploy_quantized.rst.txt
index 646f0bbc5f..2a61037df3 100644
--- a/docs/_sources/how_to/deploy_models/deploy_quantized.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_quantized.rst.txt
@@ -257,7 +257,7 @@ We create a Relay VM to build and execute the model.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 26.972 seconds)
+ **Total running time of the script:** ( 1 minutes 30.320 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_quantized.py:
diff --git a/docs/_sources/how_to/deploy_models/deploy_ssd_gluoncv.rst.txt b/docs/_sources/how_to/deploy_models/deploy_ssd_gluoncv.rst.txt
index ac6121f673..735cac2590 100644
--- a/docs/_sources/how_to/deploy_models/deploy_ssd_gluoncv.rst.txt
+++ b/docs/_sources/how_to/deploy_models/deploy_ssd_gluoncv.rst.txt
@@ -170,7 +170,7 @@ Convert and compile model for CPU.
data: None
input_sym_arg_type = in_param.infer_type()[0]
Downloading /workspace/.mxnet/models/ssd_512_resnet50_v1_voc-9c8b225a.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/ssd_512_resnet50_v1_voc-9c8b225a.zip...
-
0%| | 0/132723 [00:00<?, ?KB/s]
5%|4 | 6095/132723 [00:00<00:02, 60939.45KB/s]
11%|#1 | 14707/132723 [00:00<00:01, 75748.30KB/s]
17%|#7 | 23086/132723 [00:00<00:01, 79417.20KB/s]
24%|##3 | 31568/132723 [00:00<00:01, 81546.47KB/s]
30%|### | 40094/132723 [00:00<00:01, 82878.91KB/s]
37%|###6 | 48651/132723 [00:00<00:01, 83791.92KB/s]
43%|####3 | 57139/132723 [00:00<00:00, 84143.51KB/s]
49%|####9 | 65688/132723 [00:00<00:00, 84568.56KB/s]
56%|#####5 | 74240/132723 [00:00<00:00, 84861.38KB/s]
62%|######2 | 82876/132723 [00:01<00:00, 85320.62KB/s]
69%|######8 | 91463/132723 [00:01<00:00, 85486.02KB/s]
75%|#######5 | 100147/132723 [00:01<00:00, 85895.48KB/s]
82%|########1 | 108781/132723 [00:01<00:00, 86027.48KB/s]
88%|########8 | 117454/132723 [00:01<00:00, 86235.77KB/s]
95%|#########5| 126092/132723 [00:01<00:00, 86277.16KB/s]
100%|#######
###| 132723/132723 [00:01<00:00, 84062.38KB/s]
+
0%| | 0/132723 [00:00<?, ?KB/s]
5%|4 | 6269/132723 [00:00<00:02, 62663.47KB/s]
11%|# | 14327/132723 [00:00<00:01, 73190.96KB/s]
17%|#6 | 22357/132723 [00:00<00:01, 76433.51KB/s]
23%|##2 | 30450/132723 [00:00<00:01, 78199.94KB/s]
29%|##9 | 38520/132723 [00:00<00:01, 79098.45KB/s]
35%|###5 | 46502/132723 [00:00<00:01, 79341.91KB/s]
41%|####1 | 54473/132723 [00:00<00:00, 79457.92KB/s]
47%|####7 | 62501/132723 [00:00<00:00, 79715.33KB/s]
53%|#####3 | 70544/132723 [00:00<00:00, 79937.08KB/s]
59%|#####9 | 78549/132723 [00:01<00:00, 79968.55KB/s]
65%|######5 | 86546/132723 [00:01<00:00, 79827.14KB/s]
71%|#######1 | 94608/132723 [00:01<00:00, 80066.65KB/s]
77%|#######7 | 102669/132723 [00:01<00:00, 80229.44KB/s]
83%|########3 | 110726/132723 [00:01<00:00, 80325.62KB/s]
90%|########9 | 118800/132723 [00:01<00:00, 80447.13KB/s]
96%|########
#5| 126845/132723 [00:01<00:00, 80300.26KB/s]
100%|##########| 132723/132723 [00:01<00:00, 79243.42KB/s]
@@ -246,7 +246,7 @@ Display result
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 3 minutes 36.504 seconds)
+ **Total running time of the script:** ( 3 minutes 31.204 seconds)
.. _sphx_glr_download_how_to_deploy_models_deploy_ssd_gluoncv.py:
diff --git a/docs/_sources/how_to/deploy_models/sg_execution_times.rst.txt b/docs/_sources/how_to/deploy_models/sg_execution_times.rst.txt
index f00f780e97..ddf00a5f8a 100644
--- a/docs/_sources/how_to/deploy_models/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/deploy_models/sg_execution_times.rst.txt
@@ -5,26 +5,26 @@
Computation times
=================
-**15:03.815** total execution time for **how_to_deploy_models** files:
+**14:50.812** total execution time for **how_to_deploy_models** files:
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``) | 03:36.504 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``) | 03:31.204 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_object_detection_pytorch.py` (``deploy_object_detection_pytorch.py``) | 03:34.987 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_object_detection_pytorch.py` (``deploy_object_detection_pytorch.py``) | 03:30.093 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized_tflite.py` (``deploy_prequantized_tflite.py``) | 02:29.779 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized_tflite.py` (``deploy_prequantized_tflite.py``) | 02:28.294 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_quantized.py` (``deploy_quantized.py``) | 01:26.972 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_quantized.py` (``deploy_quantized.py``) | 01:30.320 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized.py` (``deploy_prequantized.py``) | 01:14.706 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized.py` (``deploy_prequantized.py``) | 01:13.807 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_adreno.py` (``deploy_model_on_adreno.py``) | 01:05.197 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_adreno.py` (``deploy_model_on_adreno.py``) | 01:03.250 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_android.py` (``deploy_model_on_android.py``) | 00:41.204 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_android.py` (``deploy_model_on_android.py``) | 00:40.356 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_nano.py` (``deploy_model_on_nano.py``) | 00:27.427 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_nano.py` (``deploy_model_on_nano.py``) | 00:26.963 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``) | 00:27.034 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``) | 00:26.521 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_how_to_deploy_models_deploy_sparse.py` (``deploy_sparse.py``) | 00:00.006 | 0.0 MB |
+------------------------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/extend_tvm/bring_your_own_datatypes.rst.txt b/docs/_sources/how_to/extend_tvm/bring_your_own_datatypes.rst.txt
index 5589ed2e99..679595e89c 100644
--- a/docs/_sources/how_to/extend_tvm/bring_your_own_datatypes.rst.txt
+++ b/docs/_sources/how_to/extend_tvm/bring_your_own_datatypes.rst.txt
@@ -463,7 +463,7 @@ First let us define two helper functions to get the mobilenet model and a cat im
.. code-block:: none
- Downloading /workspace/.mxnet/models/mobilenet0.25-9f83e440.zip6226acea-b49c-4426-8032-9e81465ff59a from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
+ Downloading /workspace/.mxnet/models/mobilenet0.25-9f83e440.zip8aed1037-c0db-424d-a486-9c23a0465e2a from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
diff --git a/docs/_sources/how_to/extend_tvm/sg_execution_times.rst.txt b/docs/_sources/how_to/extend_tvm/sg_execution_times.rst.txt
index 65b768cad3..a11b7ad941 100644
--- a/docs/_sources/how_to/extend_tvm/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/extend_tvm/sg_execution_times.rst.txt
@@ -5,14 +5,14 @@
Computation times
=================
-**00:55.017** total execution time for **how_to_extend_tvm** files:
+**00:53.382** total execution time for **how_to_extend_tvm** files:
+-------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_extend_tvm_bring_your_own_datatypes.py` (``bring_your_own_datatypes.py``) | 00:51.133 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_bring_your_own_datatypes.py` (``bring_your_own_datatypes.py``) | 00:49.580 | 0.0 MB |
+-------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_extend_tvm_use_pass_instrument.py` (``use_pass_instrument.py``) | 00:02.764 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_use_pass_instrument.py` (``use_pass_instrument.py``) | 00:02.708 | 0.0 MB |
+-------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_extend_tvm_use_pass_infra.py` (``use_pass_infra.py``) | 00:01.113 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_use_pass_infra.py` (``use_pass_infra.py``) | 00:01.087 | 0.0 MB |
+-------------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_how_to_extend_tvm_low_level_custom_pass.py` (``low_level_custom_pass.py``) | 00:00.007 | 0.0 MB |
+-------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/extend_tvm/use_pass_instrument.rst.txt b/docs/_sources/how_to/extend_tvm/use_pass_instrument.rst.txt
index ed4f61af6b..2a47b66deb 100644
--- a/docs/_sources/how_to/extend_tvm/use_pass_instrument.rst.txt
+++ b/docs/_sources/how_to/extend_tvm/use_pass_instrument.rst.txt
@@ -220,10 +220,10 @@ profile the execution time of each passes.
.. code-block:: none
Printing results of timing profile...
- InferType: 18990us [18990us] (48.38%; 48.38%)
- FoldScaleAxis: 20260us [10us] (51.62%; 51.62%)
- FoldConstant: 20250us [1863us] (51.59%; 99.95%)
- InferType: 18387us [18387us] (46.85%; 90.80%)
+ InferType: 18404us [18404us] (48.41%; 48.41%)
+ FoldScaleAxis: 19615us [8us] (51.59%; 51.59%)
+ FoldConstant: 19607us [1783us] (51.57%; 99.96%)
+ InferType: 17823us [17823us] (46.88%; 90.91%)
@@ -262,10 +262,10 @@ Refer to following sections and :py:func:`tvm.instrument.pass_instrument` for th
.. code-block:: none
Printing results of timing profile...
- InferType: 18830us [18830us] (47.13%; 47.13%)
- FoldScaleAxis: 21126us [9us] (52.87%; 52.87%)
- FoldConstant: 21117us [1949us] (52.85%; 99.96%)
- InferType: 19168us [19168us] (47.97%; 90.77%)
+ InferType: 18040us [18040us] (47.49%; 47.49%)
+ FoldScaleAxis: 19946us [8us] (52.51%; 52.51%)
+ FoldConstant: 19938us [1773us] (52.49%; 99.96%)
+ InferType: 18166us [18166us] (47.82%; 91.11%)
diff --git a/docs/_sources/how_to/optimize_operators/opt_conv_cuda.rst.txt b/docs/_sources/how_to/optimize_operators/opt_conv_cuda.rst.txt
index 3a048dcc1d..2b7a493a31 100644
--- a/docs/_sources/how_to/optimize_operators/opt_conv_cuda.rst.txt
+++ b/docs/_sources/how_to/optimize_operators/opt_conv_cuda.rst.txt
@@ -331,7 +331,7 @@ latency of convolution.
.. code-block:: none
- Convolution: 54.172832 ms
+ Convolution: 54.222881 ms
diff --git a/docs/_sources/how_to/optimize_operators/opt_conv_tensorcore.rst.txt b/docs/_sources/how_to/optimize_operators/opt_conv_tensorcore.rst.txt
index b6acd053ec..8ed86f5ded 100644
--- a/docs/_sources/how_to/optimize_operators/opt_conv_tensorcore.rst.txt
+++ b/docs/_sources/how_to/optimize_operators/opt_conv_tensorcore.rst.txt
@@ -602,7 +602,7 @@ be able to run on our build server
.. code-block:: none
- conv2d with tensor core: 13.370595 ms
+ conv2d with tensor core: 6.710870 ms
diff --git a/docs/_sources/how_to/optimize_operators/opt_gemm.rst.txt b/docs/_sources/how_to/optimize_operators/opt_gemm.rst.txt
index bfb92429da..93784d0e76 100644
--- a/docs/_sources/how_to/optimize_operators/opt_gemm.rst.txt
+++ b/docs/_sources/how_to/optimize_operators/opt_gemm.rst.txt
@@ -134,8 +134,8 @@ Then we write a baseline implementation, the simplest way to write a matrix mult
.. code-block:: none
- Numpy running time: 0.019813
- Baseline: 3.269114
+ Numpy running time: 0.019114
+ Baseline: 3.291007
@@ -224,7 +224,7 @@ fill 32 * 32 * sizeof(float) which is 4KB in the cache whose total size is 32KB
.. code-block:: none
- Opt1: 0.326293
+ Opt1: 0.303538
@@ -312,7 +312,7 @@ In this tutorial, we chose to vectorize the inner loop row data since it is cach
.. code-block:: none
- Opt2: 0.351003
+ Opt2: 0.344968
@@ -397,7 +397,7 @@ the access pattern for A matrix is more cache friendly.
.. code-block:: none
- Opt3: 0.126392
+ Opt3: 0.118944
@@ -511,7 +511,7 @@ flattening.
.. code-block:: none
- Opt4: 0.110417
+ Opt4: 0.109755
@@ -620,7 +620,7 @@ write to C when all the block results are ready.
.. code-block:: none
- Opt5: 0.112664
+ Opt5: 0.111921
@@ -730,7 +730,7 @@ Furthermore, we can also utilize multi-core processors to do the thread-level pa
.. code-block:: none
- Opt6: 0.148654
+ Opt6: 0.146599
diff --git a/docs/_sources/how_to/optimize_operators/sg_execution_times.rst.txt b/docs/_sources/how_to/optimize_operators/sg_execution_times.rst.txt
index 6b8ab6e423..fd3c004561 100644
--- a/docs/_sources/how_to/optimize_operators/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/optimize_operators/sg_execution_times.rst.txt
@@ -5,12 +5,12 @@
Computation times
=================
-**00:35.612** total execution time for **how_to_optimize_operators** files:
+**00:34.720** total execution time for **how_to_optimize_operators** files:
+-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_gemm.py` (``opt_gemm.py``) | 00:32.780 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_gemm.py` (``opt_gemm.py``) | 00:32.194 | 0.0 MB |
+-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``) | 00:01.637 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``) | 00:01.448 | 0.0 MB |
+-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_cuda.py` (``opt_conv_cuda.py``) | 00:01.195 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_cuda.py` (``opt_conv_cuda.py``) | 00:01.077 | 0.0 MB |
+-----------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/tune_with_autoscheduler/sg_execution_times.rst.txt b/docs/_sources/how_to/tune_with_autoscheduler/sg_execution_times.rst.txt
index d0a5edfa29..c5a09e05f7 100644
--- a/docs/_sources/how_to/tune_with_autoscheduler/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/tune_with_autoscheduler/sg_execution_times.rst.txt
@@ -5,18 +5,18 @@
Computation times
=================
-**09:53.666** total execution time for **how_to_tune_with_autoscheduler** files:
+**09:26.610** total execution time for **how_to_tune_with_autoscheduler** files:
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_conv2d_layer_cuda.py` (``tune_conv2d_layer_cuda.py``) | 05:58.328 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_conv2d_layer_cuda.py` (``tune_conv2d_layer_cuda.py``) | 05:48.187 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_x86.py` (``tune_network_x86.py``) | 01:41.158 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_x86.py` (``tune_network_x86.py``) | 01:38.178 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_cuda.py` (``tune_network_cuda.py``) | 01:05.726 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_cuda.py` (``tune_network_cuda.py``) | 01:05.333 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_sparse_x86.py` (``tune_sparse_x86.py``) | 00:41.572 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_sparse_x86.py` (``tune_sparse_x86.py``) | 00:27.967 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_arm.py` (``tune_network_arm.py``) | 00:14.008 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_arm.py` (``tune_network_arm.py``) | 00:14.033 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_mali.py` (``tune_network_mali.py``) | 00:12.875 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_mali.py` (``tune_network_mali.py``) | 00:12.913 | 0.0 MB |
+----------------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.rst.txt b/docs/_sources/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.rst.txt
index d9313b5609..0927cfae15 100644
--- a/docs/_sources/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.rst.txt
+++ b/docs/_sources/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.rst.txt
@@ -209,6 +209,13 @@ file and apply it.
+.. rst-class:: sphx-glr-script-out
+
+ .. code-block:: none
+
+ .T
+
+
@@ -241,13 +248,13 @@ cooperative fetching, unrolling and operator fusion.
def main(data: T.Buffer((1, 512, 7, 7), "float32"), kernel: T.Buffer((512, 512, 3, 3), "float32"), bias: T.Buffer((1, 512, 1, 1), "float32"), compute: T.Buffer((1, 512, 7, 7), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
blockIdx_x = T.env_thread("blockIdx.x")
- T.launch_thread(blockIdx_x, 28)
- conv2d_nchw = T.allocate([14], "float32", "local")
- pad_temp_shared = T.allocate([72], "float32", "shared")
- kernel_shared = T.allocate([3072], "float32", "shared")
+ T.launch_thread(blockIdx_x, 16)
+ conv2d_nchw = T.allocate([7], "float32", "local")
+ pad_temp_shared = T.allocate([648], "float32", "shared")
+ kernel_shared = T.allocate([2304], "float32", "shared")
threadIdx_x = T.env_thread("threadIdx.x")
- T.launch_thread(threadIdx_x, 64)
- conv2d_nchw_1 = T.buffer_decl((14,), data=conv2d_nchw, scope="local", align=32)
+ T.launch_thread(threadIdx_x, 224)
+ conv2d_nchw_1 = T.buffer_decl((7,), data=conv2d_nchw, scope="local", align=16)
conv2d_nchw_1[0] = T.float32(0)
conv2d_nchw_1[1] = T.float32(0)
conv2d_nchw_1[2] = T.float32(0)
@@ -255,467 +262,553 @@ cooperative fetching, unrolling and operator fusion.
conv2d_nchw_1[4] = T.float32(0)
conv2d_nchw_1[5] = T.float32(0)
conv2d_nchw_1[6] = T.float32(0)
- conv2d_nchw_1[7] = T.float32(0)
- conv2d_nchw_1[8] = T.float32(0)
- conv2d_nchw_1[9] = T.float32(0)
- conv2d_nchw_1[10] = T.float32(0)
- conv2d_nchw_1[11] = T.float32(0)
- conv2d_nchw_1[12] = T.float32(0)
- conv2d_nchw_1[13] = T.float32(0)
- for rc_outer_outer, ry_outer_outer in T.grid(64, 3):
- cse_var_2: T.int32 = rc_outer_outer * 72
- cse_var_1: T.int32 = ry_outer_outer * 3
+ for rc_outer_outer in range(64):
+ cse_var_2: T.int32 = rc_outer_outer * 392
+ cse_var_1: T.int32 = rc_outer_outer * 72
threadIdx_x_1 = T.env_thread("threadIdx.x")
- pad_temp_shared_1 = T.buffer_decl((72,), data=pad_temp_shared, scope="shared")
- with T.launch_thread(threadIdx_x_1, 64):
- data_1 = T.buffer_decl((25088,), data=data.data)
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= threadIdx_x_1 * 4 % 9 and threadIdx_x_1 * 4 % 9 < 8, data_1[rc_outer_outer * 392 + threadIdx_x_1 * 4 // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + threadIdx_x_1 * 4 % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 1] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 1) % 9 and (threadIdx_x_1 * 4 + 1) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 1) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 1) % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 2] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 2) % 9 and (threadIdx_x_1 * 4 + 2) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 2) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 2) % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 3] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 3) % 9 and (threadIdx_x_1 * 4 + 3) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 3) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 3) % 9 - 8], T.float32(0))
+ pad_temp_shared_1 = T.buffer_decl((648,), data=pad_temp_shared, scope="shared")
+ data_1 = T.buffer_decl((25088,), data=data.data)
+ with T.launch_thread(threadIdx_x_1, 224):
+ pad_temp_shared_1[threadIdx_x_1] = T.if_then_else(9 <= threadIdx_x_1 % 81 and threadIdx_x_1 % 81 < 72 and 1 <= threadIdx_x_1 % 9 and threadIdx_x_1 % 9 < 8, data_1[cse_var_2 + threadIdx_x_1 // 81 * 49 + threadIdx_x_1 % 81 // 9 * 7 + threadIdx_x_1 % 9 - 8], T.float32(0))
+ with T.launch_thread(threadIdx_x_1, 224):
+ pad_temp_shared_1[threadIdx_x_1 + 224] = T.if_then_else(9 <= (threadIdx_x_1 + 62) % 81 and (threadIdx_x_1 + 62) % 81 < 72 and 1 <= (threadIdx_x_1 + 8) % 9 and (threadIdx_x_1 + 8) % 9 < 8, data_1[cse_var_2 + (threadIdx_x_1 + 224) // 81 * 49 + (threadIdx_x_1 + 62) % 81 // 9 * 7 + (threadIdx_x_1 + 8) % 9 - 8], T.float32(0))
+ with T.launch_thread(threadIdx_x_1, 224):
+ if T.likely(threadIdx_x_1 < 200):
+ pad_temp_shared_1[threadIdx_x_1 + 448] = T.if_then_else(9 <= (threadIdx_x_1 + 43) % 81 and (threadIdx_x_1 + 43) % 81 < 72 and 1 <= (threadIdx_x_1 + 7) % 9 and (threadIdx_x_1 + 7) % 9 < 8, data_1[cse_var_2 + (threadIdx_x_1 + 448) // 81 * 49 + (threadIdx_x_1 + 43) % 81 // 9 * 7 + (threadIdx_x_1 + 7) % 9 - 8], T.float32(0))
threadIdx_x_2 = T.env_thread("threadIdx.x")
- kernel_shared_1 = T.buffer_decl((3072,), data=kernel_shared, scope="shared")
+ kernel_shared_1 = T.buffer_decl((2304,), data=kernel_shared, scope="shared")
kernel_1 = T.buffer_decl((2359296,), data=kernel.data)
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 64] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 64) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 128] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 128) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 192] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 36864]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 256] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 256) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 320] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 320) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 384] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 73728]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 448] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 448) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 512] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 512) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 576] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 110592]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 640] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 640) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 704] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 704) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 768] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 147456]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 832] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 832) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 896] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 896) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 960] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 184320]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1024] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1024) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1088] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1088) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1152] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 221184]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1216] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1216) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1280] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1280) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1344] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 258048]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1408] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1408) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1472] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1472) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1536] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 294912]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1600] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1600) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1664] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1664) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1728] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 331776]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1792] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1792) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1856] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1856) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1920] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 368640]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1984] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1984) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2048] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2048) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2112] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 405504]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2176] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2176) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2240] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2240) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2304] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 442368]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2368] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2368) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2432] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2432) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2496] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 479232]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2560] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2560) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2624] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2624) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2688] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 516096]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2752] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2752) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2816] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2816) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2880] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 552960]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2944] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2944) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 3008] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 3008) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 47]
- for i1_inner, i3_inner in T.grid(2, 7):
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2] = kernel_1[blockIdx_x * 147456 + threadIdx_x_2 // 72 * 4608 + cse_var_1 + threadIdx_x_2 % 72]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 224] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 224) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 8) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 448] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 448) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 16) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 672] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 672) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 // 3 + 8) % 24 * 3 + threadIdx_x_2 % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 896] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 896) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 32) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1120] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1120) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 40) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1344] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1344) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 // 3 + 16) % 24 * 3 + threadIdx_x_2 % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1568] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1568) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 56) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1792] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1792) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 64) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 2016] = kernel_1[blockIdx_x * 147456 + threadIdx_x_2 // 72 * 4608 + cse_var_1 + threadIdx_x_2 % 72 + 129024]
+ with T.launch_thread(threadIdx_x_2, 224):
+ if T.likely(threadIdx_x_2 < 64):
+ kernel_shared_1[threadIdx_x_2 + 2240] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 2240) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 8) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 1] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 81] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 82] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 1] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 82] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 7] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 88] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 7] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 8] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 88] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 89] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 9] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 10] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 90] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 91] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 10] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 91] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 16] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 97] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 16] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 17] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 97] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 98] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 18] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 19] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 99] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 100] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 19] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 100] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 25] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 106] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 25] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 26] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 106] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 107] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 162] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 163] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 243] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 244] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 163] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 244] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 169] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 250] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 169] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 170] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 250] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 251] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 171] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 172] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 252] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 253] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 172] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 253] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 178] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 259] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 178] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 179] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 259] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 260] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 180] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 181] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 261] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 262] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 181] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 262] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 187] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 268] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 187] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 188] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 268] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 269] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 324] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 325] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 405] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 406] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 325] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 406] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 331] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 412] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 331] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 332] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 412] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 413] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 333] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 334] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 414] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 415] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 334] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 415] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 340] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 421] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 340] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 341] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 421] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 422] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 342] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 343] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 423] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 424] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 343] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 424] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 349] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 430] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 349] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 350] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 430] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 431] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 486] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 487] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 567] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 568] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 487] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 568] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 493] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 574] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 493] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 494] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 574] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 575] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 495] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 496] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 576] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 577] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 496] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 577] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 502] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 583] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 502] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 503] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 583] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 584] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 504] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 505] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 585] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 586] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 505] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 586] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 511] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 592] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 511] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 512] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 592] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 593] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ for i3_inner in range(7):
compute_1 = T.buffer_decl((25088,), data=compute.data)
bias_1 = T.buffer_decl((512,), data=bias.data)
- compute_1[blockIdx_x // 7 * 6272 + threadIdx_x * 98 + i1_inner * 49 + blockIdx_x % 7 * 7 + i3_inner] = T.max(conv2d_nchw_1[i1_inner * 7 + i3_inner] + bias_1[blockIdx_x // 7 * 128 + threadIdx_x * 2 + i1_inner], T.float32(0))
+ compute_1[blockIdx_x * 1568 + threadIdx_x * 7 + i3_inner] = T.max(conv2d_nchw_1[i3_inner] + bias_1[blockIdx_x * 32 + threadIdx_x // 7], T.float32(0))
@@ -765,7 +858,7 @@ We build the binary and check its correctness and performance.
.. code-block:: none
- Execution time of this operator: 0.356 ms
+ Execution time of this operator: 0.245 ms
@@ -814,12 +907,12 @@ They can be used for debugging and learning the behavior of the auto-scheduler.
conv2d_nchw_nn_o_o_o_i, conv2d_nchw_nn_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_i, factor=1)
conv2d_nchw_nn_o_o_o_o, conv2d_nchw_nn_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_o_i, factor=1)
conv2d_nchw_ff_o_i, conv2d_nchw_ff_i = s[conv2d_nchw].split(conv2d_nchw_ff, factor=1)
- conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=2)
- conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=64)
+ conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=1)
+ conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=32)
conv2d_nchw_ff_o_o_o_o, conv2d_nchw_ff_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_o_i, factor=1)
conv2d_nchw_yy_o_i, conv2d_nchw_yy_i = s[conv2d_nchw].split(conv2d_nchw_yy, factor=1)
conv2d_nchw_yy_o_o_i, conv2d_nchw_yy_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_i, factor=1)
- conv2d_nchw_yy_o_o_o_i, conv2d_nchw_yy_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_i, factor=1)
+ conv2d_nchw_yy_o_o_o_i, conv2d_nchw_yy_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_i, factor=7)
conv2d_nchw_yy_o_o_o_o, conv2d_nchw_yy_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_o_i, factor=1)
conv2d_nchw_xx_o_i, conv2d_nchw_xx_i = s[conv2d_nchw].split(conv2d_nchw_xx, factor=1)
conv2d_nchw_xx_o_o_i, conv2d_nchw_xx_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_i, factor=7)
@@ -828,18 +921,18 @@ They can be used for debugging and learning the behavior of the auto-scheduler.
conv2d_nchw_rc_o_i, conv2d_nchw_rc_i = s[conv2d_nchw].split(conv2d_nchw_rc, factor=2)
conv2d_nchw_rc_o_o, conv2d_nchw_rc_o_i = s[conv2d_nchw].split(conv2d_nchw_rc_o_i, factor=4)
conv2d_nchw_ry_o_i, conv2d_nchw_ry_i = s[conv2d_nchw].split(conv2d_nchw_ry, factor=1)
- conv2d_nchw_ry_o_o, conv2d_nchw_ry_o_i = s[conv2d_nchw].split(conv2d_nchw_ry_o_i, factor=1)
- conv2d_nchw_rx_o_i, conv2d_nchw_rx_i = s[conv2d_nchw].split(conv2d_nchw_rx, factor=1)
- conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_o_i, factor=3)
+ conv2d_nchw_ry_o_o, conv2d_nchw_ry_o_i = s[conv2d_nchw].split(conv2d_nchw_ry_o_i, factor=3)
+ conv2d_nchw_rx_o_i, conv2d_nchw_rx_i = s[conv2d_nchw].split(conv2d_nchw_rx, factor=3)
+ conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_o_i, factor=1)
s[conv2d_nchw].reorder(conv2d_nchw_nn_o_o_o_o, conv2d_nchw_ff_o_o_o_o, conv2d_nchw_yy_o_o_o_o, conv2d_nchw_xx_o_o_o_o, conv2d_nchw_nn_o_o_o_i, conv2d_nchw_ff_o_o_o_i, conv2d_nchw_yy_o_o_o_i, conv2d_nchw_xx_o_o_o_i, conv2d_nchw_nn_o_o_i, conv2d_nchw_ff_o_o_i, conv2d_nchw_yy_o_o_i, conv2d_nchw_xx_o_o_i, conv2d_nchw_rc_o_o, conv2d_nchw_ry_o_o, conv2d_nchw_rx_o_o, conv2d_nchw_rc_o_i, conv2d_nchw_ry_o_i, conv2d_nchw_rx_o_i, conv2d_nchw_nn_o_i, conv2d_nchw_ff_o_i, conv2d_nchw_yy_o_i, conv2 [...]
compute_i0_o_i, compute_i0_i = s[compute].split(compute_i0, factor=1)
compute_i0_o_o_i, compute_i0_o_i = s[compute].split(compute_i0_o_i, factor=1)
compute_i0_o_o_o, compute_i0_o_o_i = s[compute].split(compute_i0_o_o_i, factor=1)
- compute_i1_o_i, compute_i1_i = s[compute].split(compute_i1, factor=2)
- compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=64)
+ compute_i1_o_i, compute_i1_i = s[compute].split(compute_i1, factor=1)
+ compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=32)
compute_i1_o_o_o, compute_i1_o_o_i = s[compute].split(compute_i1_o_o_i, factor=1)
compute_i2_o_i, compute_i2_i = s[compute].split(compute_i2, factor=1)
- compute_i2_o_o_i, compute_i2_o_i = s[compute].split(compute_i2_o_i, factor=1)
+ compute_i2_o_o_i, compute_i2_o_i = s[compute].split(compute_i2_o_i, factor=7)
compute_i2_o_o_o, compute_i2_o_o_i = s[compute].split(compute_i2_o_o_i, factor=1)
compute_i3_o_i, compute_i3_i = s[compute].split(compute_i3, factor=7)
compute_i3_o_o_i, compute_i3_o_i = s[compute].split(compute_i3_o_i, factor=1)
@@ -862,14 +955,14 @@ They can be used for debugging and learning the behavior of the auto-scheduler.
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[kernel_shared].fuse(kernel_shared_ax0, kernel_shared_ax1, kernel_shared_ax2, kernel_shared_ax3)
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=1)
s[kernel_shared].vectorize(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
- kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
+ kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=224)
s[kernel_shared].bind(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[pad_temp_shared].fuse(pad_temp_shared_ax0, pad_temp_shared_ax1, pad_temp_shared_ax2, pad_temp_shared_ax3)
- pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=4)
+ pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=1)
s[pad_temp_shared].vectorize(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
- pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
+ pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=224)
s[pad_temp_shared].bind(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
- s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "auto_unroll_max_step", 512)
+ s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "auto_unroll_max_step", 1024)
s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "unroll_explicit", True)
CUDA source code:
@@ -887,10 +980,10 @@ They can be used for debugging and learning the behavior of the auto-scheduler.
#define int64_t long long
#define uint64_t unsigned long long
#endif
- extern "C" __global__ void __launch_bounds__(64) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
- float conv2d_nchw[14];
- __shared__ float pad_temp_shared[72];
- __shared__ float kernel_shared[3072];
+ extern "C" __global__ void __launch_bounds__(224) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
+ float conv2d_nchw[7];
+ __shared__ float pad_temp_shared[648];
+ __shared__ float kernel_shared[2304];
conv2d_nchw[0] = 0.000000e+00f;
conv2d_nchw[1] = 0.000000e+00f;
conv2d_nchw[2] = 0.000000e+00f;
@@ -898,419 +991,534 @@ They can be used for debugging and learning the behavior of the auto-scheduler.
conv2d_nchw[4] = 0.000000e+00f;
conv2d_nchw[5] = 0.000000e+00f;
conv2d_nchw[6] = 0.000000e+00f;
- conv2d_nchw[7] = 0.000000e+00f;
- conv2d_nchw[8] = 0.000000e+00f;
- conv2d_nchw[9] = 0.000000e+00f;
- conv2d_nchw[10] = 0.000000e+00f;
- conv2d_nchw[11] = 0.000000e+00f;
- conv2d_nchw[12] = 0.000000e+00f;
- conv2d_nchw[13] = 0.000000e+00f;
for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) {
- for (int ry_outer_outer = 0; ry_outer_outer < 3; ++ry_outer_outer) {
- __syncthreads();
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[(((int)threadIdx.x) * 4)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= ((((int)threadIdx.x) * 4) % 9))) && (((((int)threadIdx.x) * 4) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + (((((int)threadIdx.x) * 4) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + ((((int)threadIdx.x) * 4) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 1)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 1) % 9))) && ((((((int)threadIdx.x) * 4) + 1) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 1) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 1) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 2)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 2) % 9))) && ((((((int)threadIdx.x) * 4) + 2) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 2) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 2) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 3)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 3) % 9))) && ((((((int)threadIdx.x) * 4) + 3) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 3) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 3) % 9)) - 8)] : 0.000000e+00f);
- }
- kernel_shared[((int)threadIdx.x)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3))];
- kernel_shared[(((int)threadIdx.x) + 64)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 64) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 128)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 128) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 192)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 36864)];
- kernel_shared[(((int)threadIdx.x) + 256)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 256) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 320)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 320) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 384)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 73728)];
- kernel_shared[(((int)threadIdx.x) + 448)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 448) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 512)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 512) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 576)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 110592)];
- kernel_shared[(((int)threadIdx.x) + 640)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 640) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 704)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 704) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 768)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 147456)];
- kernel_shared[(((int)threadIdx.x) + 832)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 832) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 896)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 896) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 960)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 184320)];
- kernel_shared[(((int)threadIdx.x) + 1024)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1024) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1088)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1088) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1152)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 221184)];
- kernel_shared[(((int)threadIdx.x) + 1216)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1216) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1280)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1280) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1344)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 258048)];
- kernel_shared[(((int)threadIdx.x) + 1408)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1408) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1472)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1472) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1536)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 294912)];
- kernel_shared[(((int)threadIdx.x) + 1600)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1600) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1664)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1664) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1728)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 331776)];
- kernel_shared[(((int)threadIdx.x) + 1792)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1792) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1856)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1856) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1920)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 368640)];
- kernel_shared[(((int)threadIdx.x) + 1984)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1984) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2048)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2048) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2112)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 405504)];
- kernel_shared[(((int)threadIdx.x) + 2176)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2176) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2240)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2240) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2304)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 442368)];
- kernel_shared[(((int)threadIdx.x) + 2368)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2368) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2432)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2432) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2496)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 479232)];
- kernel_shared[(((int)threadIdx.x) + 2560)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2560) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2624)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2624) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2688)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 516096)];
- kernel_shared[(((int)threadIdx.x) + 2752)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2752) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2816)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2816) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2880)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 552960)];
- kernel_shared[(((int)threadIdx.x) + 2944)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2944) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 3008)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 3008) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- __syncthreads();
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[0] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[1] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[2] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[3] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[4] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[5] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[6] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[0] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
+ __syncthreads();
+ pad_temp_shared[((int)threadIdx.x)] = (((((9 <= (((int)threadIdx.x) % 81)) && ((((int)threadIdx.x) % 81) < 72)) && (1 <= (((int)threadIdx.x) % 9))) && ((((int)threadIdx.x) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + ((((int)threadIdx.x) / 81) * 49)) + (((((int)threadIdx.x) % 81) / 9) * 7)) + (((int)threadIdx.x) % 9)) - 8)] : 0.000000e+00f);
+ pad_temp_shared[(((int)threadIdx.x) + 224)] = (((((9 <= ((((int)threadIdx.x) + 62) % 81)) && (((((int)threadIdx.x) + 62) % 81) < 72)) && (1 <= ((((int)threadIdx.x) + 8) % 9))) && (((((int)threadIdx.x) + 8) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 224) / 81) * 49)) + ((((((int)threadIdx.x) + 62) % 81) / 9) * 7)) + ((((int)threadIdx.x) + 8) % 9)) - 8)] : 0.000000e+00f);
+ if (((int)threadIdx.x) < 200) {
+ pad_temp_shared[(((int)threadIdx.x) + 448)] = (((((9 <= ((((int)threadIdx.x) + 43) % 81)) && (((((int)threadIdx.x) + 43) % 81) < 72)) && (1 <= ((((int)threadIdx.x) + 7) % 9))) && (((((int)threadIdx.x) + 7) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 448) / 81) * 49)) + ((((((int)threadIdx.x) + 43) % 81) / 9) * 7)) + ((((int)threadIdx.x) + 7) % 9)) - 8)] : 0.000000e+00f);
}
- }
- for (int i1_inner = 0; i1_inner < 2; ++i1_inner) {
- for (int i3_inner = 0; i3_inner < 7; ++i3_inner) {
- compute[((((((((int)blockIdx.x) / 7) * 6272) + (((int)threadIdx.x) * 98)) + (i1_inner * 49)) + ((((int)blockIdx.x) % 7) * 7)) + i3_inner)] = max((conv2d_nchw[((i1_inner * 7) + i3_inner)] + bias[((((((int)blockIdx.x) / 7) * 128) + (((int)threadIdx.x) * 2)) + i1_inner)]), 0.000000e+00f);
+ kernel_shared[((int)threadIdx.x)] = kernel[((((((int)blockIdx.x) * 147456) + ((((int)threadIdx.x) / 72) * 4608)) + (rc_outer_outer * 72)) + (((int)threadIdx.x) % 72))];
+ kernel_shared[(((int)threadIdx.x) + 224)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 224) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 448)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 448) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 672)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 672) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) / 3) + 8) % 24) * 3)) + (((int)threadIdx.x) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 896)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 896) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 32) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1120)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1120) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 40) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1344)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1344) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) / 3) + 16) % 24) * 3)) + (((int)threadIdx.x) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1568)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1568) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 56) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1792)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1792) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 64) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 2016)] = kernel[(((((((int)blockIdx.x) * 147456) + ((((int)threadIdx.x) / 72) * 4608)) + (rc_outer_outer * 72)) + (((int)threadIdx.x) % 72)) + 129024)];
+ if (((int)threadIdx.x) < 64) {
+ kernel_shared[(((int)threadIdx.x) + 2240)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 2240) / 72) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) + 8) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
}
+ __syncthreads();
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) * 9)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 1)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 81)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 1)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 7)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 88)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 7)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 8)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 88)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 89)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 9)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 16)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 97)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 16)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 17)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 97)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 98)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 25)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 106)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 25)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 26)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 106)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 107)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 162)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 163)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 243)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 244)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 163)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 244)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 169)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 250)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 169)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 170)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 250)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 251)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 171)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 172)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 252)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 253)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 172)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 253)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 178)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 259)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 178)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 179)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 259)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 260)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 180)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 181)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 261)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 262)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 181)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 262)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 187)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 268)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 187)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 188)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 268)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 269)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 324)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 325)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 405)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 406)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 325)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 406)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 331)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 412)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 331)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 332)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 412)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 413)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 333)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 334)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 414)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 415)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 334)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 415)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 340)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 421)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 340)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 341)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 421)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 422)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 342)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 343)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 423)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 424)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 343)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 424)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 349)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 430)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 349)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 350)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 430)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 431)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 486)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 487)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 567)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 568)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 487)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 568)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 493)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 574)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 493)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 494)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 574)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 575)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 495)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 496)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 576)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 577)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 496)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 577)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 502)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 583)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 502)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 503)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 583)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 584)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 504)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 505)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 585)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 586)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 505)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 586)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 511)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 592)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 511)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 512)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 592)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 593)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ }
+ for (int i3_inner = 0; i3_inner < 7; ++i3_inner) {
+ compute[(((((int)blockIdx.x) * 1568) + (((int)threadIdx.x) * 7)) + i3_inner)] = max((conv2d_nchw[i3_inner] + bias[((((int)blockIdx.x) * 32) + (((int)threadIdx.x) / 7))]), 0.000000e+00f);
}
}
@@ -1364,7 +1572,7 @@ In the example below we resume the status and do more 5 trials.
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/xgboost/training.py:17: UserWarning: Old style callback is deprecated. See: https://xgboost.readthedocs.io/en/latest/python/callbacks.html
warnings.warn(f'Old style callback is deprecated. See: {link}', UserWarning)
Get devices for measurement successfully!
- .T
+
@@ -1372,7 +1580,7 @@ In the example below we resume the status and do more 5 trials.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 5 minutes 58.328 seconds)
+ **Total running time of the script:** ( 5 minutes 48.187 seconds)
.. _sphx_glr_download_how_to_tune_with_autoscheduler_tune_conv2d_layer_cuda.py:
diff --git a/docs/_sources/how_to/tune_with_autoscheduler/tune_network_cuda.rst.txt b/docs/_sources/how_to/tune_with_autoscheduler/tune_network_cuda.rst.txt
index 32b0a3f7a6..ecc7bce088 100644
--- a/docs/_sources/how_to/tune_with_autoscheduler/tune_network_cuda.rst.txt
+++ b/docs/_sources/how_to/tune_with_autoscheduler/tune_network_cuda.rst.txt
@@ -647,7 +647,7 @@ so we can read the log file and load the best schedules.
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 7.8804 7.8801 7.8819 7.8793 0.0011
+ 7.8523 7.8530 7.8578 7.8460 0.0048
@@ -675,7 +675,7 @@ Other Tips
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 5.726 seconds)
+ **Total running time of the script:** ( 1 minutes 5.333 seconds)
.. _sphx_glr_download_how_to_tune_with_autoscheduler_tune_network_cuda.py:
diff --git a/docs/_sources/how_to/tune_with_autoscheduler/tune_network_x86.rst.txt b/docs/_sources/how_to/tune_with_autoscheduler/tune_network_x86.rst.txt
index 3141f25dae..9fa56dfe4a 100644
--- a/docs/_sources/how_to/tune_with_autoscheduler/tune_network_x86.rst.txt
+++ b/docs/_sources/how_to/tune_with_autoscheduler/tune_network_x86.rst.txt
@@ -666,7 +666,7 @@ so we can read the log file and load the best schedules.
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 786.5811 789.7259 790.2548 779.7626 4.8262
+ 752.2132 751.8728 754.3859 750.3809 1.6526
@@ -694,7 +694,7 @@ Other Tips
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 41.158 seconds)
+ **Total running time of the script:** ( 1 minutes 38.178 seconds)
.. _sphx_glr_download_how_to_tune_with_autoscheduler_tune_network_x86.py:
diff --git a/docs/_sources/how_to/tune_with_autoscheduler/tune_sparse_x86.rst.txt b/docs/_sources/how_to/tune_with_autoscheduler/tune_sparse_x86.rst.txt
index 63f0f1ba88..711d020e2d 100644
--- a/docs/_sources/how_to/tune_with_autoscheduler/tune_sparse_x86.rst.txt
+++ b/docs/_sources/how_to/tune_with_autoscheduler/tune_sparse_x86.rst.txt
@@ -386,74 +386,27 @@ layout transformation, parallelization, vectorization, unrolling, and operator f
@T.prim_func
def main(placeholder: T.Buffer((128, 256), "float32"), placeholder_1: T.Buffer((4916, 16, 1), "float32"), placeholder_2: T.Buffer((4916,), "int32"), placeholder_3: T.Buffer((33,), "int32"), placeholder_4: T.Buffer((128, 512), "float32"), compute: T.Buffer((128, 512), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
- for i0_outer_i1_outer_fused in T.parallel(64):
- compute_1 = T.allocate([1024], "float32", "global")
- compute_2 = T.buffer_decl((1024,), data=compute_1)
- for i_outer_inner, nb_j_inner in T.grid(2, 2):
- for i_inner_init in range(16):
- cse_var_1: T.int32 = i_outer_inner * 512 + i_inner_init * 32 + nb_j_inner * 16
- compute_2[cse_var_1] = T.float32(0)
- compute_2[cse_var_1 + 1] = T.float32(0)
- compute_2[cse_var_1 + 2] = T.float32(0)
- compute_2[cse_var_1 + 3] = T.float32(0)
- compute_2[cse_var_1 + 4] = T.float32(0)
- compute_2[cse_var_1 + 5] = T.float32(0)
- compute_2[cse_var_1 + 6] = T.float32(0)
- compute_2[cse_var_1 + 7] = T.float32(0)
- compute_2[cse_var_1 + 8] = T.float32(0)
- compute_2[cse_var_1 + 9] = T.float32(0)
- compute_2[cse_var_1 + 10] = T.float32(0)
- compute_2[cse_var_1 + 11] = T.float32(0)
- compute_2[cse_var_1 + 12] = T.float32(0)
- compute_2[cse_var_1 + 13] = T.float32(0)
- compute_2[cse_var_1 + 14] = T.float32(0)
- compute_2[cse_var_1 + 15] = T.float32(0)
- for elem_idx, i_inner in T.grid(T.let(cse_var_2, i0_outer_i1_outer_fused % 16 * 2 + nb_j_inner, placeholder_5[cse_var_2 + 1] - placeholder_5[cse_var_2]), 16):
- cse_var_2 = T.var("int32")
+ for i0_outer_i1_outer_fused in T.parallel(512):
+ compute_1 = T.allocate([128], "float32", "global")
+ compute_2 = T.buffer_decl((128,), data=compute_1)
+ for i_outer_inner in range(2):
+ for i_inner_init, j_init in T.grid(4, 16):
+ compute_2[i_outer_inner * 64 + i_inner_init * 16 + j_init] = T.float32(0)
+ for elem_idx, i_inner, j in T.grid(T.let(cse_var_1, i0_outer_i1_outer_fused % 32, placeholder_5[cse_var_1 + 1] - placeholder_5[cse_var_1]), 4, 16):
+ cse_var_1 = T.var("int32")
placeholder_5 = T.buffer_decl((33,), "int32", data=placeholder_3.data)
- cse_var_21: T.int32 = elem_idx * 16
- cse_var_20: T.int32 = i0_outer_i1_outer_fused % 16 * 2 + nb_j_inner
- cse_var_19: T.int32 = i_outer_inner * 512 + i_inner * 32 + nb_j_inner * 16
- cse_var_18: T.int32 = i0_outer_i1_outer_fused // 16 * 8192 + i_outer_inner * 4096 + i_inner * 256
- cse_var_17: T.int32 = cse_var_19 + 9
- cse_var_16: T.int32 = cse_var_19 + 8
- cse_var_15: T.int32 = cse_var_19 + 7
- cse_var_14: T.int32 = cse_var_19 + 6
- cse_var_13: T.int32 = cse_var_19 + 5
- cse_var_12: T.int32 = cse_var_19 + 4
- cse_var_11: T.int32 = cse_var_19 + 3
- cse_var_10: T.int32 = cse_var_19 + 2
- cse_var_9: T.int32 = cse_var_19 + 15
- cse_var_8: T.int32 = cse_var_19 + 14
- cse_var_7: T.int32 = cse_var_19 + 13
- cse_var_6: T.int32 = cse_var_19 + 12
- cse_var_5: T.int32 = cse_var_19 + 11
- cse_var_4: T.int32 = cse_var_19 + 10
- cse_var_3: T.int32 = cse_var_19 + 1
- placeholder_6 = T.buffer_decl((78656,), data=placeholder_1.data)
- placeholder_7 = T.buffer_decl((32768,), data=placeholder.data)
- placeholder_8 = T.buffer_decl((4916,), "int32", data=placeholder_2.data)
- compute_2[cse_var_19] = compute_2[cse_var_19] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_3] = compute_2[cse_var_3] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 1] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_10] = compute_2[cse_var_10] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 2] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_11] = compute_2[cse_var_11] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 3] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_12] = compute_2[cse_var_12] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 4] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_13] = compute_2[cse_var_13] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 5] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_14] = compute_2[cse_var_14] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 6] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_15] = compute_2[cse_var_15] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 7] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_16] = compute_2[cse_var_16] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 8] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_17] = compute_2[cse_var_17] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 9] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_4] = compute_2[cse_var_4] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 10] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_5] = compute_2[cse_var_5] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 11] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_6] = compute_2[cse_var_6] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 12] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_7] = compute_2[cse_var_7] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 13] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_8] = compute_2[cse_var_8] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 14] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_9] = compute_2[cse_var_9] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 15] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- for i0_inner, i1_inner in T.grid(32, 32):
- cse_var_22: T.int32 = i0_outer_i1_outer_fused // 16 * 16384 + i0_inner * 512 + i0_outer_i1_outer_fused % 16 * 32 + i1_inner
+ cse_var_2: T.int32 = i0_outer_i1_outer_fused % 32
+ if T.likely(elem_idx < placeholder_5[cse_var_2 + 1] - placeholder_5[cse_var_2]):
+ placeholder_6 = T.buffer_decl((78656,), data=placeholder_1.data)
+ placeholder_7 = T.buffer_decl((32768,), data=placeholder.data)
+ placeholder_8 = T.buffer_decl((4916,), "int32", data=placeholder_2.data)
+ cse_var_3: T.int32 = i_outer_inner * 64 + i_inner * 16 + j
+ compute_2[cse_var_3] = compute_2[cse_var_3] + placeholder_6[placeholder_5[cse_var_2] * 16 + elem_idx * 16 + j] * T.max(placeholder_7[i0_outer_i1_outer_fused // 32 * 2048 + i_outer_inner * 1024 + i_inner * 256 + placeholder_8[placeholder_5[cse_var_2] + elem_idx]], T.float32(0))
+ for i0_inner in range(8):
+ cse_var_4: T.int32 = i0_outer_i1_outer_fused // 32 * 4096 + i0_inner * 512 + i0_outer_i1_outer_fused % 32 * 16
compute_3 = T.buffer_decl((65536,), data=compute.data)
placeholder_5 = T.buffer_decl((65536,), data=placeholder_4.data)
- compute_3[cse_var_22] = T.max(compute_2[i0_inner * 32 + i1_inner] + placeholder_5[cse_var_22], T.float32(0))
+ compute_3[cse_var_4:cse_var_4 + 16] = T.max(compute_2[i0_inner * 16:i0_inner * 16 + 16] + placeholder_5[cse_var_4:cse_var_4 + 16], T.Broadcast(T.float32(0), 16))
@@ -503,7 +456,7 @@ We build the binary and check its correctness and performance.
.. code-block:: none
- Execution time of this operator: 1.732 ms
+ Execution time of this operator: 1.362 ms
diff --git a/docs/_sources/how_to/tune_with_autotvm/sg_execution_times.rst.txt b/docs/_sources/how_to/tune_with_autotvm/sg_execution_times.rst.txt
index a02f0eb220..6cf3d73032 100644
--- a/docs/_sources/how_to/tune_with_autotvm/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/tune_with_autotvm/sg_execution_times.rst.txt
@@ -5,16 +5,16 @@
Computation times
=================
-**00:49.910** total execution time for **how_to_tune_with_autotvm** files:
+**00:32.704** total execution time for **how_to_tune_with_autotvm** files:
+--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_conv2d_cuda.py` (``tune_conv2d_cuda.py``) | 00:49.876 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_conv2d_cuda.py` (``tune_conv2d_cuda.py``) | 00:32.668 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_x86.py` (``tune_relay_x86.py``) | 00:00.020 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_x86.py` (``tune_relay_x86.py``) | 00:00.023 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_cuda.py` (``tune_relay_cuda.py``) | 00:00.005 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_arm.py` (``tune_relay_arm.py``) | 00:00.005 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_arm.py` (``tune_relay_arm.py``) | 00:00.004 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_mobile_gpu.py` (``tune_relay_mobile_gpu.py``) | 00:00.004 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/tune_with_autotvm/tune_conv2d_cuda.rst.txt b/docs/_sources/how_to/tune_with_autotvm/tune_conv2d_cuda.rst.txt
index 0ac25918c1..431cc075cb 100644
--- a/docs/_sources/how_to/tune_with_autotvm/tune_conv2d_cuda.rst.txt
+++ b/docs/_sources/how_to/tune_with_autotvm/tune_conv2d_cuda.rst.txt
@@ -268,12 +268,7 @@ for this template
waiting for device...
device available
Get devices for measurement successfully!
- No: 1 GFLOPS: 7.37/7.37 result: MeasureResult(costs=(0.0314321365,), error_no=MeasureErrorNo.NO_ERROR, all_cost=7.174636363983154, timestamp=1674259112.8968542) [('tile_f', [-1, 1, 8, 8]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,1826814
- No: 2 GFLOPS: 39.80/39.80 result: MeasureResult(costs=(0.005816791055555556,), error_no=MeasureErrorNo.NO_ERROR, all_cost=4.863697290420532, timestamp=1674259114.688778) [('tile_f', [-1, 32, 16, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,8944359
- No: 3 GFLOPS: 4.60/39.80 result: MeasureResult(costs=(0.05034223175,), error_no=MeasureErrorNo.NO_ERROR, all_cost=10.43713903427124, timestamp=1674259116.8111053) [('tile_f', [-1, 16, 1, 4]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 32]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7500784
- No: 4 GFLOPS: 45.33/45.33 result: MeasureResult(costs=(0.0051065872,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.4066011905670166, timestamp=1674259117.5673692) [('tile_f', [-1, 1, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9972820
- No: 5 GFLOPS: 35.24/45.33 result: MeasureResult(costs=(0.0065700821874999996,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.415250539779663, timestamp=1674259119.1511686) [('tile_f', [-1, 4, 4, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 1]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1164481
- No: 6 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ No: 1 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -395,8 +390,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 128, 1, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 64, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9350502
- No: 7 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 2, 16]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,276272
+ No: 2 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -518,8 +513,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 1, 4]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 8, 32]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,732264
- No: 8 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 32, 2, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6215289
+ No: 3 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -641,27 +636,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 256]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2312972
- No: 9 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
- File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 142, in build
- res = future.result()
- File "/usr/lib/python3.7/concurrent/futures/_base.py", line 435, in result
- return self.__get_result()
- File "/usr/lib/python3.7/concurrent/futures/_base.py", line 384, in __get_result
- raise self._exception
- File "/usr/lib/python3.7/concurrent/futures/thread.py", line 57, in run
- result = self.fn(*self.args, **self.kwargs)
- File "/workspace/python/tvm/contrib/popen_pool.py", line 432, in <lambda>
- worker = lambda *args: self._worker_run(*args)
- File "/workspace/python/tvm/contrib/popen_pool.py", line 401, in _worker_run
- return proc.recv()
- File "/workspace/python/tvm/contrib/popen_pool.py", line 309, in recv
- raise TimeoutError()
- TimeoutError
-
- [('tile_f', [-1, 8, 4, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5661335
- No: 10 GFLOPS: 12.84/45.33 result: MeasureResult(costs=(0.018029925,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3179340362548828, timestamp=1674259131.2934096) [('tile_f', [-1, 1, 4, 8]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3305429
- No: 11 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 8, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 16]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7689467
+ No: 4 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -783,8 +759,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 32]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6286468
- No: 12 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 32, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 512, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6808873
+ No: 5 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -906,8 +882,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 512, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7979629
- No: 13 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 8, 8]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 32, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4568457
+ No: 6 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1029,9 +1005,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 1, 512]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,10452874
- No: 14 GFLOPS: 195.22/195.22 result: MeasureResult(costs=(0.001185834181818182,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.4841997623443604, timestamp=1674259134.1751637) [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 1]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,198706
- No: 15 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 128, 1, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 32]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5963207
+ No: 7 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1153,9 +1128,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 16, 8]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7375440
- No: 16 GFLOPS: 63.25/195.22 result: MeasureResult(costs=(0.0036600480714285713,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3547515869140625, timestamp=1674259134.9343917) [('tile_f', [-1, 2, 16, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5460119
- No: 17 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4614419
+ No: 8 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1277,8 +1251,8 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 32, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 8, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6823832
- No: 18 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 4, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 16]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6714109
+ No: 9 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1400,9 +1374,9 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 64, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,83868
- No: 19 GFLOPS: 7.73/195.22 result: MeasureResult(costs=(0.029934080000000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.8496928215026855, timestamp=1674259136.9989753) [('tile_f', [-1, 16, 2, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5328248
- No: 20 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 128, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6258341
+ No: 10 GFLOPS: 29.67/29.67 result: MeasureResult(costs=(0.007802013928571429,), error_no=MeasureErrorNo.NO_ERROR, all_cost=5.3617002964019775, timestamp=1674265444.7841802) [('tile_f', [-1, 1, 4, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4741854
+ No: 11 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1524,7 +1498,992 @@ for this template
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
- tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 8, 16]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9229508
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 2, 128]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4631266
+ No: 12 GFLOPS: 5.74/29.67 result: MeasureResult(costs=(0.0403490255,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.486266613006592, timestamp=1674265445.7363422) [('tile_f', [-1, 4, 2, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 32]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,3819872
+ No: 13 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 64]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7131503
+ No: 14 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 32, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 1, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2360641
+ No: 15 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 4, 64]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4543428
+ No: 16 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 32, 16]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 8, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7077804
+ No: 17 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 8, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 2, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3170723
+ No: 18 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 32, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 64, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6447230
+ No: 19 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 32, 8, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9260712
+ No: 20 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+ tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+ Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+ tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 2, 8]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,5143525
@@ -1579,9 +2538,9 @@ and measure running time.
Finish loading 20 records
Best config:
- [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 1]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,198706
+ [('tile_f', [-1, 1, 4, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4741854
Finish loading 20 records
- Time cost of this operator: 0.001517
+ Time cost of this operator: 0.007964
diff --git a/docs/_sources/how_to/work_with_microtvm/micro_autotune.rst.txt b/docs/_sources/how_to/work_with_microtvm/micro_autotune.rst.txt
index 80bd678397..d0f4b181ad 100644
--- a/docs/_sources/how_to/work_with_microtvm/micro_autotune.rst.txt
+++ b/docs/_sources/how_to/work_with_microtvm/micro_autotune.rst.txt
@@ -363,10 +363,10 @@ Timing the untuned program
########## Build without Autotuning ##########
Node Name Ops Time(us) Time(%) Shape Inputs Outputs Measurements(us)
--------- --- -------- ------- ----- ------ ------- ----------------
- tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 312.2 98.726 (1, 2, 10, 10, 3) 2 1 [312.2]
- tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 3.043 0.962 (1, 6, 10, 10) 1 1 [3.043]
- tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.985 0.311 (1, 1, 10, 10, 3) 1 1 [0.985]
- Total_time - 316.228 - - - - -
+ tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 311.5 98.724 (1, 2, 10, 10, 3) 2 1 [311.5]
+ tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 3.026 0.959 (1, 6, 10, 10) 1 1 [3.026]
+ tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 1.001 0.317 (1, 1, 10, 10, 3) 1 1 [1.001]
+ Total_time - 315.527 - - - - -
@@ -431,10 +431,10 @@ Timing the tuned program
########## Build with Autotuning ##########
Node Name Ops Time(us) Time(%) Shape Inputs Outputs Measurements(us)
--------- --- -------- ------- ----- ------ ------- ----------------
- tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 139.3 98.131 (1, 6, 10, 10, 1) 2 1 [139.3]
- tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 1.777 1.252 (1, 6, 10, 10) 1 1 [1.777]
- tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.877 0.618 (1, 3, 10, 10, 1) 1 1 [0.877]
- Total_time - 141.954 - - - - -
+ tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 102.6 97.448 (1, 6, 10, 10, 1) 2 1 [102.6]
+ tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 1.829 1.737 (1, 6, 10, 10) 1 1 [1.829]
+ tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.858 0.815 (1, 3, 10, 10, 1) 1 1 [0.858]
+ Total_time - 105.287 - - - - -
diff --git a/docs/_sources/how_to/work_with_microtvm/micro_pytorch.rst.txt b/docs/_sources/how_to/work_with_microtvm/micro_pytorch.rst.txt
index 274964bbd0..6a56fc8dc4 100644
--- a/docs/_sources/how_to/work_with_microtvm/micro_pytorch.rst.txt
+++ b/docs/_sources/how_to/work_with_microtvm/micro_pytorch.rst.txt
@@ -117,7 +117,7 @@ download a cat image and preprocess it to use as the model input.
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torch/ao/quantization/utils.py:281: UserWarning: must run observer before calling calculate_qparams. Returning default values.
"must run observer before calling calculate_qparams. " +
Downloading: "https://download.pytorch.org/models/quantized/mobilenet_v2_qnnpack_37f702c5.pth" to /workspace/.cache/torch/hub/checkpoints/mobilenet_v2_qnnpack_37f702c5.pth
-
0%| | 0.00/3.42M [00:00<?, ?B/s]
100%|##########| 3.42M/3.42M [00:00<00:00, 81.6MB/s]
+
0%| | 0.00/3.42M [00:00<?, ?B/s]
100%|##########| 3.42M/3.42M [00:00<00:00, 57.0MB/s]
/workspace/python/tvm/relay/frontend/pytorch_utils.py:47: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
return LooseVersion(torch_ver) > ver
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
@@ -322,7 +322,7 @@ Look up prediction top 1 index in 1000 class synset.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 13.811 seconds)
+ **Total running time of the script:** ( 1 minutes 13.423 seconds)
.. _sphx_glr_download_how_to_work_with_microtvm_micro_pytorch.py:
diff --git a/docs/_sources/how_to/work_with_microtvm/micro_train.rst.txt b/docs/_sources/how_to/work_with_microtvm/micro_train.rst.txt
index 8a2f7ab704..32e7fc4bfa 100644
--- a/docs/_sources/how_to/work_with_microtvm/micro_train.rst.txt
+++ b/docs/_sources/how_to/work_with_microtvm/micro_train.rst.txt
@@ -218,7 +218,7 @@ take about **2 minutes** to download the Stanford Cars, while COCO 2017 validati
.. code-block:: none
- '/tmp/tmpciq06jqh/images/random'
+ '/tmp/tmp8v5ftiwe/images/random'
@@ -309,7 +309,7 @@ objects to other stuff? We can display some examples from our datasets using ``m
.. image-sg:: /how_to/work_with_microtvm/images/sphx_glr_micro_train_001.png
- :alt: [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0], [1.0, 0.0]
+ :alt: [0.0, 1.0], [1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0]
:srcset: /how_to/work_with_microtvm/images/sphx_glr_micro_train_001.png
:class: sphx-glr-single-img
@@ -318,8 +318,8 @@ objects to other stuff? We can display some examples from our datasets using ``m
.. code-block:: none
- /tmp/tmpciq06jqh/images/target contains 8144 images
- /tmp/tmpciq06jqh/images/random contains 5000 images
+ /tmp/tmp8v5ftiwe/images/target contains 8144 images
+ /tmp/tmp8v5ftiwe/images/random contains 5000 images
@@ -494,13 +494,13 @@ the time on our validation set).
.. code-block:: none
Epoch 1/3
- 328/328 - 47s - loss: 0.2233 - accuracy: 0.9254 - val_loss: 0.1195 - val_accuracy: 0.9607 - 47s/epoch - 144ms/step
+ 328/328 - 47s - loss: 0.2200 - accuracy: 0.9250 - val_loss: 0.1157 - val_accuracy: 0.9600 - 47s/epoch - 143ms/step
Epoch 2/3
- 328/328 - 44s - loss: 0.1003 - accuracy: 0.9639 - val_loss: 0.1247 - val_accuracy: 0.9592 - 44s/epoch - 133ms/step
+ 328/328 - 44s - loss: 0.0974 - accuracy: 0.9643 - val_loss: 0.1420 - val_accuracy: 0.9577 - 44s/epoch - 133ms/step
Epoch 3/3
- 328/328 - 44s - loss: 0.0676 - accuracy: 0.9764 - val_loss: 0.1159 - val_accuracy: 0.9637 - 44s/epoch - 133ms/step
+ 328/328 - 43s - loss: 0.0648 - accuracy: 0.9753 - val_loss: 0.1082 - val_accuracy: 0.9656 - 43s/epoch - 132ms/step
- <keras.callbacks.History object at 0x7f0a48b6c0d0>
+ <keras.callbacks.History object at 0x7fb5a9e8b910>
@@ -857,7 +857,7 @@ Arduino tutorial for how to do that `on GitHub <https://github.com/guberti/tvm-a
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 4 minutes 10.207 seconds)
+ **Total running time of the script:** ( 4 minutes 20.715 seconds)
.. _sphx_glr_download_how_to_work_with_microtvm_micro_train.py:
diff --git a/docs/_sources/how_to/work_with_microtvm/sg_execution_times.rst.txt b/docs/_sources/how_to/work_with_microtvm/sg_execution_times.rst.txt
index 12f3586883..84c2ca9fd9 100644
--- a/docs/_sources/how_to/work_with_microtvm/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/work_with_microtvm/sg_execution_times.rst.txt
@@ -5,18 +5,18 @@
Computation times
=================
-**06:34.453** total execution time for **how_to_work_with_microtvm** files:
+**06:40.241** total execution time for **how_to_work_with_microtvm** files:
+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_train.py` (``micro_train.py``) | 04:10.207 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_train.py` (``micro_train.py``) | 04:20.715 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_pytorch.py` (``micro_pytorch.py``) | 01:13.811 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_pytorch.py` (``micro_pytorch.py``) | 01:13.423 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_autotune.py` (``micro_autotune.py``) | 00:56.653 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_autotune.py` (``micro_autotune.py``) | 00:52.960 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_aot.py` (``micro_aot.py``) | 00:09.495 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_aot.py` (``micro_aot.py``) | 00:08.927 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_tflite.py` (``micro_tflite.py``) | 00:04.288 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_tflite.py` (``micro_tflite.py``) | 00:04.215 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_how_to_work_with_microtvm_micro_ethosu.py` (``micro_ethosu.py``) | 00:00.000 | 0.0 MB |
+---------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/work_with_relay/sg_execution_times.rst.txt b/docs/_sources/how_to/work_with_relay/sg_execution_times.rst.txt
index 185ea9b335..11a1cf2f2e 100644
--- a/docs/_sources/how_to/work_with_relay/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/work_with_relay/sg_execution_times.rst.txt
@@ -5,14 +5,14 @@
Computation times
=================
-**00:46.941** total execution time for **how_to_work_with_relay** files:
+**00:44.962** total execution time for **how_to_work_with_relay** files:
+----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_using_pipeline_executor.py` (``using_pipeline_executor.py``) | 00:34.745 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_using_pipeline_executor.py` (``using_pipeline_executor.py``) | 00:32.792 | 0.0 MB |
+----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_using_external_lib.py` (``using_external_lib.py``) | 00:10.468 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_using_external_lib.py` (``using_external_lib.py``) | 00:10.380 | 0.0 MB |
+----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_build_gcn.py` (``build_gcn.py``) | 00:01.721 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_build_gcn.py` (``build_gcn.py``) | 00:01.782 | 0.0 MB |
+----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_using_relay_viz.py` (``using_relay_viz.py``) | 00:00.006 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_using_relay_viz.py` (``using_relay_viz.py``) | 00:00.007 | 0.0 MB |
+----------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/work_with_schedules/intrin_math.rst.txt b/docs/_sources/how_to/work_with_schedules/intrin_math.rst.txt
index 75f206f6bb..4f7239b856 100644
--- a/docs/_sources/how_to/work_with_schedules/intrin_math.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/intrin_math.rst.txt
@@ -264,7 +264,7 @@ The following example customizes CUDA lowering rule for :code:`exp`.
.. code-block:: none
- <function my_cuda_math_rule at 0x7f0adca84560>
+ <function my_cuda_math_rule at 0x7fb5a8d075f0>
diff --git a/docs/_sources/how_to/work_with_schedules/sg_execution_times.rst.txt b/docs/_sources/how_to/work_with_schedules/sg_execution_times.rst.txt
index 720387dbc8..f4311a2b8e 100644
--- a/docs/_sources/how_to/work_with_schedules/sg_execution_times.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/sg_execution_times.rst.txt
@@ -5,22 +5,22 @@
Computation times
=================
-**00:06.377** total execution time for **how_to_work_with_schedules** files:
+**00:07.164** total execution time for **how_to_work_with_schedules** files:
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_intrin_math.py` (``intrin_math.py``) | 00:03.680 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_intrin_math.py` (``intrin_math.py``) | 00:04.655 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_tensorize.py` (``tensorize.py``) | 00:01.202 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_tensorize.py` (``tensorize.py``) | 00:01.141 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_reduction.py` (``reduction.py``) | 00:00.641 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_reduction.py` (``reduction.py``) | 00:00.583 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_scan.py` (``scan.py``) | 00:00.620 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_scan.py` (``scan.py``) | 00:00.562 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_extern_op.py` (``extern_op.py``) | 00:00.121 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_extern_op.py` (``extern_op.py``) | 00:00.117 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_schedule_primitives.py` (``schedule_primitives.py``) | 00:00.056 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_schedule_primitives.py` (``schedule_primitives.py``) | 00:00.050 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_tedd.py` (``tedd.py``) | 00:00.032 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_tedd.py` (``tedd.py``) | 00:00.031 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_tuple_inputs.py` (``tuple_inputs.py``) | 00:00.025 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_tuple_inputs.py` (``tuple_inputs.py``) | 00:00.024 | 0.0 MB |
+------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt b/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt
index f4cf766891..90dd4cefeb 100644
--- a/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt
@@ -328,7 +328,7 @@ The importing needs to happen before the tensorized GEMV being executed.
def main(A: T.Buffer((1024, 64), "float32"), B: T.Buffer((512, 64), "float32"), C: T.Buffer((1024, 512), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
i = T.var("int32")
- T.attr(T.iter_var(i, None, "DataPar", ""), "pragma_import_llvm", "; ModuleID = '/tmp/tmpns4jb9wl/input0.cc'\nsource_filename = \"/tmp/tmpns4jb9wl/input0.cc\"\ntarget datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-pc-linux-gnu\"\n\n; Function Attrs: noinline nounwind optnone uwtable\ndefine dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {\n %7 = alloca float*, align 8\n %8 = alloca float*, align 8\n %9 = alloca float*, [...]
+ T.attr(T.iter_var(i, None, "DataPar", ""), "pragma_import_llvm", "; ModuleID = '/tmp/tmpwrd4vqy_/input0.cc'\nsource_filename = \"/tmp/tmpwrd4vqy_/input0.cc\"\ntarget datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-pc-linux-gnu\"\n\n; Function Attrs: noinline nounwind optnone uwtable\ndefine dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {\n %7 = alloca float*, align 8\n %8 = alloca float*, align 8\n %9 = alloca float*, [...]
for i, j_outer in T.grid(1024, 32):
T.call_extern("int32", "gemv_update", T.tvm_access_ptr(T.type_annotation("float32"), C.data, i * 512 + j_outer * 16, 16, 2), T.tvm_access_ptr(T.type_annotation("float32"), A.data, i * 64, 64, 1), T.tvm_access_ptr(T.type_annotation("float32"), B.data, j_outer * 1024, 1024, 1), 16, 64, 64)
diff --git a/docs/_sources/topic/vta/tutorials/autotvm/sg_execution_times.rst.txt b/docs/_sources/topic/vta/tutorials/autotvm/sg_execution_times.rst.txt
index 7b25f62e19..db53193c28 100644
--- a/docs/_sources/topic/vta/tutorials/autotvm/sg_execution_times.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/autotvm/sg_execution_times.rst.txt
@@ -5,10 +5,10 @@
Computation times
=================
-**00:31.127** total execution time for **topic_vta_tutorials_autotvm** files:
+**00:30.140** total execution time for **topic_vta_tutorials_autotvm** files:
+---------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``) | 00:31.120 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``) | 00:30.134 | 0.0 MB |
+---------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_alu_vta.py` (``tune_alu_vta.py``) | 00:00.007 | 0.0 MB |
+---------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/topic/vta/tutorials/frontend/deploy_classification.rst.txt b/docs/_sources/topic/vta/tutorials/frontend/deploy_classification.rst.txt
index 8d61c6d874..ecdc671202 100644
--- a/docs/_sources/topic/vta/tutorials/frontend/deploy_classification.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/frontend/deploy_classification.rst.txt
@@ -293,7 +293,7 @@ The compilation steps are:
DeprecationWarning,
/workspace/vta/tutorials/frontend/deploy_classification.py:213: DeprecationWarning: legacy graph executor behavior of producing json / lib / params will be removed in the next release. Please see documents of tvm.contrib.graph_executor.GraphModule for the new recommended usage.
relay_prog, target=tvm.target.Target(target, host=env.target_host), params=params
- resnet18_v1 inference graph built in 33.91s!
+ resnet18_v1 inference graph built in 32.87s!
diff --git a/docs/_sources/topic/vta/tutorials/frontend/deploy_detection.rst.txt b/docs/_sources/topic/vta/tutorials/frontend/deploy_detection.rst.txt
index 2d186ee35e..2901894303 100644
--- a/docs/_sources/topic/vta/tutorials/frontend/deploy_detection.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/frontend/deploy_detection.rst.txt
@@ -337,7 +337,7 @@ The compilation steps are:
/workspace/python/tvm/relay/build_module.py:348: DeprecationWarning: Please use input parameter mod (tvm.IRModule) instead of deprecated parameter mod (tvm.relay.function.Function)
DeprecationWarning,
- yolov3-tiny inference graph built in 22.61s!
+ yolov3-tiny inference graph built in 22.38s!
diff --git a/docs/_sources/topic/vta/tutorials/frontend/sg_execution_times.rst.txt b/docs/_sources/topic/vta/tutorials/frontend/sg_execution_times.rst.txt
index 8274b2d4a5..cdaae8670b 100644
--- a/docs/_sources/topic/vta/tutorials/frontend/sg_execution_times.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/frontend/sg_execution_times.rst.txt
@@ -5,10 +5,10 @@
Computation times
=================
-**01:40.229** total execution time for **topic_vta_tutorials_frontend** files:
+**01:39.092** total execution time for **topic_vta_tutorials_frontend** files:
+------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``) | 00:50.950 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``) | 00:49.699 | 0.0 MB |
+------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``) | 00:49.279 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``) | 00:49.393 | 0.0 MB |
+------------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/topic/vta/tutorials/optimize/sg_execution_times.rst.txt b/docs/_sources/topic/vta/tutorials/optimize/sg_execution_times.rst.txt
index 036fb4a5dd..921e19e6d0 100644
--- a/docs/_sources/topic/vta/tutorials/optimize/sg_execution_times.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/optimize/sg_execution_times.rst.txt
@@ -5,10 +5,10 @@
Computation times
=================
-**00:03.277** total execution time for **topic_vta_tutorials_optimize** files:
+**00:03.173** total execution time for **topic_vta_tutorials_optimize** files:
+--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``) | 00:02.747 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``) | 00:02.710 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``) | 00:00.530 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``) | 00:00.463 | 0.0 MB |
+--------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/topic/vta/tutorials/sg_execution_times.rst.txt b/docs/_sources/topic/vta/tutorials/sg_execution_times.rst.txt
index 700a62f935..ec7dc4b7ae 100644
--- a/docs/_sources/topic/vta/tutorials/sg_execution_times.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/sg_execution_times.rst.txt
@@ -5,10 +5,10 @@
Computation times
=================
-**00:00.941** total execution time for **topic_vta_tutorials** files:
+**00:00.845** total execution time for **topic_vta_tutorials** files:
+---------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``) | 00:00.494 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``) | 00:00.457 | 0.0 MB |
+---------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_vta_get_started.py` (``vta_get_started.py``) | 00:00.447 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_vta_get_started.py` (``vta_get_started.py``) | 00:00.388 | 0.0 MB |
+---------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt b/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
index 65769d33b3..9f39314f6c 100644
--- a/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
+++ b/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
@@ -211,7 +211,7 @@ trials, we can load the best schedule from the log file and apply it.
.. code-block:: none
- .T*E
+ .T
@@ -322,7 +322,7 @@ We build the binary and check its correctness and performance.
.. code-block:: none
- Execution time of this operator: 95.337 ms
+ Execution time of this operator: 93.065 ms
@@ -440,7 +440,7 @@ operations.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 1 minutes 36.515 seconds)
+ **Total running time of the script:** ( 1 minutes 24.802 seconds)
.. _sphx_glr_download_tutorial_auto_scheduler_matmul_x86.py:
diff --git a/docs/_sources/tutorial/autotvm_matmul_x86.rst.txt b/docs/_sources/tutorial/autotvm_matmul_x86.rst.txt
index 9bb962037c..f472857a49 100644
--- a/docs/_sources/tutorial/autotvm_matmul_x86.rst.txt
+++ b/docs/_sources/tutorial/autotvm_matmul_x86.rst.txt
@@ -454,16 +454,16 @@ reduce variance, we take 5 measurements and average them.
waiting for device...
device available
Get devices for measurement successfully!
- No: 1 GFLOPS: 3.86/3.86 result: MeasureResult(costs=(0.0694778412,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.407578468322754, timestamp=1674257565.2049582) [('tile_y', [-1, 32]), ('tile_x', [-1, 16])],None,45
- No: 2 GFLOPS: 3.24/3.86 result: MeasureResult(costs=(0.0827394284,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.5787849426269531, timestamp=1674257567.6051075) [('tile_y', [-1, 2]), ('tile_x', [-1, 8])],None,31
- No: 3 GFLOPS: 2.55/3.86 result: MeasureResult(costs=(0.10528658259999998,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.9338383674621582, timestamp=1674257570.3779185) [('tile_y', [-1, 4]), ('tile_x', [-1, 4])],None,22
- No: 4 GFLOPS: 1.94/3.86 result: MeasureResult(costs=(0.1387193156,), error_no=MeasureErrorNo.NO_ERROR, all_cost=2.4521687030792236, timestamp=1674257572.855061) [('tile_y', [-1, 2]), ('tile_x', [-1, 4])],None,21
- No: 5 GFLOPS: 6.47/6.47 result: MeasureResult(costs=(0.0415190078,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.8962502479553223, timestamp=1674257573.9248767) [('tile_y', [-1, 512]), ('tile_x', [-1, 32])],None,59
- No: 6 GFLOPS: 7.76/7.76 result: MeasureResult(costs=(0.034614458599999995,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.8558461666107178, timestamp=1674257574.7479596) [('tile_y', [-1, 512]), ('tile_x', [-1, 64])],None,69
- No: 7 GFLOPS: 12.49/12.49 result: MeasureResult(costs=(0.021490214599999998,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.611403226852417, timestamp=1674257576.1711226) [('tile_y', [-1, 4]), ('tile_x', [-1, 256])],None,82
- No: 8 GFLOPS: 3.27/12.49 result: MeasureResult(costs=(0.0821217014,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.573803186416626, timestamp=1674257577.7435343) [('tile_y', [-1, 32]), ('tile_x', [-1, 8])],None,35
- No: 9 GFLOPS: 3.39/12.49 result: MeasureResult(costs=(0.0791173016,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.480146884918213, timestamp=1674257579.339175) [('tile_y', [-1, 8]), ('tile_x', [-1, 8])],None,33
- No: 10 GFLOPS: 10.63/12.49 result: MeasureResult(costs=(0.0252517118,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6148405075073242, timestamp=1674257580.0078804) [('tile_y', [-1, 1]), ('tile_x', [-1, 512])],None,90
+ No: 1 GFLOPS: 1.18/1.18 result: MeasureResult(costs=(0.2268070816,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.875448703765869, timestamp=1674263923.5000813) [('tile_y', [-1, 16]), ('tile_x', [-1, 1])],None,4
+ No: 2 GFLOPS: 12.65/12.65 result: MeasureResult(costs=(0.021222059999999997,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.5897138118743896, timestamp=1674263924.8830245) [('tile_y', [-1, 64]), ('tile_x', [-1, 512])],None,96
+ No: 3 GFLOPS: 3.66/12.65 result: MeasureResult(costs=(0.0734391612,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.4268817901611328, timestamp=1674263926.3084638) [('tile_y', [-1, 256]), ('tile_x', [-1, 16])],None,48
+ No: 4 GFLOPS: 13.46/13.46 result: MeasureResult(costs=(0.019938440600000003,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.5770606994628906, timestamp=1674263927.673345) [('tile_y', [-1, 16]), ('tile_x', [-1, 512])],None,94
+ No: 5 GFLOPS: 0.90/13.46 result: MeasureResult(costs=(0.2993066266,), error_no=MeasureErrorNo.NO_ERROR, all_cost=5.030872821807861, timestamp=1674263932.9322677) [('tile_y', [-1, 128]), ('tile_x', [-1, 2])],None,17
+ No: 6 GFLOPS: 9.89/13.46 result: MeasureResult(costs=(0.0271290068,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.7892274856567383, timestamp=1674263933.622779) [('tile_y', [-1, 8]), ('tile_x', [-1, 128])],None,73
+ No: 7 GFLOPS: 2.56/13.46 result: MeasureResult(costs=(0.10480563579999999,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.9260509014129639, timestamp=1674263936.3236682) [('tile_y', [-1, 8]), ('tile_x', [-1, 4])],None,23
+ No: 8 GFLOPS: 12.56/13.46 result: MeasureResult(costs=(0.021375914599999997,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6036906242370605, timestamp=1674263936.9227962) [('tile_y', [-1, 32]), ('tile_x', [-1, 128])],None,75
+ No: 9 GFLOPS: 2.69/13.46 result: MeasureResult(costs=(0.09967235960000001,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.8228483200073242, timestamp=1674263939.0940313) [('tile_y', [-1, 512]), ('tile_x', [-1, 8])],None,39
+ No: 10 GFLOPS: 9.78/13.46 result: MeasureResult(costs=(0.027446748799999997,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.9058380126953125, timestamp=1674263939.787594) [('tile_y', [-1, 2]), ('tile_x', [-1, 128])],None,71
diff --git a/docs/_sources/tutorial/autotvm_relay_x86.rst.txt b/docs/_sources/tutorial/autotvm_relay_x86.rst.txt
index 391e310338..bcbbec52c7 100644
--- a/docs/_sources/tutorial/autotvm_relay_x86.rst.txt
+++ b/docs/_sources/tutorial/autotvm_relay_x86.rst.txt
@@ -311,7 +311,7 @@ standard deviation.
.. code-block:: none
- {'mean': 519.8523906099763, 'median': 519.4432064999091, 'std': 1.8586737969938754}
+ {'mean': 518.649112039999, 'median': 518.6342771, 'std': 1.203910214067869}
@@ -545,31 +545,31 @@ the tuning data to.
.. code-block:: none
-
[Task 1/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 1/25] Current/Best: 17.89/ 17.89 GFLOPS | Progress: (4/20) | 9.06 s
[Task 1/25] Current/Best: 1.90/ 19.54 GFLOPS | Progress: (8/20) | 13.40 s
[Task 1/25] Current/Best: 23.03/ 23.03 GFLOPS | Progress: (12/20) | 17.01 s
[Task 1/25] Current/Best: 8.99/ 23.03 GFLOPS | Progress: (16/20) | 20.73 s
[Task 1/25] Current/Best: 7.01/ 23.03 GFLOPS | Progress: (20/20) | 22.99 s Done.
-
[Task 2/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 2/25] Current/Best: 16.71/ 18.36 GFLOPS | Progress: (4/20) | 3.44 s
[Task 2/25] Current/Best: 21.56/ 21.56 GFLOPS | Progress: (8/20) | 4.96 s
[Task 2/25] Current/Best: 15.69/ 21.56 GFLOPS | Progress: (12/20) | 6.75 s
[Task 2/25] Current/Best: 17.21/ 21.56 GFLOPS | Progress: (16/20) | 8.44 s
[Task 2/25] Current/Best: 4.33/ 21.56 GFLOPS | Progress: (20/20) | 10.36 s Done.
-
[Task 3/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 3/25] Current/Best: 15.01/ 16.21 GFLOPS | Progress: (4/20) | 4.81 s
[Task 3/25] Current/Best: 6.47/ 18.56 GFLOPS | Progress: (8/20) | 7.28 s
[Task 3/25] Current/Best: 17.13/ 19.00 GFLOPS | Progress: (12/20) | 9.36 s
[Task 3/25] Current/Best: 8.34/ 19.00 GFLOPS | Progress: (16/20) | 11.76 s
[Task 3/25] Current/Best: 14.91/ 19.00 GFLOPS | Progress: (20/20) | 14.33 s Done.
-
[Task 4/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 4/25] Current/Best: 12.66/ 14.08 GFLOPS | Progress: (4/20) | 7.37 s
[Task 4/25] Current/Best: 7.95/ 14.08 GFLOPS | Progress: (8/20) | 9.68 s
[Task 4/25] Current/Best: 14.81/ 18.98 GFLOPS | Progress: (12/20) | 14.43 s
[Task 4/25] Current/Best: 16.63/ 18.98 GFLOPS | Progress: (16/20) | 16.61 s
[Task 4/25] Current/Best: 12.26/ 18.98 GFLOPS | Progress: (20/20) | 27.69 s
[Task 5/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
-
[Task 5/25] Current/Best: 9.61/ 14.52 GFLOPS | Progress: (4/20) | 4.50 s
[Task 5/25] Current/Best: 13.40/ 15.84 GFLOPS | Progress: (8/20) | 6.88 s
[Task 5/25] Current/Best: 13.44/ 15.84 GFLOPS | Progress: (12/20) | 9.44 s
[Task 5/25] Current/Best: 12.27/ 17.62 GFLOPS | Progress: (16/20) | 11.41 s
[Task 5/25] Current/Best: 8.29/ 17.62 GFLOPS | Progress: (20/20) | 13.34 s Done.
-
[Task 6/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 6/25] Current/Best: 10.75/ 15.03 GFLOPS | Progress: (4/20) | 5.14 s
[Task 6/25] Current/Best: 8.06/ 23.38 GFLOPS | Progress: (8/20) | 8.16 s
[Task 6/25] Current/Best: 16.93/ 23.38 GFLOPS | Progress: (12/20) | 15.10 s
[Task 6/25] Current/Best: 4.26/ 23.38 GFLOPS | Progress: (16/20) | 19.27 s
[Task 6/25] Current/Best: 12.57/ 23.38 GFLOPS | Progress: (20/20) | 21.91 s Done.
-
[Task 7/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 7/25] Current/Best: 12.79/ 18.20 GFLOPS | Progress: (4/20) | 4.34 s
[Task 7/25] Current/Best: 6.17/ 18.20 GFLOPS | Progress: (8/20) | 6.64 s
[Task 7/25] Current/Best: 17.20/ 19.77 GFLOPS | Progress: (12/20) | 10.12 s
[Task 7/25] Current/Best: 13.93/ 19.77 GFLOPS | Progress: (16/20) | 12.53 s
[Task 7/25] Current/Best: 20.98/ 20.98 GFLOPS | Progress: (20/20) | 15.16 s Done.
-
[Task 8/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 8/25] Current/Best: 13.83/ 16.99 GFLOPS | Progress: (4/20) | 4.04 s
[Task 8/25] Current/Best: 17.69/ 17.69 GFLOPS | Progress: (8/20) | 9.81 s
[Task 8/25] Current/Best: 6.14/ 17.69 GFLOPS | Progress: (12/20) | 15.59 s
[Task 8/25] Current/Best: 11.78/ 17.69 GFLOPS | Progress: (16/20) | 27.18 s
[Task 8/25] Current/Best: 13.71/ 17.69 GFLOPS | Progress: (20/20) | 30.59 s
[Task 9/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 9/25] Current/Best: 3.45/ 16.44 GFLOPS | Progress: (4/20) | 8.49 s
[Task 9/25] Current/Best: 12.90/ 20.55 GFLOPS | Progress: (8/20) | 11.45 s
[Task 9/25] Current/Best: 18.35/ 20.57 GFLOPS | Progress: (12/20) | 18.53 s
[Task 9/25] Current/Best: 16.42/ 20.57 GFLOPS | Progress: (16/20) | 20.23 s
[Task 9/25] Current/Best: 12.96/ 20.57 GFLOPS | Progress: (20/2
0) | 22.56 s Done.
-
[Task 10/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 10/25] Current/Best: 1.51/ 19.33 GFLOPS | Progress: (4/20) | 4.70 s
[Task 10/25] Current/Best: 11.57/ 19.33 GFLOPS | Progress: (8/20) | 7.25 s
[Task 10/25] Current/Best: 5.28/ 19.33 GFLOPS | Progress: (12/20) | 9.06 s
[Task 10/25] Current/Best: 17.01/ 19.33 GFLOPS | Progress: (16/20) | 11.19 s
[Task 10/25] Current/Best: 3.04/ 19.33 GFLOPS | Progress: (20/20) | 13.36 s Done.
-
[Task 11/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 11/25] Current/Best: 14.97/ 19.06 GFLOPS | Progress: (4/20) | 4.86 s
[Task 11/25] Current/Best: 8.11/ 19.95 GFLOPS | Progress: (8/20) | 8.43 s
[Task 11/25] Current/Best: 23.20/ 23.20 GFLOPS | Progress: (12/20) | 11.51 s
[Task 11/25] Current/Best: 6.28/ 23.20 GFLOPS | Progress: (16/20) | 14.29 s
[Task 11/25] Current/Best: 14.89/ 23.20 GFLOPS | Progress: (20/20) | 16.26 s Done.
-
[Task 12/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 12/25] Current/Best: 5.21/ 15.09 GFLOPS | Progress: (4/20) | 4.87 s
[Task 12/25] Current/Best: 9.58/ 15.25 GFLOPS | Progress: (8/20) | 8.13 s
[Task 12/25] Current/Best: 19.00/ 21.11 GFLOPS | Progress: (12/20) | 9.90 s
[Task 12/25] Current/Best: 12.81/ 21.11 GFLOPS | Progress: (16/20) | 11.93 s
[Task 12/25] Current/Best: 11.74/ 21.11 GFLOPS | Progress: (20/20) | 14.83 s Done.
-
[Task 13/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 13/25] Current/Best: 4.18/ 12.69 GFLOPS | Progress: (4/20) | 5.05 s
[Task 13/25] Current/Best: 14.81/ 14.81 GFLOPS | Progress: (8/20) | 7.88 s
[Task 13/25] Current/Best: 12.95/ 20.39 GFLOPS | Progress: (12/20) | 11.00 s
[Task 13/25] Current/Best: 9.07/ 20.39 GFLOPS | Progress: (16/20) | 14.29 s
[Task 13/25] Current/Best: 7.05/ 22.49 GFLOPS | Progress: (20/20) | 17.42 s Done.
-
[Task 14/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 14/25] Current/Best: 15.66/ 15.66 GFLOPS | Progress: (4/20) | 4.80 s
[Task 14/25] Current/Best: 21.23/ 21.23 GFLOPS | Progress: (8/20) | 9.85 s
[Task 14/25] Current/Best: 5.50/ 21.23 GFLOPS | Progress: (12/20) | 14.78 s
[Task 14/25] Current/Best: 10.92/ 21.23 GFLOPS | Progress: (16/20) | 17.78 s
[Task 14/25] Current/Best: 20.10/ 21.23 GFLOPS | Progress: (20/20) | 20.58 s
[Task 15/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
+
[Task 1/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 1/25] Current/Best: 17.57/ 17.57 GFLOPS | Progress: (4/20) | 7.32 s
[Task 1/25] Current/Best: 9.54/ 17.57 GFLOPS | Progress: (8/20) | 11.39 s
[Task 1/25] Current/Best: 15.10/ 22.57 GFLOPS | Progress: (12/20) | 13.59 s
[Task 1/25] Current/Best: 15.96/ 22.57 GFLOPS | Progress: (16/20) | 17.51 s
[Task 1/25] Current/Best: 12.88/ 22.57 GFLOPS | Progress: (20/20) | 19.89 s Done.
+
[Task 2/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 2/25] Current/Best: 6.42/ 18.65 GFLOPS | Progress: (4/20) | 3.46 s
[Task 2/25] Current/Best: 7.92/ 18.65 GFLOPS | Progress: (8/20) | 4.94 s
[Task 2/25] Current/Best: 6.06/ 23.26 GFLOPS | Progress: (12/20) | 6.42 s
[Task 2/25] Current/Best: 15.24/ 23.26 GFLOPS | Progress: (16/20) | 8.32 s
[Task 2/25] Current/Best: 12.43/ 23.26 GFLOPS | Progress: (20/20) | 9.63 s Done.
+
[Task 3/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 3/25] Current/Best: 8.37/ 23.86 GFLOPS | Progress: (4/20) | 4.12 s
[Task 3/25] Current/Best: 16.12/ 23.86 GFLOPS | Progress: (8/20) | 5.96 s
[Task 3/25] Current/Best: 9.38/ 23.86 GFLOPS | Progress: (12/20) | 9.79 s
[Task 3/25] Current/Best: 20.02/ 23.86 GFLOPS | Progress: (16/20) | 11.89 s
[Task 3/25] Current/Best: 14.46/ 23.86 GFLOPS | Progress: (20/20) | 14.31 s Done.
+
[Task 4/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 4/25] Current/Best: 12.35/ 17.21 GFLOPS | Progress: (4/20) | 4.74 s
[Task 4/25] Current/Best: 17.19/ 17.21 GFLOPS | Progress: (8/20) | 7.81 s
[Task 4/25] Current/Best: 7.76/ 17.87 GFLOPS | Progress: (12/20) | 9.81 s
[Task 4/25] Current/Best: 12.60/ 20.03 GFLOPS | Progress: (16/20) | 12.59 s
[Task 4/25] Current/Best: 12.72/ 20.03 GFLOPS | Progress: (20/20) | 15.06 s Done.
+
[Task 5/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 5/25] Current/Best: 14.37/ 14.37 GFLOPS | Progress: (4/20) | 3.81 s
[Task 5/25] Current/Best: 12.28/ 14.55 GFLOPS | Progress: (8/20) | 6.11 s
[Task 5/25] Current/Best: 19.20/ 19.20 GFLOPS | Progress: (12/20) | 7.90 s
[Task 5/25] Current/Best: 15.81/ 19.20 GFLOPS | Progress: (16/20) | 10.01 s
[Task 5/25] Current/Best: 11.28/ 19.20 GFLOPS | Progress: (20/20) | 12.06 s Done.
+
[Task 6/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 6/25] Current/Best: 11.53/ 17.86 GFLOPS | Progress: (4/20) | 4.59 s
[Task 6/25] Current/Best: 4.43/ 17.86 GFLOPS | Progress: (8/20) | 7.36 s
[Task 6/25] Current/Best: 16.77/ 18.78 GFLOPS | Progress: (12/20) | 9.53 s
[Task 6/25] Current/Best: 17.88/ 18.78 GFLOPS | Progress: (16/20) | 16.21 s
[Task 6/25] Current/Best: 5.40/ 18.78 GFLOPS | Progress: (20/20) | 18.85 s Done.
+
[Task 7/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 7/25] Current/Best: 5.44/ 14.39 GFLOPS | Progress: (4/20) | 4.53 s
[Task 7/25] Current/Best: 9.68/ 20.00 GFLOPS | Progress: (8/20) | 6.65 s
[Task 7/25] Current/Best: 13.78/ 20.00 GFLOPS | Progress: (12/20) | 10.80 s
[Task 7/25] Current/Best: 22.98/ 22.98 GFLOPS | Progress: (16/20) | 13.41 s
[Task 7/25] Current/Best: 13.32/ 22.98 GFLOPS | Progress: (20/20) | 15.65 s Done.
+
[Task 8/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 8/25] Current/Best: 11.91/ 13.21 GFLOPS | Progress: (4/20) | 7.93 s
[Task 8/25] Current/Best: 7.24/ 13.21 GFLOPS | Progress: (8/20) | 14.39 s
[Task 8/25] Current/Best: 18.13/ 18.13 GFLOPS | Progress: (12/20) | 17.26 s
[Task 8/25] Current/Best: 9.82/ 18.37 GFLOPS | Progress: (16/20) | 21.48 s
[Task 8/25] Current/Best: 12.03/ 18.37 GFLOPS | Progress: (20/20) | 24.82 s Done.
+
[Task 9/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 9/25] Current/Best: 12.98/ 12.98 GFLOPS | Progress: (4/20) | 4.13 s
[Task 9/25] Current/Best: 13.71/ 13.71 GFLOPS | Progress: (8/20) | 7.19 s
[Task 9/25] Current/Best: 6.35/ 13.71 GFLOPS | Progress: (12/20) | 10.40 s
[Task 9/25] Current/Best: 10.31/ 14.70 GFLOPS | Progress: (16/20) | 17.53 s
[Task 9/25] Current/Best: 13.74/ 15.94 GFLOPS | Progress: (20/20) | 19.55 s Done.
+
[Task 10/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 10/25] Current/Best: 15.10/ 15.10 GFLOPS | Progress: (4/20) | 3.92 s
[Task 10/25] Current/Best: 5.69/ 16.29 GFLOPS | Progress: (8/20) | 5.92 s
[Task 10/25] Current/Best: 8.64/ 16.29 GFLOPS | Progress: (12/20) | 7.69 s
[Task 10/25] Current/Best: 16.86/ 16.86 GFLOPS | Progress: (16/20) | 10.88 s
[Task 10/25] Current/Best: 9.39/ 16.86 GFLOPS | Progress: (20/20) | 13.78 s Done.
+
[Task 11/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 11/25] Current/Best: 11.73/ 22.47 GFLOPS | Progress: (4/20) | 5.37 s
[Task 11/25] Current/Best: 10.55/ 22.47 GFLOPS | Progress: (8/20) | 7.84 s
[Task 11/25] Current/Best: 18.27/ 22.47 GFLOPS | Progress: (12/20) | 10.69 s
[Task 11/25] Current/Best: 16.20/ 22.47 GFLOPS | Progress: (16/20) | 14.38 s
[Task 11/25] Current/Best: 5.09/ 22.47 GFLOPS | Progress: (20/20) | 16.97 s Done.
+
[Task 12/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 12/25] Current/Best: 14.22/ 14.22 GFLOPS | Progress: (4/20) | 4.92 s
[Task 12/25] Current/Best: 14.42/ 19.01 GFLOPS | Progress: (8/20) | 9.21 s
[Task 12/25] Current/Best: 13.56/ 19.01 GFLOPS | Progress: (12/20) | 12.25 s
[Task 12/25] Current/Best: 9.65/ 20.71 GFLOPS | Progress: (16/20) | 15.44 s
[Task 12/25] Current/Best: 18.13/ 20.71 GFLOPS | Progress: (20/20) | 20.13 s Done.
+
[Task 13/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 13/25] Current/Best: 11.04/ 20.67 GFLOPS | Progress: (4/20) | 5.12 s
[Task 13/25] Current/Best: 13.79/ 20.67 GFLOPS | Progress: (8/20) | 9.32 s
[Task 13/25] Current/Best: 12.71/ 21.96 GFLOPS | Progress: (12/20) | 11.80 s
[Task 13/25] Current/Best: 19.07/ 21.96 GFLOPS | Progress: (16/20) | 15.42 s
[Task 13/25] Current/Best: 18.25/ 21.96 GFLOPS | Progress: (20/20) | 18.39 s Done.
+
[Task 14/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 14/25] Current/Best: 13.76/ 16.72 GFLOPS | Progress: (4/20) | 4.18 s
[Task 14/25] Current/Best: 14.91/ 18.91 GFLOPS | Progress: (8/20) | 8.75 s
[Task 14/25] Current/Best: 13.40/ 18.91 GFLOPS | Progress: (12/20) | 13.95 s
[Task 14/25] Current/Best: 15.61/ 20.12 GFLOPS | Progress: (16/20) | 16.04 s
[Task 14/25] Current/Best: 11.08/ 21.85 GFLOPS | Progress: (20/20) | 18.07 s
[Task 15/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 15/25] Current/Best: 4.94/ 12.82 GFLOPS | Progress: (4/20) | 4.55 s
[Task 15/25] Current/Best: 13.14/ 15.30 GFLOPS | Progress: (8/20) | 7.14 s
[Task 15/25] Current/Best: 17.13/ 17.13 GFLOPS | Progress: (12/20) | 9.60 s
[Task 15/25] Current/Best: 15.26/ 19.13 GFLOPS | Progress: (16/20) | 11.95 s
[Task 15/25] Current/Best: 4.78/ 19.13 GFLOPS | Progress: (20/20)
| 13.76 s Done.
+
[Task 16/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 16/25] Current/Best: 18.20/ 18.20 GFLOPS | Progress: (4/20) | 4.83 s
[Task 16/25] Current/Best: 13.07/ 18.51 GFLOPS | Progress: (8/20) | 6.36 s
[Task 16/25] Current/Best: 15.00/ 18.51 GFLOPS | Progress: (12/20) | 8.12 s
[Task 16/25] Current/Best: 19.16/ 19.16 GFLOPS | Progress: (16/20) | 11.23 s
[Task 16/25] Current/Best: 10.15/ 19.16 GFLOPS | Progress: (20/20) | 14.86 s Done.
+
[Task 17/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 17/25] Current/Best: 20.01/ 22.59 GFLOPS | Progress: (4/20) | 4.19 s
[Task 17/25] Current/Best: 11.15/ 22.59 GFLOPS | Progress: (8/20) | 6.50 s
[Task 17/25] Current/Best: 9.92/ 22.59 GFLOPS | Progress: (12/20) | 9.54 s
[Task 17/25] Current/Best: 18.67/ 23.83 GFLOPS | Progress: (16/20) | 13.51 s
[Task 17/25] Current/Best: 6.19/ 23.83 GFLOPS | Progress: (20/20) | 15.77 s Done.
+
[Task 18/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 18/25] Current/Best: 15.34/ 15.34 GFLOPS | Progress: (4/20) | 3.99 s
[Task 18/25] Current/Best: 15.40/ 17.78 GFLOPS | Progress: (8/20) | 6.45 s
[Task 18/25] Current/Best: 16.15/ 17.78 GFLOPS | Progress: (12/20) | 9.62 s
[Task 18/25] Current/Best: 12.88/ 17.78 GFLOPS | Progress: (16/20) | 12.31 s
[Task 18/25] Current/Best: 6.31/ 17.78 GFLOPS | Progress: (20/20) | 16.06 s Done.
+
[Task 19/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 19/25] Current/Best: 8.12/ 17.11 GFLOPS | Progress: (4/20) | 5.54 s
[Task 19/25] Current/Best: 16.97/ 19.85 GFLOPS | Progress: (8/20) | 9.33 s
[Task 19/25] Current/Best: 10.67/ 19.85 GFLOPS | Progress: (12/20) | 16.09 s
[Task 19/25] Current/Best: 18.60/ 20.94 GFLOPS | Progress: (16/20) | 18.24 s
[Task 19/25] Current/Best: 19.36/ 20.94 GFLOPS | Progress: (20/20) | 22.19 s Done.
+
[Task 20/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 20/25] Current/Best: 5.85/ 6.69 GFLOPS | Progress: (4/20) | 4.63 s Done.
+
[Task 20/25] Current/Best: 10.28/ 12.17 GFLOPS | Progress: (8/20) | 7.89 s
[Task 20/25] Current/Best: 11.05/ 12.17 GFLOPS | Progress: (12/20) | 10.06 s
[Task 20/25] Current/Best: 2.32/ 16.68 GFLOPS | Progress: (16/20) | 14.81 s
[Task 20/25] Current/Best: 15.44/ 16.68 GFLOPS | Progress: (20/20) | 16.10 s
[Task 21/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 21/25] Current/Best: 2.71/ 19.29 GFLOPS | Progress: (4/20) | 5.47 s
[Task 21/25] Current/Best: 10.33/ 19.29 GFLOPS | Progress: (8/20) | 8.07 s
[Task 21/25] Current/Best: 14.76/ 20.32 GFLOPS | Progress: (12/20) | 10.26 s
[Task 21/25] Current/Best: 20.03/ 20.32 GFLOPS | Progress: (16/20) | 13.31 s
[Task 21/25] Current/Best: 12.13/ 20.39 GFLOPS | Progress: (20/20) | 15.19 s
[Task 22/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
Done.
-
[Task 15/25] Current/Best: 8.29/ 15.40 GFLOPS | Progress: (4/20) | 9.16 s
[Task 15/25] Current/Best: 8.92/ 20.08 GFLOPS | Progress: (8/20) | 15.77 s
[Task 15/25] Current/Best: 15.42/ 20.67 GFLOPS | Progress: (12/20) | 17.49 s
[Task 15/25] Current/Best: 4.83/ 20.67 GFLOPS | Progress: (16/20) | 20.90 s
[Task 15/25] Current/Best: 6.61/ 20.67 GFLOPS | Progress: (20/20) | 24.04 s Done.
-
[Task 16/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 16/25] Current/Best: 6.90/ 15.17 GFLOPS | Progress: (4/20) | 4.05 s
[Task 16/25] Current/Best: 6.80/ 15.17 GFLOPS | Progress: (8/20) | 6.61 s
[Task 16/25] Current/Best: 19.47/ 19.47 GFLOPS | Progress: (12/20) | 8.90 s
[Task 16/25] Current/Best: 10.42/ 19.47 GFLOPS | Progress: (16/20) | 11.99 s
[Task 16/25] Current/Best: 13.17/ 19.58 GFLOPS | Progress: (20/20) | 13.71 s Done.
-
[Task 17/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 17/25] Current/Best: 12.46/ 19.83 GFLOPS | Progress: (4/20) | 4.31 s
[Task 17/25] Current/Best: 17.70/ 19.83 GFLOPS | Progress: (8/20) | 7.14 s
[Task 17/25] Current/Best: 7.82/ 19.83 GFLOPS | Progress: (12/20) | 9.57 s
[Task 17/25] Current/Best: 21.76/ 21.76 GFLOPS | Progress: (16/20) | 12.59 s
[Task 17/25] Current/Best: 22.37/ 22.37 GFLOPS | Progress: (20/20) | 14.94 s Done.
-
[Task 18/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 18/25] Current/Best: 19.32/ 19.32 GFLOPS | Progress: (4/20) | 5.67 s
[Task 18/25] Current/Best: 12.29/ 19.32 GFLOPS | Progress: (8/20) | 9.85 s
[Task 18/25] Current/Best: 5.96/ 19.32 GFLOPS | Progress: (12/20) | 13.15 s
[Task 18/25] Current/Best: 4.76/ 19.32 GFLOPS | Progress: (16/20) | 15.64 s
[Task 18/25] Current/Best: 15.72/ 19.32 GFLOPS | Progress: (20/20) | 17.90 s Done.
-
[Task 19/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 19/25] Current/Best: 18.01/ 19.29 GFLOPS | Progress: (4/20) | 5.09 s
[Task 19/25] Current/Best: 16.03/ 21.29 GFLOPS | Progress: (8/20) | 8.61 s
[Task 19/25] Current/Best: 11.72/ 21.29 GFLOPS | Progress: (12/20) | 11.76 s
[Task 19/25] Current/Best: 18.31/ 21.86 GFLOPS | Progress: (16/20) | 15.80 s
[Task 19/25] Current/Best: 13.37/ 21.86 GFLOPS | Progress: (20/20) | 18.32 s Done.
-
[Task 20/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 20/25] Current/Best: 10.40/ 17.24 GFLOPS | Progress: (4/20) | 3.86 s
[Task 20/25] Current/Best: 16.01/ 17.24 GFLOPS | Progress: (8/20) | 6.96 s
[Task 20/25] Current/Best: 12.18/ 17.24 GFLOPS | Progress: (12/20) | 10.74 s
[Task 20/25] Current/Best: 12.72/ 17.24 GFLOPS | Progress: (16/20) | 13.42 s
[Task 20/25] Current/Best: 18.36/ 18.36 GFLOPS | Progress: (20/20) | 16.01 s
[Task 21/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 21/25] Current/Best: 11.56/ 12.69 GFLOPS | Progress: (4/20) | 4.97 s
[Task 21/25] Current/Best: 17.44/ 17.90 GFLOPS | Progress: (8/20) | 7.93 s Done.
-
[Task 21/25] Current/Best: 14.80/ 18.67 GFLOPS | Progress: (12/20) | 10.36 s
[Task 21/25] Current/Best: 8.98/ 20.70 GFLOPS | Progress: (16/20) | 11.88 s
[Task 21/25] Current/Best: 15.91/ 20.70 GFLOPS | Progress: (20/20) | 14.49 s
[Task 22/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 22/25] Current/Best: 8.90/ 10.51 GFLOPS | Progress: (4/20) | 4.69 s
[Task 22/25] Current/Best: 17.73/ 17.73 GFLOPS | Progress: (8/20) | 6.98 s
[Task 22/25] Current/Best: 10.54/ 19.50 GFLOPS | Progress: (12/20) | 8.54 s
[Task 22/25] Current/Best: 13.37/ 19.50 GFLOPS | Progress: (16/20) | 10.38 s
[Task 22/25] Current/Best: 17.35/ 19.50 GFLOPS | Progress: (20/20) | 12.81 s Done.
-
[Task 23/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 23/25] Current/Best: 3.08/ 15.14 GFLOPS | Progress: (4/20) | 6.42 s
[Task 23/25] Current/Best: 20.72/ 20.72 GFLOPS | Progress: (8/20) | 9.04 s
[Task 23/25] Current/Best: 19.48/ 20.77 GFLOPS | Progress: (12/20) | 12.69 s
[Task 23/25] Current/Best: 8.67/ 20.77 GFLOPS | Progress: (16/20) | 15.36 s
[Task 23/25] Current/Best: 1.55/ 20.77 GFLOPS | Progress: (20/20) | 25.03 s Done.
-
[Task 24/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 24/25] Current/Best: 1.66/ 4.79 GFLOPS | Progress: (4/20) | 12.84 s
[Task 24/25] Current/Best: 1.93/ 7.13 GFLOPS | Progress: (8/20) | 24.36 s
[Task 24/25] Current/Best: 8.01/ 8.01 GFLOPS | Progress: (12/20) | 35.32 s
[Task 24/25] Current/Best: 2.45/ 8.01 GFLOPS | Progress: (16/20) | 46.31 s
[Task 24/25] Current/Best: 3.75/ 9.68 GFLOPS | Progress: (20/20) | 51.87 s
[Task 25/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
- Done.
-
[Task 25/25] Current/Best: 1.53/ 2.98 GFLOPS | Progress: (4/20) | 3.34 s
[Task 25/25] Current/Best: 3.40/ 8.41 GFLOPS | Progress: (8/20) | 14.02 s
[Task 25/25] Current/Best: 7.53/ 8.99 GFLOPS | Progress: (12/20) | 19.56 s
[Task 25/25] Current/Best: 9.55/ 9.55 GFLOPS | Progress: (16/20) | 25.63 s
[Task 25/25] Current/Best: 8.84/ 9.55 GFLOPS | Progress: (20/20) | 27.38 s
+
[Task 22/25] Current/Best: 5.32/ 18.54 GFLOPS | Progress: (4/20) | 3.55 s
[Task 22/25] Current/Best: 10.30/ 20.91 GFLOPS | Progress: (8/20) | 6.37 s
[Task 22/25] Current/Best: 9.52/ 20.91 GFLOPS | Progress: (12/20) | 8.42 s
[Task 22/25] Current/Best: 12.93/ 20.91 GFLOPS | Progress: (16/20) | 10.42 s
[Task 22/25] Current/Best: 15.93/ 20.91 GFLOPS | Progress: (20/20) | 11.98 s Done.
+
[Task 23/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 23/25] Current/Best: 21.37/ 21.37 GFLOPS | Progress: (4/20) | 4.30 s
[Task 23/25] Current/Best: 11.05/ 21.37 GFLOPS | Progress: (8/20) | 9.79 s
[Task 23/25] Current/Best: 22.78/ 22.78 GFLOPS | Progress: (12/20) | 12.43 s
[Task 23/25] Current/Best: 11.51/ 22.78 GFLOPS | Progress: (16/20) | 15.36 s
[Task 23/25] Current/Best: 12.17/ 22.78 GFLOPS | Progress: (20/20) | 22.68 s Done.
+
[Task 24/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 24/25] Current/Best: 2.26/ 2.86 GFLOPS | Progress: (4/20) | 12.49 s
[Task 24/25] Current/Best: 8.19/ 8.19 GFLOPS | Progress: (8/20) | 21.99 s
[Task 24/25] Current/Best: 5.70/ 8.19 GFLOPS | Progress: (12/20) | 24.90 s
[Task 24/25] Current/Best: 3.68/ 8.19 GFLOPS | Progress: (16/20) | 31.37 s
[Task 24/25] Current/Best: 7.26/ 8.19 GFLOPS | Progress: (20/20) | 33.04 s
[Task 25/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/20) | 0.00 s
[Task 25/25] Current/Best: 2.69/ 7.89 GFLOPS | Progress: (4/20) | 6.56 s
[Task 25/25] Current/Best: 8.74/ 9.54 GFLOPS | Progress: (8/20) | 8.41 s
[Task 25/25] Current/Best: 1.55/ 9.54 GFLOPS | Progress: (12/20) | 9.74 s
[Task 25/25] Current/Best: 3.98/ 9.54 GFLOPS | Progress: (16/20) | 14.81 s
[Task 25/25] Current/Best: 3.02/ 9.54 GFLOPS | Progress: (20/2
0) | 16.54 s Done.
+
@@ -665,7 +665,7 @@ Verify that the optimized model runs and produces the same results:
.. code-block:: none
- class='n02123045 tabby, tabby cat' with probability=0.621102
+ class='n02123045 tabby, tabby cat' with probability=0.621103
class='n02123159 tiger cat' with probability=0.356379
class='n02124075 Egyptian cat' with probability=0.019712
class='n02129604 tiger, Panthera tigris' with probability=0.001215
@@ -723,8 +723,8 @@ improvement in comparing the optimized model to the unoptimized model.
.. code-block:: none
- optimized: {'mean': 421.18828668999413, 'median': 422.2439039001074, 'std': 3.874548039091391}
- unoptimized: {'mean': 519.8523906099763, 'median': 519.4432064999091, 'std': 1.8586737969938754}
+ optimized: {'mean': 429.38185770999553, 'median': 429.40295524999783, 'std': 2.5111785413661645}
+ unoptimized: {'mean': 518.649112039999, 'median': 518.6342771, 'std': 1.203910214067869}
@@ -747,7 +747,7 @@ profiling/benchmarking.
.. rst-class:: sphx-glr-timing
- **Total running time of the script:** ( 12 minutes 3.168 seconds)
+ **Total running time of the script:** ( 11 minutes 1.246 seconds)
.. _sphx_glr_download_tutorial_autotvm_relay_x86.py:
diff --git a/docs/_sources/tutorial/cross_compilation_and_rpc.rst.txt b/docs/_sources/tutorial/cross_compilation_and_rpc.rst.txt
index 619c25c6fe..935b77ca32 100644
--- a/docs/_sources/tutorial/cross_compilation_and_rpc.rst.txt
+++ b/docs/_sources/tutorial/cross_compilation_and_rpc.rst.txt
@@ -274,7 +274,7 @@ device and returns the measured cost. Network overhead is excluded.
.. code-block:: none
- 1.269e-07 secs/op
+ 1.255e-07 secs/op
diff --git a/docs/_sources/tutorial/intro_topi.rst.txt b/docs/_sources/tutorial/intro_topi.rst.txt
index 022934e01c..4426ba4d3c 100644
--- a/docs/_sources/tutorial/intro_topi.rst.txt
+++ b/docs/_sources/tutorial/intro_topi.rst.txt
@@ -268,7 +268,7 @@ As you can see, scheduled stages of computation have been accumulated and we can
.. code-block:: none
- [stage(a, placeholder(a, 0x1995a900)), stage(b, placeholder(b, 0xb598b80)), stage(T_add, compute(T_add, body=[a[ax0, ax1, ax2] + b[ax1, ax2]], axis=[T.iter_var(ax0, T.Range(0, 100), "DataPar", ""), T.iter_var(ax1, T.Range(0, 10), "DataPar", ""), T.iter_var(ax2, T.Range(0, 10), "DataPar", "")], reduce_axis=[], tag=broadcast, attrs={})), stage(T_multiply, compute(T_multiply, body=[a[ax0, ax1, ax2] * b[ax1, ax2]], axis=[T.iter_var(ax0, T.Range(0, 100), "DataPar", ""), T.iter_var(ax1, T. [...]
+ [stage(a, placeholder(a, 0x205a8cf0)), stage(b, placeholder(b, 0xdb8f120)), stage(T_add, compute(T_add, body=[a[ax0, ax1, ax2] + b[ax1, ax2]], axis=[T.iter_var(ax0, T.Range(0, 100), "DataPar", ""), T.iter_var(ax1, T.Range(0, 10), "DataPar", ""), T.iter_var(ax2, T.Range(0, 10), "DataPar", "")], reduce_axis=[], tag=broadcast, attrs={})), stage(T_multiply, compute(T_multiply, body=[a[ax0, ax1, ax2] * b[ax1, ax2]], axis=[T.iter_var(ax0, T.Range(0, 100), "DataPar", ""), T.iter_var(ax1, T. [...]
diff --git a/docs/_sources/tutorial/sg_execution_times.rst.txt b/docs/_sources/tutorial/sg_execution_times.rst.txt
index b24496b183..e0dcaba876 100644
--- a/docs/_sources/tutorial/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorial/sg_execution_times.rst.txt
@@ -5,32 +5,32 @@
Computation times
=================
-**15:39.003** total execution time for **tutorial** files:
+**14:28.097** total execution time for **tutorial** files:
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_autotvm_relay_x86.py` (``autotvm_relay_x86.py``) | 12:03.168 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_autotvm_relay_x86.py` (``autotvm_relay_x86.py``) | 11:01.246 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_auto_scheduler_matmul_x86.py` (``auto_scheduler_matmul_x86.py``) | 01:36.515 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_auto_scheduler_matmul_x86.py` (``auto_scheduler_matmul_x86.py``) | 01:24.802 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_tensor_expr_get_started.py` (``tensor_expr_get_started.py``) | 01:01.125 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_tensor_expr_get_started.py` (``tensor_expr_get_started.py``) | 00:59.309 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_relay_quick_start.py` (``relay_quick_start.py``) | 00:35.861 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_relay_quick_start.py` (``relay_quick_start.py``) | 00:35.672 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_autotvm_matmul_x86.py` (``autotvm_matmul_x86.py``) | 00:20.701 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_autotvm_matmul_x86.py` (``autotvm_matmul_x86.py``) | 00:24.461 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_intro_topi.py` (``intro_topi.py``) | 00:00.833 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_tensor_ir_blitz_course.py` (``tensor_ir_blitz_course.py``) | 00:01.614 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_tensor_ir_blitz_course.py` (``tensor_ir_blitz_course.py``) | 00:00.616 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_intro_topi.py` (``intro_topi.py``) | 00:00.824 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``) | 00:00.185 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``) | 00:00.168 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_tutorial_uma.py` (``uma.py``) | 00:00.000 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_tutorial_introduction.py` (``introduction.py``) | 00:00.000 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_tvmc_python.py` (``tvmc_python.py``) | 00:00.000 | 0.0 MB |
-+------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_tutorial_tvmc_command_line_driver.py` (``tvmc_command_line_driver.py``) | 00:00.000 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
+| :ref:`sphx_glr_tutorial_tvmc_python.py` (``tvmc_python.py``) | 00:00.000 | 0.0 MB |
++------------------------------------------------------------------------------------------+-----------+--------+
| :ref:`sphx_glr_tutorial_install.py` (``install.py``) | 00:00.000 | 0.0 MB |
+------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/tutorial/tensor_expr_get_started.rst.txt b/docs/_sources/tutorial/tensor_expr_get_started.rst.txt
index bbad5ccbba..9f3a1d2ed7 100644
--- a/docs/_sources/tutorial/tensor_expr_get_started.rst.txt
+++ b/docs/_sources/tutorial/tensor_expr_get_started.rst.txt
@@ -286,7 +286,7 @@ helper function to run a profile of the TVM generated code.
.. code-block:: none
Numpy running time: 0.000007
- naive: 0.000008
+ naive: 0.000007
@@ -498,10 +498,10 @@ We can now compare the different schedules
.. code-block:: none
Operator Timing Performance
- numpy 7.386109991784906e-06 1.0
- naive 7.8251e-06 1.0594345343764653
- parallel 6.998100000000001e-06 0.9474676125570207
- vector 2.46932e-05 3.3431941884787326
+ numpy 7.255450000229757e-06 1.0
+ naive 6.952500000000001e-06 0.9582451811782642
+ parallel 6.9663e-06 0.9601471996608617
+ vector 2.46045e-05 3.3911749097879325
@@ -922,7 +922,7 @@ matrix multiplication.
.. code-block:: none
- Numpy running time: 0.019327
+ Numpy running time: 0.018666
@@ -980,7 +980,7 @@ optimizations.
.. code-block:: none
- none: 3.342093
+ none: 3.265791
@@ -1077,7 +1077,7 @@ schedule.
.. code-block:: none
- blocking: 0.330309
+ blocking: 0.306217
@@ -1158,7 +1158,7 @@ already cache friendly from our previous optimizations.
.. code-block:: none
- vectorization: 0.347634
+ vectorization: 0.341605
@I.ir_module
class Module:
@T.prim_func
@@ -1221,7 +1221,7 @@ more cache friendly.
.. code-block:: none
- loop permutation: 0.140605
+ loop permutation: 0.120006
@I.ir_module
class Module:
@T.prim_func
@@ -1309,7 +1309,7 @@ optimized schedule.
.. code-block:: none
- array packing: 0.109399
+ array packing: 0.109400
@I.ir_module
class Module:
@T.prim_func
@@ -1389,7 +1389,7 @@ to `C` when all the block results are ready.
.. code-block:: none
- block caching: 0.112114
+ block caching: 0.110613
@I.ir_module
class Module:
@T.prim_func
@@ -1460,7 +1460,7 @@ of thread-level parallelization.
.. code-block:: none
- parallelization: 0.148669
+ parallelization: 0.146919
@I.ir_module
class Module:
@T.prim_func
@@ -1527,13 +1527,13 @@ working, we can compare the results.
.. code-block:: none
Operator Timing Performance
- none 3.3420934389 1.0
- blocking 0.3303085537 0.09883283030193078
- vectorization 0.3476336397 0.10401673264240582
- loop permutation 0.14060549579999998 0.042071084597287076
- array packing 0.10939919809999998 0.032733734140002704
- block caching 0.1121140604 0.03354605801712733
- parallelization 0.1486689602 0.044483783268768264
+ none 3.2657908377 1.0
+ blocking 0.3062168686 0.09376499715323457
+ vectorization 0.3416048459 0.10460095666769101
+ loop permutation 0.12000589290000001 0.03674634992378036
+ array packing 0.10939959129999999 0.033498652160175356
+ block caching 0.11061271609999998 0.03387011648850765
+ parallelization 0.146918643 0.04498715634326124
@@ -1573,11 +1573,6 @@ operations with tunable parameters that allows you to automatically optimize
the computation for specific platforms.
-.. rst-class:: sphx-glr-timing
-
- **Total running time of the script:** ( 1 minutes 1.125 seconds)
-
-
.. _sphx_glr_download_tutorial_tensor_expr_get_started.py:
.. only:: html
diff --git a/docs/commit_hash b/docs/commit_hash
index f5c5c29fa4..2ade7a41f9 100644
--- a/docs/commit_hash
+++ b/docs/commit_hash
@@ -1 +1 @@
-efa5a388ce3a4a8d6fac503300380b940b7543bd
+d907de3b358e3b0052e0c39b71e65f4a509e6459
diff --git a/docs/how_to/compile_models/from_darknet.html b/docs/how_to/compile_models/from_darknet.html
index 0085c0f608..e00ebec3b2 100644
--- a/docs/how_to/compile_models/from_darknet.html
+++ b/docs/how_to/compile_models/from_darknet.html
@@ -585,7 +585,7 @@ class:['truck 0.9266'] left:471 top:83 right:689 bottom:169
class:['bicycle 0.9984'] left:111 top:113 right:577 bottom:447
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 16.080 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 16.875 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-compile-models-from-darknet-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/7716f96385bd5abb6e822041e285be54/from_darknet.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">from_darknet.py</span></code></a></p>
diff --git a/docs/how_to/compile_models/from_keras.html b/docs/how_to/compile_models/from_keras.html
index 4d6ce32743..09bd4103ee 100644
--- a/docs/how_to/compile_models/from_keras.html
+++ b/docs/how_to/compile_models/from_keras.html
@@ -506,7 +506,7 @@ Tensorflow is also required since it’s used as the default backend of keras.</
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Relay top-1 id: 285, class name: Egyptian cat
1/1 [==============================] - ETA: 0s
-1/1 [==============================] - 1s 964ms/step
+1/1 [==============================] - 1s 947ms/step
Keras top-1 id: 285, class name: Egyptian cat
</pre></div>
</div>
diff --git a/docs/how_to/compile_models/from_mxnet.html b/docs/how_to/compile_models/from_mxnet.html
index 497707559e..15eb1ff6f0 100644
--- a/docs/how_to/compile_models/from_mxnet.html
+++ b/docs/how_to/compile_models/from_mxnet.html
@@ -439,7 +439,7 @@
<span class="nb">print</span><span class="p">(</span><span class="s2">"x"</span><span class="p">,</span> <a href="https://docs.python.org/3/library/stdtypes.html#tuple" title="builtins.tuple" class="sphx-glr-backref-module-builtins sphx-glr-backref-type-py-class sphx-glr-backref-instance"><span class="n">x</span><span class="o">.</span><span class="n">shape</span></a><span class="p">)</span>
</pre></div>
</div>
-<img src="../../_images/sphx_glr_from_mxnet_001.png" srcset="../../_images/sphx_glr_from_mxnet_001.png" alt="from mxnet" class = "sphx-glr-single-img"/><div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading /workspace/.mxnet/models/resnet18_v1-a0666292.zipadb64d9b-d983-4e24-9005-220557c91021 from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
+<img src="../../_images/sphx_glr_from_mxnet_001.png" srcset="../../_images/sphx_glr_from_mxnet_001.png" alt="from mxnet" class = "sphx-glr-single-img"/><div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading /workspace/.mxnet/models/resnet18_v1-a0666292.zip7772c4e3-b5bb-4af5-84d2-2e2de09354bb from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
x (1, 3, 224, 224)
</pre></div>
</div>
diff --git a/docs/how_to/compile_models/from_oneflow.html b/docs/how_to/compile_models/from_oneflow.html
index 2a46e4b9f2..0cab90ff21 100644
--- a/docs/how_to/compile_models/from_oneflow.html
+++ b/docs/how_to/compile_models/from_oneflow.html
@@ -449,12 +449,12 @@ Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdo
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading: "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/ResNet/resnet18.zip" to /workspace/.oneflow/flowvision_cache/resnet18.zip
0%| | 0.00/41.5M [00:00<?, ?B/s]
- 19%|#9 | 7.99M/41.5M [00:00<00:00, 60.9MB/s]
- 39%|###8 | 16.0M/41.5M [00:00<00:00, 63.5MB/s]
- 58%|#####7 | 24.0M/41.5M [00:00<00:00, 56.6MB/s]
- 77%|#######7 | 32.0M/41.5M [00:00<00:00, 62.0MB/s]
- 96%|#########6| 40.0M/41.5M [00:00<00:00, 61.8MB/s]
-100%|##########| 41.5M/41.5M [00:00<00:00, 62.9MB/s]
+ 19%|#9 | 7.99M/41.5M [00:00<00:00, 82.3MB/s]
+ 39%|###8 | 16.0M/41.5M [00:00<00:00, 57.7MB/s]
+ 58%|#####7 | 24.0M/41.5M [00:00<00:00, 55.3MB/s]
+ 77%|#######7 | 32.0M/41.5M [00:00<00:00, 62.1MB/s]
+ 96%|#########6| 40.0M/41.5M [00:00<00:00, 61.6MB/s]
+100%|##########| 41.5M/41.5M [00:00<00:00, 62.6MB/s]
</pre></div>
</div>
</div>
diff --git a/docs/how_to/compile_models/from_pytorch.html b/docs/how_to/compile_models/from_pytorch.html
index dcea784a42..7bd154fea4 100644
--- a/docs/how_to/compile_models/from_pytorch.html
+++ b/docs/how_to/compile_models/from_pytorch.html
@@ -432,10 +432,10 @@ be unstable.</p>
Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to /workspace/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth
0%| | 0.00/44.7M [00:00<?, ?B/s]
- 27%|##7 | 12.1M/44.7M [00:00<00:00, 127MB/s]
- 54%|#####4 | 24.2M/44.7M [00:00<00:00, 108MB/s]
- 78%|#######7 | 34.7M/44.7M [00:00<00:00, 105MB/s]
-100%|##########| 44.7M/44.7M [00:00<00:00, 106MB/s]
+ 27%|##7 | 12.3M/44.7M [00:00<00:00, 129MB/s]
+ 55%|#####4 | 24.5M/44.7M [00:00<00:00, 110MB/s]
+ 90%|########9 | 40.0M/44.7M [00:00<00:00, 119MB/s]
+100%|##########| 44.7M/44.7M [00:00<00:00, 120MB/s]
</pre></div>
</div>
</div>
diff --git a/docs/how_to/compile_models/from_tensorflow.html b/docs/how_to/compile_models/from_tensorflow.html
index cc6b7d69ef..78107ae574 100644
--- a/docs/how_to/compile_models/from_tensorflow.html
+++ b/docs/how_to/compile_models/from_tensorflow.html
@@ -649,7 +649,7 @@ banana (score = 0.00022)
desk (score = 0.00019)
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 19.756 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 20.526 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-compile-models-from-tensorflow-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/7f1d3d1b878694c201c614c807cdebc8/from_tensorflow.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">from_tensorflow.py</span></code></a></p>
diff --git a/docs/how_to/compile_models/sg_execution_times.html b/docs/how_to/compile_models/sg_execution_times.html
index 7e45e7e7e1..f54a977128 100644
--- a/docs/how_to/compile_models/sg_execution_times.html
+++ b/docs/how_to/compile_models/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-compile-models-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>06:17.311</strong> total execution time for <strong>how_to_compile_models</strong> files:</p>
+<p><strong>06:17.675</strong> total execution time for <strong>how_to_compile_models</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 81%" />
@@ -349,43 +349,43 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="from_tensorflow.html#sphx-glr-how-to-compile-models-from-tensorflow-py"><span class="std std-ref">Compile Tensorflow Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_tensorflow.py</span></code>)</p></td>
-<td><p>01:19.756</p></td>
+<td><p>01:20.526</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="from_darknet.html#sphx-glr-how-to-compile-models-from-darknet-py"><span class="std std-ref">Compile YOLO-V2 and YOLO-V3 in DarkNet Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_darknet.py</span></code>)</p></td>
-<td><p>01:16.080</p></td>
+<td><p>01:16.875</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="from_paddle.html#sphx-glr-how-to-compile-models-from-paddle-py"><span class="std std-ref">Compile PaddlePaddle Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_paddle.py</span></code>)</p></td>
-<td><p>00:51.867</p></td>
+<td><p>00:52.063</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="from_oneflow.html#sphx-glr-how-to-compile-models-from-oneflow-py"><span class="std std-ref">Compile OneFlow Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_oneflow.py</span></code>)</p></td>
-<td><p>00:34.823</p></td>
+<td><p>00:35.509</p></td>
<td><p>0.0 MB</p></td>
</tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="from_coreml.html#sphx-glr-how-to-compile-models-from-coreml-py"><span class="std std-ref">Compile CoreML Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_coreml.py</span></code>)</p></td>
-<td><p>00:30.953</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="from_mxnet.html#sphx-glr-how-to-compile-models-from-mxnet-py"><span class="std std-ref">Compile MXNet Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_mxnet.py</span></code>)</p></td>
+<td><p>00:30.426</p></td>
<td><p>0.0 MB</p></td>
</tr>
-<tr class="row-even"><td><p><a class="reference internal" href="from_mxnet.html#sphx-glr-how-to-compile-models-from-mxnet-py"><span class="std std-ref">Compile MXNet Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_mxnet.py</span></code>)</p></td>
-<td><p>00:30.871</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="from_coreml.html#sphx-glr-how-to-compile-models-from-coreml-py"><span class="std std-ref">Compile CoreML Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_coreml.py</span></code>)</p></td>
+<td><p>00:30.080</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="from_tflite.html#sphx-glr-how-to-compile-models-from-tflite-py"><span class="std std-ref">Compile TFLite Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_tflite.py</span></code>)</p></td>
-<td><p>00:26.285</p></td>
+<td><p>00:26.201</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="from_pytorch.html#sphx-glr-how-to-compile-models-from-pytorch-py"><span class="std std-ref">Compile PyTorch Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_pytorch.py</span></code>)</p></td>
-<td><p>00:24.216</p></td>
+<td><p>00:23.765</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="from_keras.html#sphx-glr-how-to-compile-models-from-keras-py"><span class="std std-ref">Compile Keras Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_keras.py</span></code>)</p></td>
-<td><p>00:19.821</p></td>
+<td><p>00:19.631</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="from_onnx.html#sphx-glr-how-to-compile-models-from-onnx-py"><span class="std std-ref">Compile ONNX Models</span></a> (<code class="docutils literal notranslate"><span class="pre">from_onnx.py</span></code>)</p></td>
-<td><p>00:02.639</p></td>
+<td><p>00:02.598</p></td>
<td><p>0.0 MB</p></td>
</tr>
</tbody>
diff --git a/docs/how_to/deploy_models/deploy_model_on_adreno.html b/docs/how_to/deploy_models/deploy_model_on_adreno.html
index 4965b8741b..0d9e64a440 100644
--- a/docs/how_to/deploy_models/deploy_model_on_adreno.html
+++ b/docs/how_to/deploy_models/deploy_model_on_adreno.html
@@ -920,10 +920,10 @@ Top5 predictions:
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 3485.6176 3484.7953 3490.6544 3482.3343 2.6236
+ 3340.4418 3339.6930 3345.6862 3337.8607 2.4028
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 5.197 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 3.250 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-model-on-adreno-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/2387d8448da213eb625e6b3d916327d4/deploy_model_on_adreno.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_model_on_adreno.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/deploy_model_on_android.html b/docs/how_to/deploy_models/deploy_model_on_android.html
index 02b21a2dae..9752f12a4a 100644
--- a/docs/how_to/deploy_models/deploy_model_on_android.html
+++ b/docs/how_to/deploy_models/deploy_model_on_android.html
@@ -662,7 +662,7 @@ to the remote android device.</p>
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 16.3794 16.3462 16.5128 16.2863 0.0881
+ 16.3099 16.1822 17.2296 15.9700 0.4217
</pre></div>
</div>
</div>
diff --git a/docs/how_to/deploy_models/deploy_object_detection_pytorch.html b/docs/how_to/deploy_models/deploy_object_detection_pytorch.html
index 7ad7c40d4e..349cfe173a 100644
--- a/docs/how_to/deploy_models/deploy_object_detection_pytorch.html
+++ b/docs/how_to/deploy_models/deploy_object_detection_pytorch.html
@@ -454,23 +454,28 @@ be unstable.</p>
Downloading: "https://download.pytorch.org/models/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth" to /workspace/.cache/torch/hub/checkpoints/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth
0%| | 0.00/170M [00:00<?, ?B/s]
- 7%|7 | 12.4M/170M [00:00<00:01, 130MB/s]
- 15%|#4 | 24.9M/170M [00:00<00:01, 108MB/s]
- 21%|## | 35.4M/170M [00:00<00:01, 107MB/s]
- 27%|##6 | 45.6M/170M [00:00<00:01, 104MB/s]
- 33%|###2 | 55.7M/170M [00:00<00:01, 103MB/s]
- 39%|###8 | 65.5M/170M [00:00<00:01, 102MB/s]
- 44%|####4 | 75.2M/170M [00:00<00:00, 101MB/s]
- 50%|####9 | 84.9M/170M [00:00<00:00, 101MB/s]
- 56%|#####5 | 94.5M/170M [00:00<00:00, 101MB/s]
- 61%|######1 | 104M/170M [00:01<00:00, 99.1MB/s]
- 67%|######7 | 114M/170M [00:01<00:00, 101MB/s]
- 73%|#######2 | 124M/170M [00:01<00:00, 101MB/s]
- 79%|#######8 | 133M/170M [00:01<00:00, 100MB/s]
- 84%|########4 | 143M/170M [00:01<00:00, 100MB/s]
- 90%|########9 | 152M/170M [00:01<00:00, 99.3MB/s]
- 96%|#########5| 162M/170M [00:01<00:00, 100MB/s]
-100%|##########| 170M/170M [00:01<00:00, 102MB/s]
+ 5%|4 | 7.99M/170M [00:00<00:03, 44.4MB/s]
+ 7%|7 | 12.2M/170M [00:00<00:04, 38.3MB/s]
+ 9%|9 | 16.0M/170M [00:00<00:04, 37.6MB/s]
+ 13%|#2 | 21.4M/170M [00:00<00:03, 43.7MB/s]
+ 17%|#6 | 28.8M/170M [00:00<00:02, 54.3MB/s]
+ 20%|## | 34.2M/170M [00:00<00:03, 38.8MB/s]
+ 24%|##4 | 41.3M/170M [00:00<00:02, 47.3MB/s]
+ 30%|### | 51.1M/170M [00:01<00:02, 61.5MB/s]
+ 36%|###6 | 61.3M/170M [00:01<00:01, 73.4MB/s]
+ 41%|#### | 69.0M/170M [00:01<00:02, 52.5MB/s]
+ 44%|####4 | 75.3M/170M [00:01<00:01, 53.1MB/s]
+ 48%|####7 | 81.3M/170M [00:01<00:01, 51.1MB/s]
+ 52%|#####2 | 89.1M/170M [00:01<00:01, 58.1MB/s]
+ 60%|###### | 103M/170M [00:01<00:00, 78.8MB/s]
+ 65%|######5 | 111M/170M [00:02<00:00, 78.1MB/s]
+ 71%|####### | 120M/170M [00:02<00:00, 73.3MB/s]
+ 76%|#######6 | 129M/170M [00:02<00:00, 79.5MB/s]
+ 81%|######## | 137M/170M [00:02<00:00, 67.7MB/s]
+ 85%|########5 | 144M/170M [00:02<00:00, 62.3MB/s]
+ 90%|########9 | 153M/170M [00:02<00:00, 68.4MB/s]
+ 94%|#########4| 160M/170M [00:02<00:00, 67.5MB/s]
+100%|##########| 170M/170M [00:02<00:00, 62.1MB/s]
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torch/nn/functional.py:3897: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
for i in range(dim)
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/torchvision/models/detection/anchor_utils.py:124: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode=& [...]
@@ -568,7 +573,7 @@ torchvision rcnn models.</p>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Get 9 valid boxes
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 3 minutes 34.987 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 3 minutes 30.093 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-object-detection-pytorch-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/7795da4b258c8feff986668b95ef57ad/deploy_object_detection_pytorch.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_object_detection_pytorch.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/deploy_prequantized.html b/docs/how_to/deploy_models/deploy_prequantized.html
index ae7daecd4d..598a1e1d07 100644
--- a/docs/how_to/deploy_models/deploy_prequantized.html
+++ b/docs/how_to/deploy_models/deploy_prequantized.html
@@ -495,8 +495,8 @@ training. Other models require a full post training calibration.</p>
Downloading: "https://download.pytorch.org/models/mobilenet_v2-b0353104.pth" to /workspace/.cache/torch/hub/checkpoints/mobilenet_v2-b0353104.pth
0%| | 0.00/13.6M [00:00<?, ?B/s]
- 59%|#####8 | 7.99M/13.6M [00:00<00:00, 80.0MB/s]
-100%|##########| 13.6M/13.6M [00:00<00:00, 101MB/s]
+ 59%|#####8 | 7.99M/13.6M [00:00<00:00, 58.8MB/s]
+100%|##########| 13.6M/13.6M [00:00<00:00, 48.1MB/s]
</pre></div>
</div>
</div>
@@ -587,7 +587,7 @@ output values are identical out of 1000 outputs from mobilenet v2.</p>
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 90.4587 90.3497 92.1262 90.1692 0.2914
+ 90.4858 90.3190 95.8622 90.0693 0.7481
</pre></div>
</div>
<div class="admonition note">
@@ -626,7 +626,7 @@ This includes support for the VNNI 8 bit dot product instruction (CascadeLake or
<div class="section" id="deploy-a-quantized-tflite-model">
<h2>Deploy a quantized TFLite Model<a class="headerlink" href="#deploy-a-quantized-tflite-model" title="Permalink to this headline">¶</a></h2>
<p>TODO</p>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 14.706 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 13.807 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-prequantized-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/fb8217c13f4351224c6cf3aacf1a87fc/deploy_prequantized.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_prequantized.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/deploy_prequantized_tflite.html b/docs/how_to/deploy_models/deploy_prequantized_tflite.html
index b335241bff..1a743a68d9 100644
--- a/docs/how_to/deploy_models/deploy_prequantized_tflite.html
+++ b/docs/how_to/deploy_models/deploy_prequantized_tflite.html
@@ -580,7 +580,7 @@ TFLite Top-5 labels: [387 102 386 341 349]
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 123.0591 122.7188 147.8364 121.6682 2.5565
+ 120.8559 120.8256 123.1634 119.9825 0.4285
</pre></div>
</div>
<div class="admonition note">
@@ -608,7 +608,7 @@ network for ARM CPU</span></a>.</p></li>
</ul>
</div></blockquote>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 2 minutes 29.779 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 2 minutes 28.294 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-prequantized-tflite-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/56691c7a27d45da61d112276334640d3/deploy_prequantized_tflite.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_prequantized_tflite.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/deploy_quantized.html b/docs/how_to/deploy_models/deploy_quantized.html
index c3b149912c..7cfad5c888 100644
--- a/docs/how_to/deploy_models/deploy_quantized.html
+++ b/docs/how_to/deploy_models/deploy_quantized.html
@@ -521,7 +521,7 @@ for calibration. But the accuracy might be impacted.</p>
DeprecationWarning,
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 26.972 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 30.320 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-quantized-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/7810ecf51bfc05f7d5e8a400ac3e815d/deploy_quantized.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_quantized.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/deploy_ssd_gluoncv.html b/docs/how_to/deploy_models/deploy_ssd_gluoncv.html
index f624b48b79..4a65e0ff23 100644
--- a/docs/how_to/deploy_models/deploy_ssd_gluoncv.html
+++ b/docs/how_to/deploy_models/deploy_ssd_gluoncv.html
@@ -463,22 +463,23 @@ to your device.</p>
Downloading /workspace/.mxnet/models/ssd_512_resnet50_v1_voc-9c8b225a.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/ssd_512_resnet50_v1_voc-9c8b225a.zip...
0%| | 0/132723 [00:00<?, ?KB/s]
- 5%|4 | 6095/132723 [00:00<00:02, 60939.45KB/s]
- 11%|#1 | 14707/132723 [00:00<00:01, 75748.30KB/s]
- 17%|#7 | 23086/132723 [00:00<00:01, 79417.20KB/s]
- 24%|##3 | 31568/132723 [00:00<00:01, 81546.47KB/s]
- 30%|### | 40094/132723 [00:00<00:01, 82878.91KB/s]
- 37%|###6 | 48651/132723 [00:00<00:01, 83791.92KB/s]
- 43%|####3 | 57139/132723 [00:00<00:00, 84143.51KB/s]
- 49%|####9 | 65688/132723 [00:00<00:00, 84568.56KB/s]
- 56%|#####5 | 74240/132723 [00:00<00:00, 84861.38KB/s]
- 62%|######2 | 82876/132723 [00:01<00:00, 85320.62KB/s]
- 69%|######8 | 91463/132723 [00:01<00:00, 85486.02KB/s]
- 75%|#######5 | 100147/132723 [00:01<00:00, 85895.48KB/s]
- 82%|########1 | 108781/132723 [00:01<00:00, 86027.48KB/s]
- 88%|########8 | 117454/132723 [00:01<00:00, 86235.77KB/s]
- 95%|#########5| 126092/132723 [00:01<00:00, 86277.16KB/s]
-100%|##########| 132723/132723 [00:01<00:00, 84062.38KB/s]
+ 5%|4 | 6269/132723 [00:00<00:02, 62663.47KB/s]
+ 11%|# | 14327/132723 [00:00<00:01, 73190.96KB/s]
+ 17%|#6 | 22357/132723 [00:00<00:01, 76433.51KB/s]
+ 23%|##2 | 30450/132723 [00:00<00:01, 78199.94KB/s]
+ 29%|##9 | 38520/132723 [00:00<00:01, 79098.45KB/s]
+ 35%|###5 | 46502/132723 [00:00<00:01, 79341.91KB/s]
+ 41%|####1 | 54473/132723 [00:00<00:00, 79457.92KB/s]
+ 47%|####7 | 62501/132723 [00:00<00:00, 79715.33KB/s]
+ 53%|#####3 | 70544/132723 [00:00<00:00, 79937.08KB/s]
+ 59%|#####9 | 78549/132723 [00:01<00:00, 79968.55KB/s]
+ 65%|######5 | 86546/132723 [00:01<00:00, 79827.14KB/s]
+ 71%|#######1 | 94608/132723 [00:01<00:00, 80066.65KB/s]
+ 77%|#######7 | 102669/132723 [00:01<00:00, 80229.44KB/s]
+ 83%|########3 | 110726/132723 [00:01<00:00, 80325.62KB/s]
+ 90%|########9 | 118800/132723 [00:01<00:00, 80447.13KB/s]
+ 96%|#########5| 126845/132723 [00:01<00:00, 80300.26KB/s]
+100%|##########| 132723/132723 [00:01<00:00, 79243.42KB/s]
</pre></div>
</div>
<p>Create TVM runtime and do inference
@@ -517,7 +518,7 @@ Downloading /workspace/.mxnet/models/ssd_512_resnet50_v1_voc-9c8b225a.zip from h
<span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
</pre></div>
</div>
-<img src="../../_images/sphx_glr_deploy_ssd_gluoncv_001.png" srcset="../../_images/sphx_glr_deploy_ssd_gluoncv_001.png" alt="deploy ssd gluoncv" class = "sphx-glr-single-img"/><p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 3 minutes 36.504 seconds)</p>
+<img src="../../_images/sphx_glr_deploy_ssd_gluoncv_001.png" srcset="../../_images/sphx_glr_deploy_ssd_gluoncv_001.png" alt="deploy ssd gluoncv" class = "sphx-glr-single-img"/><p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 3 minutes 31.204 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-deploy-models-deploy-ssd-gluoncv-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/cccb17d28e5e8b2e94ea8cd5ec59f6ed/deploy_ssd_gluoncv.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">deploy_ssd_gluoncv.py</span></code></a></p>
diff --git a/docs/how_to/deploy_models/sg_execution_times.html b/docs/how_to/deploy_models/sg_execution_times.html
index 880c3591e4..895ca6c7c1 100644
--- a/docs/how_to/deploy_models/sg_execution_times.html
+++ b/docs/how_to/deploy_models/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-deploy-models-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>15:03.815</strong> total execution time for <strong>how_to_deploy_models</strong> files:</p>
+<p><strong>14:50.812</strong> total execution time for <strong>how_to_deploy_models</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 86%" />
@@ -349,39 +349,39 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="deploy_ssd_gluoncv.html#sphx-glr-how-to-deploy-models-deploy-ssd-gluoncv-py"><span class="std std-ref">Deploy Single Shot Multibox Detector(SSD) model</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_ssd_gluoncv.py</span></code>)</p></td>
-<td><p>03:36.504</p></td>
+<td><p>03:31.204</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="deploy_object_detection_pytorch.html#sphx-glr-how-to-deploy-models-deploy-object-detection-pytorch-py"><span class="std std-ref">Compile PyTorch Object Detection Models</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_object_detection_pytorch.py</span></code>)</p></td>
-<td><p>03:34.987</p></td>
+<td><p>03:30.093</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="deploy_prequantized_tflite.html#sphx-glr-how-to-deploy-models-deploy-prequantized-tflite-py"><span class="std std-ref">Deploy a Framework-prequantized Model with TVM - Part 3 (TFLite)</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_prequantized_tflite.py</span></code>)</p></td>
-<td><p>02:29.779</p></td>
+<td><p>02:28.294</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="deploy_quantized.html#sphx-glr-how-to-deploy-models-deploy-quantized-py"><span class="std std-ref">Deploy a Quantized Model on Cuda</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_quantized.py</span></code>)</p></td>
-<td><p>01:26.972</p></td>
+<td><p>01:30.320</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="deploy_prequantized.html#sphx-glr-how-to-deploy-models-deploy-prequantized-py"><span class="std std-ref">Deploy a Framework-prequantized Model with TVM</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_prequantized.py</span></code>)</p></td>
-<td><p>01:14.706</p></td>
+<td><p>01:13.807</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="deploy_model_on_adreno.html#sphx-glr-how-to-deploy-models-deploy-model-on-adreno-py"><span class="std std-ref">Deploy the Pretrained Model on Adreno</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_model_on_adreno.py</span></code>)</p></td>
-<td><p>01:05.197</p></td>
+<td><p>01:03.250</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="deploy_model_on_android.html#sphx-glr-how-to-deploy-models-deploy-model-on-android-py"><span class="std std-ref">Deploy the Pretrained Model on Android</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_model_on_android.py</span></code>)</p></td>
-<td><p>00:41.204</p></td>
+<td><p>00:40.356</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="deploy_model_on_nano.html#sphx-glr-how-to-deploy-models-deploy-model-on-nano-py"><span class="std std-ref">Deploy the Pretrained Model on Jetson Nano</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_model_on_nano.py</span></code>)</p></td>
-<td><p>00:27.427</p></td>
+<td><p>00:26.963</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="deploy_model_on_rasp.html#sphx-glr-how-to-deploy-models-deploy-model-on-rasp-py"><span class="std std-ref">Deploy the Pretrained Model on Raspberry Pi</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_model_on_rasp.py</span></code>)</p></td>
-<td><p>00:27.034</p></td>
+<td><p>00:26.521</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="deploy_sparse.html#sphx-glr-how-to-deploy-models-deploy-sparse-py"><span class="std std-ref">Deploy a Hugging Face Pruned Model on CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">deploy_sparse.py</span></code>)</p></td>
diff --git a/docs/how_to/extend_tvm/bring_your_own_datatypes.html b/docs/how_to/extend_tvm/bring_your_own_datatypes.html
index 3fca8787ee..4eb1a29bd8 100644
--- a/docs/how_to/extend_tvm/bring_your_own_datatypes.html
+++ b/docs/how_to/extend_tvm/bring_your_own_datatypes.html
@@ -619,7 +619,7 @@ In this alpha state of the Bring Your Own Datatypes framework, we have not imple
<span class="n">module</span><span class="p">,</span> <a href="https://docs.python.org/3/library/stdtypes.html#dict" title="builtins.dict" class="sphx-glr-backref-module-builtins sphx-glr-backref-type-py-class sphx-glr-backref-instance"><span class="n">params</span></a> <span class="o">=</span> <span class="n">get_mobilenet</span><span class="p">()</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading /workspace/.mxnet/models/mobilenet0.25-9f83e440.zip6226acea-b49c-4426-8032-9e81465ff59a from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading /workspace/.mxnet/models/mobilenet0.25-9f83e440.zip8aed1037-c0db-424d-a486-9c23a0465e2a from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
</pre></div>
</div>
<p>It’s easy to execute MobileNet with native TVM:</p>
diff --git a/docs/how_to/extend_tvm/sg_execution_times.html b/docs/how_to/extend_tvm/sg_execution_times.html
index 367e103fe2..8f71aebd37 100644
--- a/docs/how_to/extend_tvm/sg_execution_times.html
+++ b/docs/how_to/extend_tvm/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-extend-tvm-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>00:55.017</strong> total execution time for <strong>how_to_extend_tvm</strong> files:</p>
+<p><strong>00:53.382</strong> total execution time for <strong>how_to_extend_tvm</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 84%" />
@@ -349,15 +349,15 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="bring_your_own_datatypes.html#sphx-glr-how-to-extend-tvm-bring-your-own-datatypes-py"><span class="std std-ref">Bring Your Own Datatypes to TVM</span></a> (<code class="docutils literal notranslate"><span class="pre">bring_your_own_datatypes.py</span></code>)</p></td>
-<td><p>00:51.133</p></td>
+<td><p>00:49.580</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="use_pass_instrument.html#sphx-glr-how-to-extend-tvm-use-pass-instrument-py"><span class="std std-ref">How to Use TVM Pass Instrument</span></a> (<code class="docutils literal notranslate"><span class="pre">use_pass_instrument.py</span></code>)</p></td>
-<td><p>00:02.764</p></td>
+<td><p>00:02.708</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="use_pass_infra.html#sphx-glr-how-to-extend-tvm-use-pass-infra-py"><span class="std std-ref">How to Use TVM Pass Infra</span></a> (<code class="docutils literal notranslate"><span class="pre">use_pass_infra.py</span></code>)</p></td>
-<td><p>00:01.113</p></td>
+<td><p>00:01.087</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="low_level_custom_pass.html#sphx-glr-how-to-extend-tvm-low-level-custom-pass-py"><span class="std std-ref">Writing a Customized Pass</span></a> (<code class="docutils literal notranslate"><span class="pre">low_level_custom_pass.py</span></code>)</p></td>
diff --git a/docs/how_to/extend_tvm/use_pass_instrument.html b/docs/how_to/extend_tvm/use_pass_instrument.html
index c3960e0da9..c51ad08a2c 100644
--- a/docs/how_to/extend_tvm/use_pass_instrument.html
+++ b/docs/how_to/extend_tvm/use_pass_instrument.html
@@ -526,10 +526,10 @@ profile the execution time of each passes.</p>
</pre></div>
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Printing results of timing profile...
-InferType: 18990us [18990us] (48.38%; 48.38%)
-FoldScaleAxis: 20260us [10us] (51.62%; 51.62%)
- FoldConstant: 20250us [1863us] (51.59%; 99.95%)
- InferType: 18387us [18387us] (46.85%; 90.80%)
+InferType: 18404us [18404us] (48.41%; 48.41%)
+FoldScaleAxis: 19615us [8us] (51.59%; 51.59%)
+ FoldConstant: 19607us [1783us] (51.57%; 99.96%)
+ InferType: 17823us [17823us] (46.88%; 90.91%)
</pre></div>
</div>
</div>
@@ -551,10 +551,10 @@ Refer to following sections and <a class="reference internal" href="../../refere
</pre></div>
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Printing results of timing profile...
-InferType: 18830us [18830us] (47.13%; 47.13%)
-FoldScaleAxis: 21126us [9us] (52.87%; 52.87%)
- FoldConstant: 21117us [1949us] (52.85%; 99.96%)
- InferType: 19168us [19168us] (47.97%; 90.77%)
+InferType: 18040us [18040us] (47.49%; 47.49%)
+FoldScaleAxis: 19946us [8us] (52.51%; 52.51%)
+ FoldConstant: 19938us [1773us] (52.49%; 99.96%)
+ InferType: 18166us [18166us] (47.82%; 91.11%)
</pre></div>
</div>
<p>Register empty list to clear existing instruments.</p>
diff --git a/docs/how_to/optimize_operators/opt_conv_cuda.html b/docs/how_to/optimize_operators/opt_conv_cuda.html
index 439c520623..fb763321c1 100644
--- a/docs/how_to/optimize_operators/opt_conv_cuda.html
+++ b/docs/how_to/optimize_operators/opt_conv_cuda.html
@@ -575,7 +575,7 @@ latency of convolution.</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Convolution: </span><span class="si">%f</span><span class="s2"> ms"</span> <span class="o">%</span> <span class="p">(</span><span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">w</span><span class="p">,</span> <span class="n">b</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span> <span class="o">*</span> <span cl [...]
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Convolution: 54.172832 ms
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Convolution: 54.222881 ms
</pre></div>
</div>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-optimize-operators-opt-conv-cuda-py">
diff --git a/docs/how_to/optimize_operators/opt_conv_tensorcore.html b/docs/how_to/optimize_operators/opt_conv_tensorcore.html
index 0c7d3ceadb..3fb299a9ae 100644
--- a/docs/how_to/optimize_operators/opt_conv_tensorcore.html
+++ b/docs/how_to/optimize_operators/opt_conv_tensorcore.html
@@ -861,7 +861,7 @@ be able to run on our build server</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"conv2d with tensor core: </span><span class="si">%f</span><span class="s2"> ms"</span> <span class="o">%</span> <span class="p">(</span><span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">w</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span> <span class="o">* [...]
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>conv2d with tensor core: 13.370595 ms
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>conv2d with tensor core: 6.710870 ms
</pre></div>
</div>
</div>
diff --git a/docs/how_to/optimize_operators/opt_gemm.html b/docs/how_to/optimize_operators/opt_gemm.html
index f73840c155..22098bdc7e 100644
--- a/docs/how_to/optimize_operators/opt_gemm.html
+++ b/docs/how_to/optimize_operators/opt_gemm.html
@@ -472,8 +472,8 @@ Then we write a baseline implementation, the simplest way to write a matrix mult
<span class="nb">print</span><span class="p">(</span><span class="s2">"Baseline: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Numpy running time: 0.019813
-Baseline: 3.269114
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Numpy running time: 0.019114
+Baseline: 3.291007
</pre></div>
</div>
<p>In TVM, we can always inspect lower level IR to debug or optimize our schedule.
@@ -529,7 +529,7 @@ fill 32 * 32 * sizeof(float) which is 4KB in the cache whose total size is 32KB
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt1: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt1: 0.326293
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt1: 0.303538
</pre></div>
</div>
<p>Here is the generated IR after blocking.</p>
@@ -583,7 +583,7 @@ vastly.</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt2: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt2: 0.351003
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt2: 0.344968
</pre></div>
</div>
<p>Here is the generated IR after vectorization.</p>
@@ -635,7 +635,7 @@ the access pattern for A matrix is more cache friendly.</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt3: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt3: 0.126392
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt3: 0.118944
</pre></div>
</div>
<p>Here is the generated IR after loop permutation.</p>
@@ -709,7 +709,7 @@ flattening.</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt4: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt4: 0.110417
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt4: 0.109755
</pre></div>
</div>
<p>Here is the generated IR after array packing.</p>
@@ -784,7 +784,7 @@ write to C when all the block results are ready.</p>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt5: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">evaluator</span><span class="p">(</span><span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">)</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt5: 0.112664
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt5: 0.111921
</pre></div>
</div>
<p>Here is the generated IR after blocking.</p>
@@ -861,7 +861,7 @@ class Module:
<span class="nb">print</span><span class="p">(</span><span class="s2">"Opt6: </span><span class="si">%f</span><span class="s2">"</span> <span class="o">%</span> <span class="n">opt6_time</span><span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt6: 0.148654
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Opt6: 0.146599
</pre></div>
</div>
<p>Here is the generated IR after parallelization.</p>
diff --git a/docs/how_to/optimize_operators/sg_execution_times.html b/docs/how_to/optimize_operators/sg_execution_times.html
index 8d1c9e3891..64eab275a2 100644
--- a/docs/how_to/optimize_operators/sg_execution_times.html
+++ b/docs/how_to/optimize_operators/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-optimize-operators-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>00:35.612</strong> total execution time for <strong>how_to_optimize_operators</strong> files:</p>
+<p><strong>00:34.720</strong> total execution time for <strong>how_to_optimize_operators</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 83%" />
@@ -349,15 +349,15 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="opt_gemm.html#sphx-glr-how-to-optimize-operators-opt-gemm-py"><span class="std std-ref">How to optimize GEMM on CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">opt_gemm.py</span></code>)</p></td>
-<td><p>00:32.780</p></td>
+<td><p>00:32.194</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="opt_conv_tensorcore.html#sphx-glr-how-to-optimize-operators-opt-conv-tensorcore-py"><span class="std std-ref">How to optimize convolution using TensorCores</span></a> (<code class="docutils literal notranslate"><span class="pre">opt_conv_tensorcore.py</span></code>)</p></td>
-<td><p>00:01.637</p></td>
+<td><p>00:01.448</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="opt_conv_cuda.html#sphx-glr-how-to-optimize-operators-opt-conv-cuda-py"><span class="std std-ref">How to optimize convolution on GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">opt_conv_cuda.py</span></code>)</p></td>
-<td><p>00:01.195</p></td>
+<td><p>00:01.077</p></td>
<td><p>0.0 MB</p></td>
</tr>
</tbody>
diff --git a/docs/how_to/tune_with_autoscheduler/sg_execution_times.html b/docs/how_to/tune_with_autoscheduler/sg_execution_times.html
index 3a4f89abe7..7fe4406b4d 100644
--- a/docs/how_to/tune_with_autoscheduler/sg_execution_times.html
+++ b/docs/how_to/tune_with_autoscheduler/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-tune-with-autoscheduler-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>09:53.666</strong> total execution time for <strong>how_to_tune_with_autoscheduler</strong> files:</p>
+<p><strong>09:26.610</strong> total execution time for <strong>how_to_tune_with_autoscheduler</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 85%" />
@@ -349,27 +349,27 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_conv2d_layer_cuda.html#sphx-glr-how-to-tune-with-autoscheduler-tune-conv2d-layer-cuda-py"><span class="std std-ref">Auto-scheduling a Convolution Layer for GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_conv2d_layer_cuda.py</span></code>)</p></td>
-<td><p>05:58.328</p></td>
+<td><p>05:48.187</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tune_network_x86.html#sphx-glr-how-to-tune-with-autoscheduler-tune-network-x86-py"><span class="std std-ref">Auto-scheduling a Neural Network for x86 CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_network_x86.py</span></code>)</p></td>
-<td><p>01:41.158</p></td>
+<td><p>01:38.178</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_network_cuda.html#sphx-glr-how-to-tune-with-autoscheduler-tune-network-cuda-py"><span class="std std-ref">Auto-scheduling a Neural Network for NVIDIA GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_network_cuda.py</span></code>)</p></td>
-<td><p>01:05.726</p></td>
+<td><p>01:05.333</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tune_sparse_x86.html#sphx-glr-how-to-tune-with-autoscheduler-tune-sparse-x86-py"><span class="std std-ref">Auto-scheduling Sparse Matrix Multiplication on CPU with Custom Sketch Rule</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_sparse_x86.py</span></code>)</p></td>
-<td><p>00:41.572</p></td>
+<td><p>00:27.967</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_network_arm.html#sphx-glr-how-to-tune-with-autoscheduler-tune-network-arm-py"><span class="std std-ref">Auto-scheduling a Neural Network for ARM CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_network_arm.py</span></code>)</p></td>
-<td><p>00:14.008</p></td>
+<td><p>00:14.033</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tune_network_mali.html#sphx-glr-how-to-tune-with-autoscheduler-tune-network-mali-py"><span class="std std-ref">Auto-scheduling a Neural Network for mali GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_network_mali.py</span></code>)</p></td>
-<td><p>00:12.875</p></td>
+<td><p>00:12.913</p></td>
<td><p>0.0 MB</p></td>
</tr>
</tbody>
diff --git a/docs/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.html b/docs/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.html
index 511f670b1c..f8ec4e16c9 100644
--- a/docs/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.html
+++ b/docs/how_to/tune_with_autoscheduler/tune_conv2d_layer_cuda.html
@@ -489,6 +489,9 @@ file and apply it.</p>
<span class="k">del</span> <span class="n">measure_ctx</span>
</pre></div>
</div>
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>.T
+</pre></div>
+</div>
<p>We can lower the schedule to see the IR after auto-scheduling.
The auto-scheduler correctly performs optimizations including multi-level tiling,
cooperative fetching, unrolling and operator fusion.</p>
@@ -503,13 +506,13 @@ class Module:
def main(data: T.Buffer((1, 512, 7, 7), "float32"), kernel: T.Buffer((512, 512, 3, 3), "float32"), bias: T.Buffer((1, 512, 1, 1), "float32"), compute: T.Buffer((1, 512, 7, 7), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
blockIdx_x = T.env_thread("blockIdx.x")
- T.launch_thread(blockIdx_x, 28)
- conv2d_nchw = T.allocate([14], "float32", "local")
- pad_temp_shared = T.allocate([72], "float32", "shared")
- kernel_shared = T.allocate([3072], "float32", "shared")
+ T.launch_thread(blockIdx_x, 16)
+ conv2d_nchw = T.allocate([7], "float32", "local")
+ pad_temp_shared = T.allocate([648], "float32", "shared")
+ kernel_shared = T.allocate([2304], "float32", "shared")
threadIdx_x = T.env_thread("threadIdx.x")
- T.launch_thread(threadIdx_x, 64)
- conv2d_nchw_1 = T.buffer_decl((14,), data=conv2d_nchw, scope="local", align=32)
+ T.launch_thread(threadIdx_x, 224)
+ conv2d_nchw_1 = T.buffer_decl((7,), data=conv2d_nchw, scope="local", align=16)
conv2d_nchw_1[0] = T.float32(0)
conv2d_nchw_1[1] = T.float32(0)
conv2d_nchw_1[2] = T.float32(0)
@@ -517,467 +520,553 @@ class Module:
conv2d_nchw_1[4] = T.float32(0)
conv2d_nchw_1[5] = T.float32(0)
conv2d_nchw_1[6] = T.float32(0)
- conv2d_nchw_1[7] = T.float32(0)
- conv2d_nchw_1[8] = T.float32(0)
- conv2d_nchw_1[9] = T.float32(0)
- conv2d_nchw_1[10] = T.float32(0)
- conv2d_nchw_1[11] = T.float32(0)
- conv2d_nchw_1[12] = T.float32(0)
- conv2d_nchw_1[13] = T.float32(0)
- for rc_outer_outer, ry_outer_outer in T.grid(64, 3):
- cse_var_2: T.int32 = rc_outer_outer * 72
- cse_var_1: T.int32 = ry_outer_outer * 3
+ for rc_outer_outer in range(64):
+ cse_var_2: T.int32 = rc_outer_outer * 392
+ cse_var_1: T.int32 = rc_outer_outer * 72
threadIdx_x_1 = T.env_thread("threadIdx.x")
- pad_temp_shared_1 = T.buffer_decl((72,), data=pad_temp_shared, scope="shared")
- with T.launch_thread(threadIdx_x_1, 64):
- data_1 = T.buffer_decl((25088,), data=data.data)
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= threadIdx_x_1 * 4 % 9 and threadIdx_x_1 * 4 % 9 < 8, data_1[rc_outer_outer * 392 + threadIdx_x_1 * 4 // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + threadIdx_x_1 * 4 % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 1] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 1) % 9 and (threadIdx_x_1 * 4 + 1) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 1) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 1) % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 2] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 2) % 9 and (threadIdx_x_1 * 4 + 2) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 2) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 2) % 9 - 8], T.float32(0))
- if T.likely(threadIdx_x_1 < 18):
- pad_temp_shared_1[threadIdx_x_1 * 4 + 3] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 3) % 9 and (threadIdx_x_1 * 4 + 3) % 9 < 8, data_1[rc_outer_outer * 392 + (threadIdx_x_1 * 4 + 3) // 9 * 49 + ry_outer_outer * 7 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 3) % 9 - 8], T.float32(0))
+ pad_temp_shared_1 = T.buffer_decl((648,), data=pad_temp_shared, scope="shared")
+ data_1 = T.buffer_decl((25088,), data=data.data)
+ with T.launch_thread(threadIdx_x_1, 224):
+ pad_temp_shared_1[threadIdx_x_1] = T.if_then_else(9 <= threadIdx_x_1 % 81 and threadIdx_x_1 % 81 < 72 and 1 <= threadIdx_x_1 % 9 and threadIdx_x_1 % 9 < 8, data_1[cse_var_2 + threadIdx_x_1 // 81 * 49 + threadIdx_x_1 % 81 // 9 * 7 + threadIdx_x_1 % 9 - 8], T.float32(0))
+ with T.launch_thread(threadIdx_x_1, 224):
+ pad_temp_shared_1[threadIdx_x_1 + 224] = T.if_then_else(9 <= (threadIdx_x_1 + 62) % 81 and (threadIdx_x_1 + 62) % 81 < 72 and 1 <= (threadIdx_x_1 + 8) % 9 and (threadIdx_x_1 + 8) % 9 < 8, data_1[cse_var_2 + (threadIdx_x_1 + 224) // 81 * 49 + (threadIdx_x_1 + 62) % 81 // 9 * 7 + (threadIdx_x_1 + 8) % 9 - 8], T.float32(0))
+ with T.launch_thread(threadIdx_x_1, 224):
+ if T.likely(threadIdx_x_1 < 200):
+ pad_temp_shared_1[threadIdx_x_1 + 448] = T.if_then_else(9 <= (threadIdx_x_1 + 43) % 81 and (threadIdx_x_1 + 43) % 81 < 72 and 1 <= (threadIdx_x_1 + 7) % 9 and (threadIdx_x_1 + 7) % 9 < 8, data_1[cse_var_2 + (threadIdx_x_1 + 448) // 81 * 49 + (threadIdx_x_1 + 43) % 81 // 9 * 7 + (threadIdx_x_1 + 7) % 9 - 8], T.float32(0))
threadIdx_x_2 = T.env_thread("threadIdx.x")
- kernel_shared_1 = T.buffer_decl((3072,), data=kernel_shared, scope="shared")
+ kernel_shared_1 = T.buffer_decl((2304,), data=kernel_shared, scope="shared")
kernel_1 = T.buffer_decl((2359296,), data=kernel.data)
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 64] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 64) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 128] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 128) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 192] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 36864]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 256] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 256) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 320] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 320) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 384] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 73728]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 448] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 448) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 512] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 512) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 576] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 110592]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 640] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 640) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 704] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 704) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 768] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 147456]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 832] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 832) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 896] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 896) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 960] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 184320]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1024] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1024) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1088] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1088) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1152] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 221184]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1216] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1216) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1280] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1280) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1344] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 258048]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1408] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1408) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1472] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1472) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1536] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 294912]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1600] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1600) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1664] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1664) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1728] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 331776]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1792] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1792) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1856] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1856) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1920] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 368640]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 1984] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 1984) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2048] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2048) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2112] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 405504]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2176] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2176) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2240] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2240) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2304] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 442368]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2368] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2368) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2432] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2432) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2496] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 479232]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2560] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2560) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2624] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2624) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2688] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 516096]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2752] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2752) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2816] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2816) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2880] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_2 // 24 * 4608 + cse_var_2 + threadIdx_x_2 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_2 % 3 + 552960]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 2944] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 2944) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 1) % 3]
- with T.launch_thread(threadIdx_x_2, 64):
- kernel_shared_1[threadIdx_x_2 + 3008] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_2 + 3008) // 24 * 4608 + cse_var_2 + (threadIdx_x_2 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_2 + 2) % 3]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 3]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 24]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 27]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 1]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 4]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 25]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 28]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 2]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 5]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 26]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 29]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 6]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 9]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 30]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 33]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 7]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 10]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 31]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 34]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 8]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 11]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 32]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 35]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 12]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 15]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 36]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 39]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 13]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 16]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 37]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 40]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 14]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 17]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 38]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 41]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 18]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 21]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 42]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 45]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 19]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 22]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 43]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 46]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 20]
- conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 23]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 47]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 44]
- conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 47]
- for i1_inner, i3_inner in T.grid(2, 7):
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2] = kernel_1[blockIdx_x * 147456 + threadIdx_x_2 // 72 * 4608 + cse_var_1 + threadIdx_x_2 % 72]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 224] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 224) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 8) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 448] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 448) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 16) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 672] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 672) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 // 3 + 8) % 24 * 3 + threadIdx_x_2 % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 896] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 896) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 32) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1120] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1120) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 40) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1344] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1344) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 // 3 + 16) % 24 * 3 + threadIdx_x_2 % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1568] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1568) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 56) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 1792] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 1792) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 64) % 72 // 3 * 3 + (threadIdx_x_2 + 1) % 3]
+ with T.launch_thread(threadIdx_x_2, 224):
+ kernel_shared_1[threadIdx_x_2 + 2016] = kernel_1[blockIdx_x * 147456 + threadIdx_x_2 // 72 * 4608 + cse_var_1 + threadIdx_x_2 % 72 + 129024]
+ with T.launch_thread(threadIdx_x_2, 224):
+ if T.likely(threadIdx_x_2 < 64):
+ kernel_shared_1[threadIdx_x_2 + 2240] = kernel_1[blockIdx_x * 147456 + (threadIdx_x_2 + 2240) // 72 * 4608 + cse_var_1 + (threadIdx_x_2 + 8) % 72 // 3 * 3 + (threadIdx_x_2 + 2) % 3]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 1] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 81] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 82] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 1] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 82] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 2] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 83] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 3] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 84] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 4] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 85] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 5] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 7] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 86] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 88] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 6] * kernel_shared_1[threadIdx_x // 7 * 72]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 7] * kernel_shared_1[threadIdx_x // 7 * 72 + 1]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 8] * kernel_shared_1[threadIdx_x // 7 * 72 + 2]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 87] * kernel_shared_1[threadIdx_x // 7 * 72 + 9]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 88] * kernel_shared_1[threadIdx_x // 7 * 72 + 10]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 89] * kernel_shared_1[threadIdx_x // 7 * 72 + 11]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 9] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 10] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 90] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 91] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 10] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 91] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 11] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 92] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 12] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 93] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 13] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 94] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 14] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 16] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 95] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 97] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 15] * kernel_shared_1[threadIdx_x // 7 * 72 + 3]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 16] * kernel_shared_1[threadIdx_x // 7 * 72 + 4]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 17] * kernel_shared_1[threadIdx_x // 7 * 72 + 5]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 96] * kernel_shared_1[threadIdx_x // 7 * 72 + 12]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 97] * kernel_shared_1[threadIdx_x // 7 * 72 + 13]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 98] * kernel_shared_1[threadIdx_x // 7 * 72 + 14]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 18] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 19] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 99] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 100] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 19] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 100] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 20] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 101] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 21] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 102] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 22] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 103] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 23] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 25] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 104] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 106] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 24] * kernel_shared_1[threadIdx_x // 7 * 72 + 6]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 25] * kernel_shared_1[threadIdx_x // 7 * 72 + 7]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 26] * kernel_shared_1[threadIdx_x // 7 * 72 + 8]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 105] * kernel_shared_1[threadIdx_x // 7 * 72 + 15]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 106] * kernel_shared_1[threadIdx_x // 7 * 72 + 16]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 107] * kernel_shared_1[threadIdx_x // 7 * 72 + 17]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 162] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 163] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 243] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 244] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 163] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 244] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 164] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 245] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 165] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 246] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 166] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 247] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 167] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 169] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 248] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 250] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 168] * kernel_shared_1[threadIdx_x // 7 * 72 + 18]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 169] * kernel_shared_1[threadIdx_x // 7 * 72 + 19]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 170] * kernel_shared_1[threadIdx_x // 7 * 72 + 20]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 249] * kernel_shared_1[threadIdx_x // 7 * 72 + 27]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 250] * kernel_shared_1[threadIdx_x // 7 * 72 + 28]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 251] * kernel_shared_1[threadIdx_x // 7 * 72 + 29]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 171] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 172] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 252] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 253] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 172] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 253] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 173] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 254] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 174] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 255] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 175] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 256] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 176] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 178] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 257] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 259] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 177] * kernel_shared_1[threadIdx_x // 7 * 72 + 21]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 178] * kernel_shared_1[threadIdx_x // 7 * 72 + 22]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 179] * kernel_shared_1[threadIdx_x // 7 * 72 + 23]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 258] * kernel_shared_1[threadIdx_x // 7 * 72 + 30]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 259] * kernel_shared_1[threadIdx_x // 7 * 72 + 31]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 260] * kernel_shared_1[threadIdx_x // 7 * 72 + 32]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 180] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 181] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 261] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 262] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 181] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 262] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 182] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 263] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 183] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 264] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 184] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 265] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 185] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 187] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 266] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 268] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 186] * kernel_shared_1[threadIdx_x // 7 * 72 + 24]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 187] * kernel_shared_1[threadIdx_x // 7 * 72 + 25]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 188] * kernel_shared_1[threadIdx_x // 7 * 72 + 26]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 267] * kernel_shared_1[threadIdx_x // 7 * 72 + 33]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 268] * kernel_shared_1[threadIdx_x // 7 * 72 + 34]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 269] * kernel_shared_1[threadIdx_x // 7 * 72 + 35]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 324] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 325] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 405] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 406] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 325] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 406] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 326] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 407] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 327] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 408] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 328] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 409] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 329] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 331] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 410] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 412] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 330] * kernel_shared_1[threadIdx_x // 7 * 72 + 36]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 331] * kernel_shared_1[threadIdx_x // 7 * 72 + 37]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 332] * kernel_shared_1[threadIdx_x // 7 * 72 + 38]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 411] * kernel_shared_1[threadIdx_x // 7 * 72 + 45]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 412] * kernel_shared_1[threadIdx_x // 7 * 72 + 46]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 413] * kernel_shared_1[threadIdx_x // 7 * 72 + 47]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 333] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 334] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 414] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 415] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 334] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 415] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 335] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 416] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 336] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 417] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 337] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 418] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 338] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 340] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 419] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 421] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 339] * kernel_shared_1[threadIdx_x // 7 * 72 + 39]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 340] * kernel_shared_1[threadIdx_x // 7 * 72 + 40]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 341] * kernel_shared_1[threadIdx_x // 7 * 72 + 41]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 420] * kernel_shared_1[threadIdx_x // 7 * 72 + 48]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 421] * kernel_shared_1[threadIdx_x // 7 * 72 + 49]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 422] * kernel_shared_1[threadIdx_x // 7 * 72 + 50]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 342] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 343] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 423] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 424] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 343] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 424] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 344] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 425] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 345] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 426] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 346] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 427] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 347] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 349] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 428] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 430] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 348] * kernel_shared_1[threadIdx_x // 7 * 72 + 42]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 349] * kernel_shared_1[threadIdx_x // 7 * 72 + 43]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 350] * kernel_shared_1[threadIdx_x // 7 * 72 + 44]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 429] * kernel_shared_1[threadIdx_x // 7 * 72 + 51]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 430] * kernel_shared_1[threadIdx_x // 7 * 72 + 52]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 431] * kernel_shared_1[threadIdx_x // 7 * 72 + 53]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 486] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 487] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 567] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 568] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 487] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 568] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 488] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 569] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 489] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 570] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 490] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 571] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 491] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 493] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 572] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 574] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 492] * kernel_shared_1[threadIdx_x // 7 * 72 + 54]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 493] * kernel_shared_1[threadIdx_x // 7 * 72 + 55]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 494] * kernel_shared_1[threadIdx_x // 7 * 72 + 56]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 573] * kernel_shared_1[threadIdx_x // 7 * 72 + 63]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 574] * kernel_shared_1[threadIdx_x // 7 * 72 + 64]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 575] * kernel_shared_1[threadIdx_x // 7 * 72 + 65]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 495] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 496] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 576] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 577] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 496] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 577] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 497] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 578] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 498] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 579] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 499] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 580] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 500] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 502] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 581] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 583] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 501] * kernel_shared_1[threadIdx_x // 7 * 72 + 57]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 502] * kernel_shared_1[threadIdx_x // 7 * 72 + 58]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 503] * kernel_shared_1[threadIdx_x // 7 * 72 + 59]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 582] * kernel_shared_1[threadIdx_x // 7 * 72 + 66]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 583] * kernel_shared_1[threadIdx_x // 7 * 72 + 67]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 584] * kernel_shared_1[threadIdx_x // 7 * 72 + 68]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 504] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 505] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 585] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 586] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 505] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 586] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 506] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 587] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 507] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 588] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 508] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 589] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 509] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 511] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 590] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 592] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 510] * kernel_shared_1[threadIdx_x // 7 * 72 + 60]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 511] * kernel_shared_1[threadIdx_x // 7 * 72 + 61]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 512] * kernel_shared_1[threadIdx_x // 7 * 72 + 62]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 591] * kernel_shared_1[threadIdx_x // 7 * 72 + 69]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 592] * kernel_shared_1[threadIdx_x // 7 * 72 + 70]
+ conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x % 7 * 9 + 593] * kernel_shared_1[threadIdx_x // 7 * 72 + 71]
+ for i3_inner in range(7):
compute_1 = T.buffer_decl((25088,), data=compute.data)
bias_1 = T.buffer_decl((512,), data=bias.data)
- compute_1[blockIdx_x // 7 * 6272 + threadIdx_x * 98 + i1_inner * 49 + blockIdx_x % 7 * 7 + i3_inner] = T.max(conv2d_nchw_1[i1_inner * 7 + i3_inner] + bias_1[blockIdx_x // 7 * 128 + threadIdx_x * 2 + i1_inner], T.float32(0))
+ compute_1[blockIdx_x * 1568 + threadIdx_x * 7 + i3_inner] = T.max(conv2d_nchw_1[i3_inner] + bias_1[blockIdx_x * 32 + threadIdx_x // 7], T.float32(0))
</pre></div>
</div>
</div>
@@ -1011,7 +1100,7 @@ class Module:
<span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time of this operator: 0.356 ms
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time of this operator: 0.245 ms
</pre></div>
</div>
</div>
@@ -1041,12 +1130,12 @@ conv2d_nchw_nn_o_o_i, conv2d_nchw_nn_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o
conv2d_nchw_nn_o_o_o_i, conv2d_nchw_nn_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_i, factor=1)
conv2d_nchw_nn_o_o_o_o, conv2d_nchw_nn_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_o_i, factor=1)
conv2d_nchw_ff_o_i, conv2d_nchw_ff_i = s[conv2d_nchw].split(conv2d_nchw_ff, factor=1)
-conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=2)
-conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=64)
+conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=1)
+conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=32)
conv2d_nchw_ff_o_o_o_o, conv2d_nchw_ff_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_o_i, factor=1)
conv2d_nchw_yy_o_i, conv2d_nchw_yy_i = s[conv2d_nchw].split(conv2d_nchw_yy, factor=1)
conv2d_nchw_yy_o_o_i, conv2d_nchw_yy_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_i, factor=1)
-conv2d_nchw_yy_o_o_o_i, conv2d_nchw_yy_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_i, factor=1)
+conv2d_nchw_yy_o_o_o_i, conv2d_nchw_yy_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_i, factor=7)
conv2d_nchw_yy_o_o_o_o, conv2d_nchw_yy_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_o_i, factor=1)
conv2d_nchw_xx_o_i, conv2d_nchw_xx_i = s[conv2d_nchw].split(conv2d_nchw_xx, factor=1)
conv2d_nchw_xx_o_o_i, conv2d_nchw_xx_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_i, factor=7)
@@ -1055,18 +1144,18 @@ conv2d_nchw_xx_o_o_o_o, conv2d_nchw_xx_o_o_o_i = s[conv2d_nchw].split(conv2d_nch
conv2d_nchw_rc_o_i, conv2d_nchw_rc_i = s[conv2d_nchw].split(conv2d_nchw_rc, factor=2)
conv2d_nchw_rc_o_o, conv2d_nchw_rc_o_i = s[conv2d_nchw].split(conv2d_nchw_rc_o_i, factor=4)
conv2d_nchw_ry_o_i, conv2d_nchw_ry_i = s[conv2d_nchw].split(conv2d_nchw_ry, factor=1)
-conv2d_nchw_ry_o_o, conv2d_nchw_ry_o_i = s[conv2d_nchw].split(conv2d_nchw_ry_o_i, factor=1)
-conv2d_nchw_rx_o_i, conv2d_nchw_rx_i = s[conv2d_nchw].split(conv2d_nchw_rx, factor=1)
-conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_o_i, factor=3)
+conv2d_nchw_ry_o_o, conv2d_nchw_ry_o_i = s[conv2d_nchw].split(conv2d_nchw_ry_o_i, factor=3)
+conv2d_nchw_rx_o_i, conv2d_nchw_rx_i = s[conv2d_nchw].split(conv2d_nchw_rx, factor=3)
+conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_o_i, factor=1)
s[conv2d_nchw].reorder(conv2d_nchw_nn_o_o_o_o, conv2d_nchw_ff_o_o_o_o, conv2d_nchw_yy_o_o_o_o, conv2d_nchw_xx_o_o_o_o, conv2d_nchw_nn_o_o_o_i, conv2d_nchw_ff_o_o_o_i, conv2d_nchw_yy_o_o_o_i, conv2d_nchw_xx_o_o_o_i, conv2d_nchw_nn_o_o_i, conv2d_nchw_ff_o_o_i, conv2d_nchw_yy_o_o_i, conv2d_nchw_xx_o_o_i, conv2d_nchw_rc_o_o, conv2d_nchw_ry_o_o, conv2d_nchw_rx_o_o, conv2d_nchw_rc_o_i, conv2d_nchw_ry_o_i, conv2d_nchw_rx_o_i, conv2d_nchw_nn_o_i, conv2d_nchw_ff_o_i, conv2d_nchw_yy_o_i, conv2d_nc [...]
compute_i0_o_i, compute_i0_i = s[compute].split(compute_i0, factor=1)
compute_i0_o_o_i, compute_i0_o_i = s[compute].split(compute_i0_o_i, factor=1)
compute_i0_o_o_o, compute_i0_o_o_i = s[compute].split(compute_i0_o_o_i, factor=1)
-compute_i1_o_i, compute_i1_i = s[compute].split(compute_i1, factor=2)
-compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=64)
+compute_i1_o_i, compute_i1_i = s[compute].split(compute_i1, factor=1)
+compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=32)
compute_i1_o_o_o, compute_i1_o_o_i = s[compute].split(compute_i1_o_o_i, factor=1)
compute_i2_o_i, compute_i2_i = s[compute].split(compute_i2, factor=1)
-compute_i2_o_o_i, compute_i2_o_i = s[compute].split(compute_i2_o_i, factor=1)
+compute_i2_o_o_i, compute_i2_o_i = s[compute].split(compute_i2_o_i, factor=7)
compute_i2_o_o_o, compute_i2_o_o_i = s[compute].split(compute_i2_o_o_i, factor=1)
compute_i3_o_i, compute_i3_i = s[compute].split(compute_i3, factor=7)
compute_i3_o_o_i, compute_i3_o_i = s[compute].split(compute_i3_o_i, factor=1)
@@ -1089,14 +1178,14 @@ s[compute].bind(compute_i0_o_i_i1_o_i_fused_i2_o_i_fused_i3_o_i_fused, te.thread
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[kernel_shared].fuse(kernel_shared_ax0, kernel_shared_ax1, kernel_shared_ax2, kernel_shared_ax3)
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=1)
s[kernel_shared].vectorize(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
-kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
+kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=224)
s[kernel_shared].bind(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[pad_temp_shared].fuse(pad_temp_shared_ax0, pad_temp_shared_ax1, pad_temp_shared_ax2, pad_temp_shared_ax3)
-pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=4)
+pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=1)
s[pad_temp_shared].vectorize(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
-pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
+pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=224)
s[pad_temp_shared].bind(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
-s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "auto_unroll_max_step", 512)
+s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "auto_unroll_max_step", 1024)
s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "unroll_explicit", True)
CUDA source code:
@@ -1114,10 +1203,10 @@ CUDA source code:
#define int64_t long long
#define uint64_t unsigned long long
#endif
-extern "C" __global__ void __launch_bounds__(64) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
- float conv2d_nchw[14];
- __shared__ float pad_temp_shared[72];
- __shared__ float kernel_shared[3072];
+extern "C" __global__ void __launch_bounds__(224) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
+ float conv2d_nchw[7];
+ __shared__ float pad_temp_shared[648];
+ __shared__ float kernel_shared[2304];
conv2d_nchw[0] = 0.000000e+00f;
conv2d_nchw[1] = 0.000000e+00f;
conv2d_nchw[2] = 0.000000e+00f;
@@ -1125,419 +1214,534 @@ extern "C" __global__ void __launch_bounds__(64) default_function_kern
conv2d_nchw[4] = 0.000000e+00f;
conv2d_nchw[5] = 0.000000e+00f;
conv2d_nchw[6] = 0.000000e+00f;
- conv2d_nchw[7] = 0.000000e+00f;
- conv2d_nchw[8] = 0.000000e+00f;
- conv2d_nchw[9] = 0.000000e+00f;
- conv2d_nchw[10] = 0.000000e+00f;
- conv2d_nchw[11] = 0.000000e+00f;
- conv2d_nchw[12] = 0.000000e+00f;
- conv2d_nchw[13] = 0.000000e+00f;
for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) {
- for (int ry_outer_outer = 0; ry_outer_outer < 3; ++ry_outer_outer) {
- __syncthreads();
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[(((int)threadIdx.x) * 4)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= ((((int)threadIdx.x) * 4) % 9))) && (((((int)threadIdx.x) * 4) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + (((((int)threadIdx.x) * 4) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + ((((int)threadIdx.x) * 4) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 1)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 1) % 9))) && ((((((int)threadIdx.x) * 4) + 1) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 1) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 1) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 2)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 2) % 9))) && ((((((int)threadIdx.x) * 4) + 2) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 2) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 2) % 9)) - 8)] : 0.000000e+00f);
- }
- if (((int)threadIdx.x) < 18) {
- pad_temp_shared[((((int)threadIdx.x) * 4) + 3)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 3) % 9))) && ((((((int)threadIdx.x) * 4) + 3) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 3) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 3) % 9)) - 8)] : 0.000000e+00f);
- }
- kernel_shared[((int)threadIdx.x)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3))];
- kernel_shared[(((int)threadIdx.x) + 64)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 64) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 128)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 128) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 192)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 36864)];
- kernel_shared[(((int)threadIdx.x) + 256)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 256) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 320)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 320) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 384)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 73728)];
- kernel_shared[(((int)threadIdx.x) + 448)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 448) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 512)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 512) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 576)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 110592)];
- kernel_shared[(((int)threadIdx.x) + 640)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 640) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 704)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 704) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 768)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 147456)];
- kernel_shared[(((int)threadIdx.x) + 832)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 832) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 896)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 896) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 960)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 184320)];
- kernel_shared[(((int)threadIdx.x) + 1024)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1024) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1088)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1088) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1152)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 221184)];
- kernel_shared[(((int)threadIdx.x) + 1216)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1216) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1280)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1280) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1344)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 258048)];
- kernel_shared[(((int)threadIdx.x) + 1408)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1408) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1472)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1472) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1536)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 294912)];
- kernel_shared[(((int)threadIdx.x) + 1600)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1600) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1664)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1664) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1728)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 331776)];
- kernel_shared[(((int)threadIdx.x) + 1792)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1792) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1856)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1856) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 1920)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 368640)];
- kernel_shared[(((int)threadIdx.x) + 1984)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1984) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2048)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2048) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2112)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 405504)];
- kernel_shared[(((int)threadIdx.x) + 2176)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2176) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2240)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2240) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2304)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 442368)];
- kernel_shared[(((int)threadIdx.x) + 2368)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2368) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2432)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2432) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2496)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 479232)];
- kernel_shared[(((int)threadIdx.x) + 2560)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2560) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2624)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2624) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2688)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 516096)];
- kernel_shared[(((int)threadIdx.x) + 2752)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2752) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2816)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2816) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- kernel_shared[(((int)threadIdx.x) + 2880)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 552960)];
- kernel_shared[(((int)threadIdx.x) + 2944)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2944) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
- kernel_shared[(((int)threadIdx.x) + 3008)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 3008) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
- __syncthreads();
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[0] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[1] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[2] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[3] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[4] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[5] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[6] * kernel_shared[(((int)threadIdx.x) * 48)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[0] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
- conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
- conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
+ __syncthreads();
+ pad_temp_shared[((int)threadIdx.x)] = (((((9 <= (((int)threadIdx.x) % 81)) && ((((int)threadIdx.x) % 81) < 72)) && (1 <= (((int)threadIdx.x) % 9))) && ((((int)threadIdx.x) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + ((((int)threadIdx.x) / 81) * 49)) + (((((int)threadIdx.x) % 81) / 9) * 7)) + (((int)threadIdx.x) % 9)) - 8)] : 0.000000e+00f);
+ pad_temp_shared[(((int)threadIdx.x) + 224)] = (((((9 <= ((((int)threadIdx.x) + 62) % 81)) && (((((int)threadIdx.x) + 62) % 81) < 72)) && (1 <= ((((int)threadIdx.x) + 8) % 9))) && (((((int)threadIdx.x) + 8) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 224) / 81) * 49)) + ((((((int)threadIdx.x) + 62) % 81) / 9) * 7)) + ((((int)threadIdx.x) + 8) % 9)) - 8)] : 0.000000e+00f);
+ if (((int)threadIdx.x) < 200) {
+ pad_temp_shared[(((int)threadIdx.x) + 448)] = (((((9 <= ((((int)threadIdx.x) + 43) % 81)) && (((((int)threadIdx.x) + 43) % 81) < 72)) && (1 <= ((((int)threadIdx.x) + 7) % 9))) && (((((int)threadIdx.x) + 7) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 448) / 81) * 49)) + ((((((int)threadIdx.x) + 43) % 81) / 9) * 7)) + ((((int)threadIdx.x) + 7) % 9)) - 8)] : 0.000000e+00f);
}
- }
- for (int i1_inner = 0; i1_inner < 2; ++i1_inner) {
- for (int i3_inner = 0; i3_inner < 7; ++i3_inner) {
- compute[((((((((int)blockIdx.x) / 7) * 6272) + (((int)threadIdx.x) * 98)) + (i1_inner * 49)) + ((((int)blockIdx.x) % 7) * 7)) + i3_inner)] = max((conv2d_nchw[((i1_inner * 7) + i3_inner)] + bias[((((((int)blockIdx.x) / 7) * 128) + (((int)threadIdx.x) * 2)) + i1_inner)]), 0.000000e+00f);
+ kernel_shared[((int)threadIdx.x)] = kernel[((((((int)blockIdx.x) * 147456) + ((((int)threadIdx.x) / 72) * 4608)) + (rc_outer_outer * 72)) + (((int)threadIdx.x) % 72))];
+ kernel_shared[(((int)threadIdx.x) + 224)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 224) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 448)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 448) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 672)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 672) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) / 3) + 8) % 24) * 3)) + (((int)threadIdx.x) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 896)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 896) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 32) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1120)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1120) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 40) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1344)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1344) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) / 3) + 16) % 24) * 3)) + (((int)threadIdx.x) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1568)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1568) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 56) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 1792)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 1792) / 72) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 64) % 72) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))];
+ kernel_shared[(((int)threadIdx.x) + 2016)] = kernel[(((((((int)blockIdx.x) * 147456) + ((((int)threadIdx.x) / 72) * 4608)) + (rc_outer_outer * 72)) + (((int)threadIdx.x) % 72)) + 129024)];
+ if (((int)threadIdx.x) < 64) {
+ kernel_shared[(((int)threadIdx.x) + 2240)] = kernel[(((((((int)blockIdx.x) * 147456) + (((((int)threadIdx.x) + 2240) / 72) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) + 8) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))];
}
+ __syncthreads();
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) * 9)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 1)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 81)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 1)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 2)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 3)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 84)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 4)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 85)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 5)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 7)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 86)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 88)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 6)] * kernel_shared[((((int)threadIdx.x) / 7) * 72)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 7)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 1)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 8)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 2)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 87)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 9)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 88)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 10)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 89)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 11)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 9)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 12)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 93)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 13)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 94)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 14)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 16)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 95)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 97)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 15)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 3)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 16)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 4)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 17)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 5)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 96)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 12)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 97)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 13)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 98)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 14)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 21)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 102)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 22)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 103)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 23)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 25)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 104)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 106)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 24)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 6)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 25)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 7)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 26)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 8)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 105)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 15)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 106)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 16)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 107)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 17)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 162)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 163)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 243)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 244)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 163)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 244)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 165)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 246)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 166)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 247)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 167)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 169)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 248)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 250)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 168)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 18)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 169)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 19)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 170)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 20)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 249)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 27)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 250)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 28)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 251)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 29)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 171)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 172)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 252)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 253)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 172)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 253)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 174)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 255)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 175)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 256)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 176)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 178)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 257)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 259)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 177)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 21)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 178)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 22)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 179)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 23)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 258)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 30)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 259)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 31)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 260)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 32)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 180)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 181)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 261)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 262)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 181)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 262)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 183)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 264)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 184)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 265)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 185)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 187)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 266)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 268)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 186)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 24)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 187)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 25)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 188)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 26)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 267)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 33)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 268)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 34)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 269)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 35)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 324)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 325)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 405)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 406)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 325)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 406)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 327)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 408)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 328)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 409)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 329)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 331)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 410)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 412)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 330)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 36)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 331)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 37)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 332)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 38)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 411)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 45)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 412)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 46)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 413)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 47)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 333)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 334)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 414)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 415)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 334)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 415)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 336)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 417)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 337)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 418)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 338)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 340)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 419)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 421)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 339)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 39)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 340)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 40)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 341)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 41)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 420)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 48)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 421)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 49)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 422)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 50)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 342)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 343)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 423)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 424)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 343)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 424)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 345)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 426)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 346)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 427)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 347)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 349)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 428)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 430)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 348)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 42)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 349)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 43)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 350)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 44)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 429)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 51)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 430)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 52)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 431)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 53)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 486)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 487)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 567)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 568)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 487)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 568)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 569)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 489)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 570)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 490)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 571)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 491)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 493)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 572)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 574)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 492)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 54)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 493)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 55)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 494)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 56)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 573)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 63)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 574)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 64)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 575)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 65)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 495)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 496)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 576)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 577)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 496)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 577)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 578)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 498)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 579)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 499)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 580)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 500)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 502)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 581)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 583)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 501)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 57)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 502)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 58)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 503)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 59)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 582)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 66)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 583)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 67)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 584)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 68)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 504)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 505)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 585)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 586)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 505)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 586)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 506)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 587)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 507)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 588)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 508)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 589)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 509)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 511)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 590)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 592)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 510)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 60)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 511)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 61)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 512)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 62)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 591)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 69)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 592)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 70)]));
+ conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[(((((int)threadIdx.x) % 7) * 9) + 593)] * kernel_shared[(((((int)threadIdx.x) / 7) * 72) + 71)]));
+ }
+ for (int i3_inner = 0; i3_inner < 7; ++i3_inner) {
+ compute[(((((int)blockIdx.x) * 1568) + (((int)threadIdx.x) * 7)) + i3_inner)] = max((conv2d_nchw[i3_inner] + bias[((((int)blockIdx.x) * 32) + (((int)threadIdx.x) / 7))]), 0.000000e+00f);
}
}
</pre></div>
@@ -1572,10 +1776,9 @@ In the example below we resume the status and do more 5 trials.</p>
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/xgboost/training.py:17: UserWarning: Old style callback is deprecated. See: https://xgboost.readthedocs.io/en/latest/python/callbacks.html
warnings.warn(f'Old style callback is deprecated. See: {link}', UserWarning)
Get devices for measurement successfully!
-.T
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 5 minutes 58.328 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 5 minutes 48.187 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-tune-with-autoscheduler-tune-conv2d-layer-cuda-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/e3e540f3b477c0c52d8eb73e674e8ffd/tune_conv2d_layer_cuda.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">tune_conv2d_layer_cuda.py</span></code></a></p>
diff --git a/docs/how_to/tune_with_autoscheduler/tune_network_cuda.html b/docs/how_to/tune_with_autoscheduler/tune_network_cuda.html
index 2eb7d0384a..173e483796 100644
--- a/docs/how_to/tune_with_autoscheduler/tune_network_cuda.html
+++ b/docs/how_to/tune_with_autoscheduler/tune_network_cuda.html
@@ -916,7 +916,7 @@ so we can read the log file and load the best schedules.</p>
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 7.8804 7.8801 7.8819 7.8793 0.0011
+ 7.8523 7.8530 7.8578 7.8460 0.0048
</pre></div>
</div>
</div>
@@ -938,7 +938,7 @@ to learn how to use the RPC Tracker and RPC Server.
To use the RPC Tracker in auto-scheduler, replace the runner in <code class="code docutils literal notranslate"><span class="pre">TuningOptions</span></code>
with <a class="reference internal" href="../../reference/api/python/auto_scheduler.html#tvm.auto_scheduler.RPCRunner" title="tvm.auto_scheduler.RPCRunner"><code class="xref any py py-class docutils literal notranslate"><span class="pre">auto_scheduler.RPCRunner</span></code></a>.</p></li>
</ol>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 5.726 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 5.333 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-tune-with-autoscheduler-tune-network-cuda-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/eafe360d52540634c9eea0fa89e804bd/tune_network_cuda.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">tune_network_cuda.py</span></code></a></p>
diff --git a/docs/how_to/tune_with_autoscheduler/tune_network_x86.html b/docs/how_to/tune_with_autoscheduler/tune_network_x86.html
index e11a1851e1..86b032d168 100644
--- a/docs/how_to/tune_with_autoscheduler/tune_network_x86.html
+++ b/docs/how_to/tune_with_autoscheduler/tune_network_x86.html
@@ -935,7 +935,7 @@ so we can read the log file and load the best schedules.</p>
Evaluate inference time cost...
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
- 786.5811 789.7259 790.2548 779.7626 4.8262
+ 752.2132 751.8728 754.3859 750.3809 1.6526
</pre></div>
</div>
</div>
@@ -957,7 +957,7 @@ to learn how to use the RPC Tracker and RPC Server.
To use the RPC Tracker in auto-scheduler, replace the runner in <code class="code docutils literal notranslate"><span class="pre">TuningOptions</span></code>
with <a class="reference internal" href="../../reference/api/python/auto_scheduler.html#tvm.auto_scheduler.RPCRunner" title="tvm.auto_scheduler.RPCRunner"><code class="xref any py py-class docutils literal notranslate"><span class="pre">auto_scheduler.RPCRunner</span></code></a>.</p></li>
</ol>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 41.158 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 38.178 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-tune-with-autoscheduler-tune-network-x86-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/e416b94ca1090b0897c0f6e0df95b911/tune_network_x86.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">tune_network_x86.py</span></code></a></p>
diff --git a/docs/how_to/tune_with_autoscheduler/tune_sparse_x86.html b/docs/how_to/tune_with_autoscheduler/tune_sparse_x86.html
index da7c5c411c..8fc75c398a 100644
--- a/docs/how_to/tune_with_autoscheduler/tune_sparse_x86.html
+++ b/docs/how_to/tune_with_autoscheduler/tune_sparse_x86.html
@@ -629,74 +629,27 @@ class Module:
@T.prim_func
def main(placeholder: T.Buffer((128, 256), "float32"), placeholder_1: T.Buffer((4916, 16, 1), "float32"), placeholder_2: T.Buffer((4916,), "int32"), placeholder_3: T.Buffer((33,), "int32"), placeholder_4: T.Buffer((128, 512), "float32"), compute: T.Buffer((128, 512), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
- for i0_outer_i1_outer_fused in T.parallel(64):
- compute_1 = T.allocate([1024], "float32", "global")
- compute_2 = T.buffer_decl((1024,), data=compute_1)
- for i_outer_inner, nb_j_inner in T.grid(2, 2):
- for i_inner_init in range(16):
- cse_var_1: T.int32 = i_outer_inner * 512 + i_inner_init * 32 + nb_j_inner * 16
- compute_2[cse_var_1] = T.float32(0)
- compute_2[cse_var_1 + 1] = T.float32(0)
- compute_2[cse_var_1 + 2] = T.float32(0)
- compute_2[cse_var_1 + 3] = T.float32(0)
- compute_2[cse_var_1 + 4] = T.float32(0)
- compute_2[cse_var_1 + 5] = T.float32(0)
- compute_2[cse_var_1 + 6] = T.float32(0)
- compute_2[cse_var_1 + 7] = T.float32(0)
- compute_2[cse_var_1 + 8] = T.float32(0)
- compute_2[cse_var_1 + 9] = T.float32(0)
- compute_2[cse_var_1 + 10] = T.float32(0)
- compute_2[cse_var_1 + 11] = T.float32(0)
- compute_2[cse_var_1 + 12] = T.float32(0)
- compute_2[cse_var_1 + 13] = T.float32(0)
- compute_2[cse_var_1 + 14] = T.float32(0)
- compute_2[cse_var_1 + 15] = T.float32(0)
- for elem_idx, i_inner in T.grid(T.let(cse_var_2, i0_outer_i1_outer_fused % 16 * 2 + nb_j_inner, placeholder_5[cse_var_2 + 1] - placeholder_5[cse_var_2]), 16):
- cse_var_2 = T.var("int32")
+ for i0_outer_i1_outer_fused in T.parallel(512):
+ compute_1 = T.allocate([128], "float32", "global")
+ compute_2 = T.buffer_decl((128,), data=compute_1)
+ for i_outer_inner in range(2):
+ for i_inner_init, j_init in T.grid(4, 16):
+ compute_2[i_outer_inner * 64 + i_inner_init * 16 + j_init] = T.float32(0)
+ for elem_idx, i_inner, j in T.grid(T.let(cse_var_1, i0_outer_i1_outer_fused % 32, placeholder_5[cse_var_1 + 1] - placeholder_5[cse_var_1]), 4, 16):
+ cse_var_1 = T.var("int32")
placeholder_5 = T.buffer_decl((33,), "int32", data=placeholder_3.data)
- cse_var_21: T.int32 = elem_idx * 16
- cse_var_20: T.int32 = i0_outer_i1_outer_fused % 16 * 2 + nb_j_inner
- cse_var_19: T.int32 = i_outer_inner * 512 + i_inner * 32 + nb_j_inner * 16
- cse_var_18: T.int32 = i0_outer_i1_outer_fused // 16 * 8192 + i_outer_inner * 4096 + i_inner * 256
- cse_var_17: T.int32 = cse_var_19 + 9
- cse_var_16: T.int32 = cse_var_19 + 8
- cse_var_15: T.int32 = cse_var_19 + 7
- cse_var_14: T.int32 = cse_var_19 + 6
- cse_var_13: T.int32 = cse_var_19 + 5
- cse_var_12: T.int32 = cse_var_19 + 4
- cse_var_11: T.int32 = cse_var_19 + 3
- cse_var_10: T.int32 = cse_var_19 + 2
- cse_var_9: T.int32 = cse_var_19 + 15
- cse_var_8: T.int32 = cse_var_19 + 14
- cse_var_7: T.int32 = cse_var_19 + 13
- cse_var_6: T.int32 = cse_var_19 + 12
- cse_var_5: T.int32 = cse_var_19 + 11
- cse_var_4: T.int32 = cse_var_19 + 10
- cse_var_3: T.int32 = cse_var_19 + 1
- placeholder_6 = T.buffer_decl((78656,), data=placeholder_1.data)
- placeholder_7 = T.buffer_decl((32768,), data=placeholder.data)
- placeholder_8 = T.buffer_decl((4916,), "int32", data=placeholder_2.data)
- compute_2[cse_var_19] = compute_2[cse_var_19] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_3] = compute_2[cse_var_3] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 1] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_10] = compute_2[cse_var_10] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 2] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_11] = compute_2[cse_var_11] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 3] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_12] = compute_2[cse_var_12] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 4] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_13] = compute_2[cse_var_13] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 5] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_14] = compute_2[cse_var_14] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 6] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_15] = compute_2[cse_var_15] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 7] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_16] = compute_2[cse_var_16] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 8] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_17] = compute_2[cse_var_17] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 9] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_4] = compute_2[cse_var_4] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 10] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_5] = compute_2[cse_var_5] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 11] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_6] = compute_2[cse_var_6] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 12] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_7] = compute_2[cse_var_7] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 13] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_8] = compute_2[cse_var_8] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 14] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- compute_2[cse_var_9] = compute_2[cse_var_9] + placeholder_6[placeholder_5[cse_var_20] * 16 + cse_var_21 + 15] * T.max(placeholder_7[cse_var_18 + placeholder_8[placeholder_5[cse_var_20] + elem_idx]], T.float32(0))
- for i0_inner, i1_inner in T.grid(32, 32):
- cse_var_22: T.int32 = i0_outer_i1_outer_fused // 16 * 16384 + i0_inner * 512 + i0_outer_i1_outer_fused % 16 * 32 + i1_inner
+ cse_var_2: T.int32 = i0_outer_i1_outer_fused % 32
+ if T.likely(elem_idx < placeholder_5[cse_var_2 + 1] - placeholder_5[cse_var_2]):
+ placeholder_6 = T.buffer_decl((78656,), data=placeholder_1.data)
+ placeholder_7 = T.buffer_decl((32768,), data=placeholder.data)
+ placeholder_8 = T.buffer_decl((4916,), "int32", data=placeholder_2.data)
+ cse_var_3: T.int32 = i_outer_inner * 64 + i_inner * 16 + j
+ compute_2[cse_var_3] = compute_2[cse_var_3] + placeholder_6[placeholder_5[cse_var_2] * 16 + elem_idx * 16 + j] * T.max(placeholder_7[i0_outer_i1_outer_fused // 32 * 2048 + i_outer_inner * 1024 + i_inner * 256 + placeholder_8[placeholder_5[cse_var_2] + elem_idx]], T.float32(0))
+ for i0_inner in range(8):
+ cse_var_4: T.int32 = i0_outer_i1_outer_fused // 32 * 4096 + i0_inner * 512 + i0_outer_i1_outer_fused % 32 * 16
compute_3 = T.buffer_decl((65536,), data=compute.data)
placeholder_5 = T.buffer_decl((65536,), data=placeholder_4.data)
- compute_3[cse_var_22] = T.max(compute_2[i0_inner * 32 + i1_inner] + placeholder_5[cse_var_22], T.float32(0))
+ compute_3[cse_var_4:cse_var_4 + 16] = T.max(compute_2[i0_inner * 16:i0_inner * 16 + 16] + placeholder_5[cse_var_4:cse_var_4 + 16], T.Broadcast(T.float32(0), 16))
</pre></div>
</div>
</div>
@@ -730,7 +683,7 @@ class Module:
<span class="p">)</span>
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time of this operator: 1.732 ms
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Execution time of this operator: 1.362 ms
</pre></div>
</div>
<div class="admonition note">
diff --git a/docs/how_to/tune_with_autotvm/sg_execution_times.html b/docs/how_to/tune_with_autotvm/sg_execution_times.html
index f2dd1a9c32..73f2f1db06 100644
--- a/docs/how_to/tune_with_autotvm/sg_execution_times.html
+++ b/docs/how_to/tune_with_autotvm/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-tune-with-autotvm-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>00:49.910</strong> total execution time for <strong>how_to_tune_with_autotvm</strong> files:</p>
+<p><strong>00:32.704</strong> total execution time for <strong>how_to_tune_with_autotvm</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 84%" />
@@ -349,11 +349,11 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_conv2d_cuda.html#sphx-glr-how-to-tune-with-autotvm-tune-conv2d-cuda-py"><span class="std std-ref">Tuning High Performance Convolution on NVIDIA GPUs</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_conv2d_cuda.py</span></code>)</p></td>
-<td><p>00:49.876</p></td>
+<td><p>00:32.668</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tune_relay_x86.html#sphx-glr-how-to-tune-with-autotvm-tune-relay-x86-py"><span class="std std-ref">Auto-tuning a Convolutional Network for x86 CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_relay_x86.py</span></code>)</p></td>
-<td><p>00:00.020</p></td>
+<td><p>00:00.023</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_relay_cuda.html#sphx-glr-how-to-tune-with-autotvm-tune-relay-cuda-py"><span class="std std-ref">Auto-tuning a Convolutional Network for NVIDIA GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_relay_cuda.py</span></code>)</p></td>
@@ -361,7 +361,7 @@
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tune_relay_arm.html#sphx-glr-how-to-tune-with-autotvm-tune-relay-arm-py"><span class="std std-ref">Auto-tuning a Convolutional Network for ARM CPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_relay_arm.py</span></code>)</p></td>
-<td><p>00:00.005</p></td>
+<td><p>00:00.004</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="tune_relay_mobile_gpu.html#sphx-glr-how-to-tune-with-autotvm-tune-relay-mobile-gpu-py"><span class="std std-ref">Auto-tuning a Convolutional Network for Mobile GPU</span></a> (<code class="docutils literal notranslate"><span class="pre">tune_relay_mobile_gpu.py</span></code>)</p></td>
diff --git a/docs/how_to/tune_with_autotvm/tune_conv2d_cuda.html b/docs/how_to/tune_with_autotvm/tune_conv2d_cuda.html
index 31a6426aeb..2a74d6e378 100644
--- a/docs/how_to/tune_with_autotvm/tune_conv2d_cuda.html
+++ b/docs/how_to/tune_with_autotvm/tune_conv2d_cuda.html
@@ -568,12 +568,7 @@ for this template</p>
waiting for device...
device available
Get devices for measurement successfully!
-No: 1 GFLOPS: 7.37/7.37 result: MeasureResult(costs=(0.0314321365,), error_no=MeasureErrorNo.NO_ERROR, all_cost=7.174636363983154, timestamp=1674259112.8968542) [('tile_f', [-1, 1, 8, 8]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,1826814
-No: 2 GFLOPS: 39.80/39.80 result: MeasureResult(costs=(0.005816791055555556,), error_no=MeasureErrorNo.NO_ERROR, all_cost=4.863697290420532, timestamp=1674259114.688778) [('tile_f', [-1, 32, 16, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,8944359
-No: 3 GFLOPS: 4.60/39.80 result: MeasureResult(costs=(0.05034223175,), error_no=MeasureErrorNo.NO_ERROR, all_cost=10.43713903427124, timestamp=1674259116.8111053) [('tile_f', [-1, 16, 1, 4]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 32]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7500784
-No: 4 GFLOPS: 45.33/45.33 result: MeasureResult(costs=(0.0051065872,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.4066011905670166, timestamp=1674259117.5673692) [('tile_f', [-1, 1, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9972820
-No: 5 GFLOPS: 35.24/45.33 result: MeasureResult(costs=(0.0065700821874999996,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.415250539779663, timestamp=1674259119.1511686) [('tile_f', [-1, 4, 4, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 1]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1164481
-No: 6 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+No: 1 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -695,8 +690,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 128, 1, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 64, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9350502
-No: 7 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 2, 16]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,276272
+No: 2 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -818,8 +813,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 1, 4]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 8, 32]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,732264
-No: 8 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 32, 2, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6215289
+No: 3 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -941,27 +936,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 256]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2312972
-No: 9 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
- File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 142, in build
- res = future.result()
- File "/usr/lib/python3.7/concurrent/futures/_base.py", line 435, in result
- return self.__get_result()
- File "/usr/lib/python3.7/concurrent/futures/_base.py", line 384, in __get_result
- raise self._exception
- File "/usr/lib/python3.7/concurrent/futures/thread.py", line 57, in run
- result = self.fn(*self.args, **self.kwargs)
- File "/workspace/python/tvm/contrib/popen_pool.py", line 432, in <lambda>
- worker = lambda *args: self._worker_run(*args)
- File "/workspace/python/tvm/contrib/popen_pool.py", line 401, in _worker_run
- return proc.recv()
- File "/workspace/python/tvm/contrib/popen_pool.py", line 309, in recv
- raise TimeoutError()
-TimeoutError
-
- [('tile_f', [-1, 8, 4, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5661335
-No: 10 GFLOPS: 12.84/45.33 result: MeasureResult(costs=(0.018029925,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3179340362548828, timestamp=1674259131.2934096) [('tile_f', [-1, 1, 4, 8]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3305429
-No: 11 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 8, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 16]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7689467
+No: 4 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1083,8 +1059,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 32]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6286468
-No: 12 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 32, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 512, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6808873
+No: 5 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1206,8 +1182,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 512, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7979629
-No: 13 GFLOPS: 0.00/45.33 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 8, 8]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 32, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4568457
+No: 6 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1329,9 +1305,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 1, 512]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,10452874
-No: 14 GFLOPS: 195.22/195.22 result: MeasureResult(costs=(0.001185834181818182,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.4841997623443604, timestamp=1674259134.1751637) [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 1]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,198706
-No: 15 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 128, 1, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 32]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5963207
+No: 7 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1453,9 +1428,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 16, 8]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7375440
-No: 16 GFLOPS: 63.25/195.22 result: MeasureResult(costs=(0.0036600480714285713,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3547515869140625, timestamp=1674259134.9343917) [('tile_f', [-1, 2, 16, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5460119
-No: 17 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4614419
+No: 8 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1577,8 +1551,8 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 32, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 8, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6823832
-No: 18 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 4, 8]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 16]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6714109
+No: 9 GFLOPS: 0.00/0.00 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1700,9 +1674,9 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 64, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,83868
-No: 19 GFLOPS: 7.73/195.22 result: MeasureResult(costs=(0.029934080000000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.8496928215026855, timestamp=1674259136.9989753) [('tile_f', [-1, 16, 2, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5328248
-No: 20 GFLOPS: 0.00/195.22 result: Traceback (most recent call last):
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 128, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6258341
+No: 10 GFLOPS: 29.67/29.67 result: MeasureResult(costs=(0.007802013928571429,), error_no=MeasureErrorNo.NO_ERROR, all_cost=5.3617002964019775, timestamp=1674265444.7841802) [('tile_f', [-1, 1, 4, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4741854
+No: 11 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
@@ -1824,7 +1798,992 @@ Traceback (most recent call last):
File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
raise InstantiationError("Skipped because of invalid gpu kernel")
-tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 2, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 8, 16]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9229508
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 2, 128]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4631266
+No: 12 GFLOPS: 5.74/29.67 result: MeasureResult(costs=(0.0403490255,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.486266613006592, timestamp=1674265445.7363422) [('tile_f', [-1, 4, 2, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 32]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,3819872
+No: 13 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 8, 1, 64]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7131503
+No: 14 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 32, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 1, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2360641
+No: 15 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 2, 4, 64]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4543428
+No: 16 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 32, 16]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 8, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7077804
+No: 17 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 16, 8, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 2, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3170723
+No: 18 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 1, 32, 4]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 64, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6447230
+No: 19 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 32, 8, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9260712
+No: 20 GFLOPS: 0.00/29.67 result: Traceback (most recent call last):
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 592, in __call__
+ func, arg_info = _build_func_common(measure_input, self.runtime, **kwargs)
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 544, in _build_func_common
+ func = build(s, args, target_host=task.target_host, runtime=runtime)
+ File "/workspace/python/tvm/driver/build_module.py", line 227, in build
+ input_mod = lower(inputs, args, name=name, binds=binds)
+ File "/workspace/python/tvm/driver/build_module.py", line 134, in lower
+ return ffi.lower_schedule(inp, args, name, binds, simple_mode)
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 331, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 276, in tvm._ffi._cy3.core.FuncCall
+ File "tvm/_ffi/_cython/./base.pxi", line 181, in tvm._ffi._cy3.core.CHECK_CALL
+tvm._ffi.base.TVMError: Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel
+
+Traceback (most recent call last):
+ 24: TVMFuncCall
+ at ../src/runtime/c_runtime_api.cc:477
+ 23: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 22: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 21: operator()
+ at ../include/tvm/runtime/packed_func.h:1730
+ 20: unpack_call<tvm::IRModule, 5, tvm::<lambda(tvm::te::Schedule, const tvm::runtime::Array<tvm::runtime::ObjectRef>&, const tvm::runtime::String&, const tvm::runtime::Map<tvm::te::Tensor, tvm::tir::Buffer>&, bool)> >
+ at ../include/tvm/runtime/packed_func.h:1670
+ 19: run<>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 18: run<tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 17: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 16: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 15: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1630
+ 14: run<tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_, tvm::runtime::TVMMovableArgValueWithContext_>
+ at ../include/tvm/runtime/packed_func.h:1645
+ 13: operator()
+ at ../src/driver/driver_api.cc:395
+ 12: tvm::LowerSchedule(tvm::te::Schedule, tvm::runtime::Array<tvm::runtime::ObjectRef, void> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::unordered_map<tvm::te::Tensor, tvm::tir::Buffer, std::hash<tvm::te::Tensor>, std::equal_to<tvm::te::Tensor>, std::allocator<std::pair<tvm::te::Tensor const, tvm::tir::Buffer> > > const&, tvm::GlobalVarSupply, bool)
+ at ../src/driver/driver_api.cc:381
+ 11: tvm::LowerWithPassList(tvm::IRModule, tvm::runtime::Array<tvm::transform::Pass, void>)
+ at ../src/driver/driver_api.cc:276
+ 10: tvm::transform::Pass::operator()(tvm::IRModule) const
+ at ../src/ir/transform.cc:258
+ 9: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 8: tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:451
+ 7: tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/ir/transform.cc:274
+ 6: tvm::tir::transform::PrimFuncPassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
+ at ../src/tir/ir/transform.cc:100
+ 5: tvm::runtime::TypedPackedFunc<tvm::tir::PrimFunc (tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext)>::operator()(tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext) const
+ at ../include/tvm/runtime/packed_func.h:1749
+ 4: tvm::tir::PrimFunc tvm::runtime::detail::typed_packed_call_dispatcher<tvm::tir::PrimFunc>::run<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::runtime::PackedFunc const&, tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&)
+ at ../include/tvm/runtime/packed_func.h:1693
+ 3: tvm::runtime::TVMRetValue tvm::runtime::PackedFunc::operator()<tvm::tir::PrimFunc, tvm::IRModule, tvm::transform::PassContext>(tvm::tir::PrimFunc&&, tvm::IRModule&&, tvm::transform::PassContext&&) const
+ at ../include/tvm/runtime/packed_func.h:1617
+ 2: tvm::runtime::PackedFuncObj::CallPacked(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const
+ at ../include/tvm/runtime/packed_func.h:1217
+ 1: Call
+ at ../include/tvm/runtime/packed_func.h:1213
+ 0: operator()
+ at ../src/runtime/c_runtime_api.cc:534
+ File "tvm/_ffi/_cython/./packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
+ File "/workspace/python/tvm/autotvm/measure/measure_methods.py", line 875, in verify_pass
+ raise InstantiationError("Skipped because of invalid gpu kernel")
+tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel [('tile_f', [-1, 4, 2, 8]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,5143525
</pre></div>
</div>
<p>Finally we can inspect the best config from log file, check correctness,
@@ -1863,9 +2822,9 @@ and measure running time.</p>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Finish loading 20 records
Best config:
-[('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 1]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,198706
+[('tile_f', [-1, 1, 4, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 8]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4741854
Finish loading 20 records
-Time cost of this operator: 0.001517
+Time cost of this operator: 0.007964
</pre></div>
</div>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-tune-with-autotvm-tune-conv2d-cuda-py">
diff --git a/docs/how_to/work_with_microtvm/micro_autotune.html b/docs/how_to/work_with_microtvm/micro_autotune.html
index bc35f6c2e7..ac54f38ae5 100644
--- a/docs/how_to/work_with_microtvm/micro_autotune.html
+++ b/docs/how_to/work_with_microtvm/micro_autotune.html
@@ -647,10 +647,10 @@ the tuned operator.</p>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>########## Build without Autotuning ##########
Node Name Ops Time(us) Time(%) Shape Inputs Outputs Measurements(us)
--------- --- -------- ------- ----- ------ ------- ----------------
-tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 312.2 98.726 (1, 2, 10, 10, 3) 2 1 [312.2]
-tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 3.043 0.962 (1, 6, 10, 10) 1 1 [3.043]
-tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.985 0.311 (1, 1, 10, 10, 3) 1 1 [0.985]
-Total_time - 316.228 - - - - -
+tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 311.5 98.724 (1, 2, 10, 10, 3) 2 1 [311.5]
+tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 3.026 0.959 (1, 6, 10, 10) 1 1 [3.026]
+tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 1.001 0.317 (1, 1, 10, 10, 3) 1 1 [1.001]
+Total_time - 315.527 - - - - -
</pre></div>
</div>
</div>
@@ -702,10 +702,10 @@ Total_time -
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>########## Build with Autotuning ##########
Node Name Ops Time(us) Time(%) Shape Inputs Outputs Measurements(us)
--------- --- -------- ------- ----- ------ ------- ----------------
-tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 139.3 98.131 (1, 6, 10, 10, 1) 2 1 [139.3]
-tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 1.777 1.252 (1, 6, 10, 10) 1 1 [1.777]
-tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.877 0.618 (1, 3, 10, 10, 1) 1 1 [0.877]
-Total_time - 141.954 - - - - -
+tvmgen_default_fused_nn_contrib_conv2d_NCHWc tvmgen_default_fused_nn_contrib_conv2d_NCHWc 102.6 97.448 (1, 6, 10, 10, 1) 2 1 [102.6]
+tvmgen_default_fused_layout_transform_1 tvmgen_default_fused_layout_transform_1 1.829 1.737 (1, 6, 10, 10) 1 1 [1.829]
+tvmgen_default_fused_layout_transform tvmgen_default_fused_layout_transform 0.858 0.815 (1, 3, 10, 10, 1) 1 1 [0.858]
+Total_time - 105.287 - - - - -
</pre></div>
</div>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-work-with-microtvm-micro-autotune-py">
diff --git a/docs/how_to/work_with_microtvm/micro_pytorch.html b/docs/how_to/work_with_microtvm/micro_pytorch.html
index 04c5e58b67..e9a91b35af 100644
--- a/docs/how_to/work_with_microtvm/micro_pytorch.html
+++ b/docs/how_to/work_with_microtvm/micro_pytorch.html
@@ -454,7 +454,7 @@ download a cat image and preprocess it to use as the model input.</p>
Downloading: "https://download.pytorch.org/models/quantized/mobilenet_v2_qnnpack_37f702c5.pth" to /workspace/.cache/torch/hub/checkpoints/mobilenet_v2_qnnpack_37f702c5.pth
0%| | 0.00/3.42M [00:00<?, ?B/s]
-100%|##########| 3.42M/3.42M [00:00<00:00, 81.6MB/s]
+100%|##########| 3.42M/3.42M [00:00<00:00, 57.0MB/s]
/workspace/python/tvm/relay/frontend/pytorch_utils.py:47: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
return LooseVersion(torch_ver) > ver
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/setuptools/_distutils/version.py:346: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
@@ -578,7 +578,7 @@ via the host <cite>main.cc`</cite> or if a Zephyr emulated board is selected as
Torch top-1 id: 282, class name: tiger cat
</pre></div>
</div>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 13.811 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes 13.423 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-work-with-microtvm-micro-pytorch-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/12b9ecc04c41abaa12022061771821d1/micro_pytorch.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">micro_pytorch.py</span></code></a></p>
diff --git a/docs/how_to/work_with_microtvm/micro_train.html b/docs/how_to/work_with_microtvm/micro_train.html
index 2af024a1e7..8ec057aabc 100644
--- a/docs/how_to/work_with_microtvm/micro_train.html
+++ b/docs/how_to/work_with_microtvm/micro_train.html
@@ -524,7 +524,7 @@ take about <strong>2 minutes</strong> to download the Stanford Cars, while COCO
<a href="https://docs.python.org/3/library/shutil.html#shutil.move" title="shutil.move" class="sphx-glr-backref-module-shutil sphx-glr-backref-type-py-function"><span class="n">shutil</span><span class="o">.</span><span class="n">move</span></a><span class="p">(</span><span class="sa">f</span><span class="s2">"</span><span class="si">{</span><a href="https://docs.python.org/3/library/stdtypes.html#str" title="builtins.str" class="sphx-glr-backref-module-builtins sphx-glr-backref-typ [...]
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>'/tmp/tmpciq06jqh/images/random'
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>'/tmp/tmp8v5ftiwe/images/random'
</pre></div>
</div>
</div>
@@ -584,8 +584,8 @@ objects to other stuff? We can display some examples from our datasets using <co
<span class="n">plt</span><span class="o">.</span><span class="n">axis</span><span class="p">(</span><span class="s2">"off"</span><span class="p">)</span>
</pre></div>
</div>
-<img src="../../_images/sphx_glr_micro_train_001.png" srcset="../../_images/sphx_glr_micro_train_001.png" alt="[1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0], [1.0, 0.0]" class = "sphx-glr-single-img"/><div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>/tmp/tmpciq06jqh/images/target contains 8144 images
-/tmp/tmpciq06jqh/images/random contains 5000 images
+<img src="../../_images/sphx_glr_micro_train_001.png" srcset="../../_images/sphx_glr_micro_train_001.png" alt="[0.0, 1.0], [1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0]" class = "sphx-glr-single-img"/><div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>/tmp/tmp8v5ftiwe/images/target contains 8144 images
+/tmp/tmp8v5ftiwe/images/random contains 5000 images
</pre></div>
</div>
</div>
@@ -697,13 +697,13 @@ the time on our validation set).</p>
</pre></div>
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Epoch 1/3
-328/328 - 47s - loss: 0.2233 - accuracy: 0.9254 - val_loss: 0.1195 - val_accuracy: 0.9607 - 47s/epoch - 144ms/step
+328/328 - 47s - loss: 0.2200 - accuracy: 0.9250 - val_loss: 0.1157 - val_accuracy: 0.9600 - 47s/epoch - 143ms/step
Epoch 2/3
-328/328 - 44s - loss: 0.1003 - accuracy: 0.9639 - val_loss: 0.1247 - val_accuracy: 0.9592 - 44s/epoch - 133ms/step
+328/328 - 44s - loss: 0.0974 - accuracy: 0.9643 - val_loss: 0.1420 - val_accuracy: 0.9577 - 44s/epoch - 133ms/step
Epoch 3/3
-328/328 - 44s - loss: 0.0676 - accuracy: 0.9764 - val_loss: 0.1159 - val_accuracy: 0.9637 - 44s/epoch - 133ms/step
+328/328 - 43s - loss: 0.0648 - accuracy: 0.9753 - val_loss: 0.1082 - val_accuracy: 0.9656 - 43s/epoch - 132ms/step
-<keras.callbacks.History object at 0x7f0a48b6c0d0>
+<keras.callbacks.History object at 0x7fb5a9e8b910>
</pre></div>
</div>
</div>
@@ -963,7 +963,7 @@ as intended.</p>
<p>From here, we could modify the model to read live images from the camera - we have another
Arduino tutorial for how to do that <a class="reference external" href="https://github.com/guberti/tvm-arduino-demos/tree/master/examples/person_detection">on GitHub</a>. Alternatively, we could also
<a class="reference external" href="https://tvm.apache.org/docs/how_to/work_with_microtvm/micro_autotune.html">use TVM’s autotuning capabilities</a> to dramatically improve the model’s performance.</p>
-<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 4 minutes 10.207 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 4 minutes 20.715 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-how-to-work-with-microtvm-micro-train-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../../_downloads/b52cec46baf4f78d6bcd94cbe269c8a6/micro_train.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">micro_train.py</span></code></a></p>
diff --git a/docs/how_to/work_with_microtvm/sg_execution_times.html b/docs/how_to/work_with_microtvm/sg_execution_times.html
index 841851aaef..b0917d2fa1 100644
--- a/docs/how_to/work_with_microtvm/sg_execution_times.html
+++ b/docs/how_to/work_with_microtvm/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-work-with-microtvm-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>06:34.453</strong> total execution time for <strong>how_to_work_with_microtvm</strong> files:</p>
+<p><strong>06:40.241</strong> total execution time for <strong>how_to_work_with_microtvm</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 83%" />
@@ -349,23 +349,23 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="micro_train.html#sphx-glr-how-to-work-with-microtvm-micro-train-py"><span class="std std-ref">Training Vision Models for microTVM on Arduino</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_train.py</span></code>)</p></td>
-<td><p>04:10.207</p></td>
+<td><p>04:20.715</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="micro_pytorch.html#sphx-glr-how-to-work-with-microtvm-micro-pytorch-py"><span class="std std-ref">microTVM PyTorch Tutorial</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_pytorch.py</span></code>)</p></td>
-<td><p>01:13.811</p></td>
+<td><p>01:13.423</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="micro_autotune.html#sphx-glr-how-to-work-with-microtvm-micro-autotune-py"><span class="std std-ref">Autotuning with microTVM</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_autotune.py</span></code>)</p></td>
-<td><p>00:56.653</p></td>
+<td><p>00:52.960</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="micro_aot.html#sphx-glr-how-to-work-with-microtvm-micro-aot-py"><span class="std std-ref">microTVM Host-Driven AoT</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_aot.py</span></code>)</p></td>
-<td><p>00:09.495</p></td>
+<td><p>00:08.927</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="micro_tflite.html#sphx-glr-how-to-work-with-microtvm-micro-tflite-py"><span class="std std-ref">microTVM with TFLite Models</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_tflite.py</span></code>)</p></td>
-<td><p>00:04.288</p></td>
+<td><p>00:04.215</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="micro_ethosu.html#sphx-glr-how-to-work-with-microtvm-micro-ethosu-py"><span class="std std-ref">Running TVM on bare metal Arm(R) Cortex(R)-M55 CPU and Ethos(TM)-U55 NPU with CMSIS-NN</span></a> (<code class="docutils literal notranslate"><span class="pre">micro_ethosu.py</span></code>)</p></td>
diff --git a/docs/how_to/work_with_relay/sg_execution_times.html b/docs/how_to/work_with_relay/sg_execution_times.html
index 3eef84e74a..a2cffd46dd 100644
--- a/docs/how_to/work_with_relay/sg_execution_times.html
+++ b/docs/how_to/work_with_relay/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-work-with-relay-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>00:46.941</strong> total execution time for <strong>how_to_work_with_relay</strong> files:</p>
+<p><strong>00:44.962</strong> total execution time for <strong>how_to_work_with_relay</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 84%" />
@@ -349,19 +349,19 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="using_pipeline_executor.html#sphx-glr-how-to-work-with-relay-using-pipeline-executor-py"><span class="std std-ref">Using Pipeline Executor in Relay</span></a> (<code class="docutils literal notranslate"><span class="pre">using_pipeline_executor.py</span></code>)</p></td>
-<td><p>00:34.745</p></td>
+<td><p>00:32.792</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="using_external_lib.html#sphx-glr-how-to-work-with-relay-using-external-lib-py"><span class="std std-ref">Using External Libraries in Relay</span></a> (<code class="docutils literal notranslate"><span class="pre">using_external_lib.py</span></code>)</p></td>
-<td><p>00:10.468</p></td>
+<td><p>00:10.380</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="build_gcn.html#sphx-glr-how-to-work-with-relay-build-gcn-py"><span class="std std-ref">Building a Graph Convolutional Network</span></a> (<code class="docutils literal notranslate"><span class="pre">build_gcn.py</span></code>)</p></td>
-<td><p>00:01.721</p></td>
+<td><p>00:01.782</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="using_relay_viz.html#sphx-glr-how-to-work-with-relay-using-relay-viz-py"><span class="std std-ref">Use Relay Visualizer to Visualize Relay</span></a> (<code class="docutils literal notranslate"><span class="pre">using_relay_viz.py</span></code>)</p></td>
-<td><p>00:00.006</p></td>
+<td><p>00:00.007</p></td>
<td><p>0.0 MB</p></td>
</tr>
</tbody>
diff --git a/docs/how_to/work_with_schedules/intrin_math.html b/docs/how_to/work_with_schedules/intrin_math.html
index ba32c2ade2..ddebd35370 100644
--- a/docs/how_to/work_with_schedules/intrin_math.html
+++ b/docs/how_to/work_with_schedules/intrin_math.html
@@ -535,7 +535,7 @@ The following example customizes CUDA lowering rule for <code class="code docuti
<a href="../../reference/api/python/ir.html#tvm.ir.register_intrin_lowering" title="tvm.ir.register_intrin_lowering" class="sphx-glr-backref-module-tvm-ir sphx-glr-backref-type-py-function"><span class="n">register_intrin_lowering</span></a><span class="p">(</span><span class="s2">"tir.exp"</span><span class="p">,</span> <span class="n">target</span><span class="o">=</span><span class="s2">"cuda"</span><span class="p">,</span> <span class="n">f</span><span class="o">= [...]
</pre></div>
</div>
-<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span><function my_cuda_math_rule at 0x7f0adca84560>
+<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span><function my_cuda_math_rule at 0x7fb5a8d075f0>
</pre></div>
</div>
<p>Register the rule to TVM with override option to override existing rule.
diff --git a/docs/how_to/work_with_schedules/sg_execution_times.html b/docs/how_to/work_with_schedules/sg_execution_times.html
index 663bd1b9c3..7315c114d0 100644
--- a/docs/how_to/work_with_schedules/sg_execution_times.html
+++ b/docs/how_to/work_with_schedules/sg_execution_times.html
@@ -340,7 +340,7 @@
<div class="section" id="computation-times">
<span id="sphx-glr-how-to-work-with-schedules-sg-execution-times"></span><h1>Computation times<a class="headerlink" href="#computation-times" title="Permalink to this headline">¶</a></h1>
-<p><strong>00:06.377</strong> total execution time for <strong>how_to_work_with_schedules</strong> files:</p>
+<p><strong>00:07.164</strong> total execution time for <strong>how_to_work_with_schedules</strong> files:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 83%" />
@@ -349,35 +349,35 @@
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="intrin_math.html#sphx-glr-how-to-work-with-schedules-intrin-math-py"><span class="std std-ref">Intrinsics and Math Functions</span></a> (<code class="docutils literal notranslate"><span class="pre">intrin_math.py</span></code>)</p></td>
-<td><p>00:03.680</p></td>
+<td><p>00:04.655</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tensorize.html#sphx-glr-how-to-work-with-schedules-tensorize-py"><span class="std std-ref">Use Tensorize to Leverage Hardware Intrinsics</span></a> (<code class="docutils literal notranslate"><span class="pre">tensorize.py</span></code>)</p></td>
-<td><p>00:01.202</p></td>
+<td><p>00:01.141</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="reduction.html#sphx-glr-how-to-work-with-schedules-reduction-py"><span class="std std-ref">Reduction</span></a> (<code class="docutils literal notranslate"><span class="pre">reduction.py</span></code>)</p></td>
-<td><p>00:00.641</p></td>
+<td><p>00:00.583</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="scan.html#sphx-glr-how-to-work-with-schedules-scan-py"><span class="std std-ref">Scan and Recurrent Kernel</span></a> (<code class="docutils literal notranslate"><span class="pre">scan.py</span></code>)</p></td>
-<td><p>00:00.620</p></td>
+<td><p>00:00.562</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="extern_op.html#sphx-glr-how-to-work-with-schedules-extern-op-py"><span class="std std-ref">External Tensor Functions</span></a> (<code class="docutils literal notranslate"><span class="pre">extern_op.py</span></code>)</p></td>
-<td><p>00:00.121</p></td>
+<td><p>00:00.117</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="schedule_primitives.html#sphx-glr-how-to-work-with-schedules-schedule-primitives-py"><span class="std std-ref">Schedule Primitives in TVM</span></a> (<code class="docutils literal notranslate"><span class="pre">schedule_primitives.py</span></code>)</p></td>
-<td><p>00:00.056</p></td>
+<td><p>00:00.050</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-odd"><td><p><a class="reference internal" href="tedd.html#sphx-glr-how-to-work-with-schedules-tedd-py"><span class="std std-ref">Use Tensor Expression Debug Display (TEDD) for Visualization</span></a> (<code class="docutils literal notranslate"><span class="pre">tedd.py</span></code>)</p></td>
-<td><p>00:00.032</p></td>
+<td><p>00:00.031</p></td>
<td><p>0.0 MB</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="tuple_inputs.html#sphx-glr-how-to-work-with-schedules-tuple-inputs-py"><span class="std std-ref">Compute and Reduce with Tuple Inputs</span></a> (<code class="docutils literal notranslate"><span class="pre">tuple_inputs.py</span></code>)</p></td>
-<td><p>00:00.025</p></td>
+<td><p>00:00.024</p></td>
<td><p>0.0 MB</p></td>
</tr>
</tbody>
diff --git a/docs/how_to/work_with_schedules/tensorize.html b/docs/how_to/work_with_schedules/tensorize.html
index b716144b88..31bba8b84f 100644
--- a/docs/how_to/work_with_schedules/tensorize.html
+++ b/docs/how_to/work_with_schedules/tensorize.html
@@ -574,7 +574,7 @@ class Module:
def main(A: T.Buffer((1024, 64), "float32"), B: T.Buffer((512, 64), "float32"), C: T.Buffer((1024, 512), "float32")):
T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
i = T.var("int32")
- T.attr(T.iter_var(i, None, "DataPar", ""), "pragma_import_llvm", "; ModuleID = '/tmp/tmpns4jb9wl/input0.cc'\nsource_filename = \"/tmp/tmpns4jb9wl/input0.cc\"\ntarget datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-pc-linux-gnu\"\n\n; Function Attrs: noinline nounwind optnone uwtable\ndefine dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {\n %7 = alloca [...]
+ T.attr(T.iter_var(i, None, "DataPar", ""), "pragma_import_llvm", "; ModuleID = '/tmp/tmpwrd4vqy_/input0.cc'\nsource_filename = \"/tmp/tmpwrd4vqy_/input0.cc\"\ntarget datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-pc-linux-gnu\"\n\n; Function Attrs: noinline nounwind optnone uwtable\ndefine dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {\n %7 = alloca [...]
for i, j_outer in T.grid(1024, 32):
T.call_extern("int32", "gemv_update", T.tvm_access_ptr(T.type_annotation("float32"), C.data, i * 512 + j_outer * 16, 16, 2), T.tvm_access_ptr(T.type_annotation("float32"), A.data, i * 64, 64, 1), T.tvm_access_ptr(T.type_annotation("float32"), B.data, j_outer * 1024, 1024, 1), 16, 64, 64)
</pre></div>
diff --git a/docs/reference/api/python/auto_scheduler.html b/docs/reference/api/python/auto_scheduler.html
index 46499d7eb1..6a928519e9 100644
--- a/docs/reference/api/python/auto_scheduler.html
+++ b/docs/reference/api/python/auto_scheduler.html
@@ -1615,7 +1615,7 @@ history states as starting point to perform Evolutionary Search).</p></li>
<dl class="py class">
<dt class="sig sig-object py" id="tvm.auto_scheduler.SketchPolicy">
-<em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">tvm.auto_scheduler.</span></span><span class="sig-name descname"><span class="pre">SketchPolicy</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">task</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">program_cost_model</span></span><span class="o"><span class="pre">=</span></span><span class="defau [...]
+<em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">tvm.auto_scheduler.</span></span><span class="sig-name descname"><span class="pre">SketchPolicy</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">task</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">program_cost_model</span></span><span class="o"><span class="pre">=</span></span><span class="defau [...]
<dd><p>The search policy that searches in a hierarchical search space defined by sketches.
The policy randomly samples programs from the space defined by sketches and use evolutionary
search to fine-tune them.</p>
@@ -1899,7 +1899,7 @@ Candidates:
<dl class="py function">
<dt class="sig sig-object py" id="tvm.auto_scheduler.auto_schedule">
-<span class="sig-prename descclassname"><span class="pre">tvm.auto_scheduler.</span></span><span class="sig-name descname"><span class="pre">auto_schedule</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">task</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">search_policy</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em clas [...]
+<span class="sig-prename descclassname"><span class="pre">tvm.auto_scheduler.</span></span><span class="sig-name descname"><span class="pre">auto_schedule</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">task</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">search_policy</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em clas [...]
<dd><p>THIS API IS DEPRECATED.</p>
<p>Run auto scheduling search for a task.</p>
<dl class="field-list simple">
diff --git a/docs/reference/api/typedoc/classes/bytestreamreader.html b/docs/reference/api/typedoc/classes/bytestreamreader.html
index efcfb01492..675b21ebdf 100644
--- a/docs/reference/api/typedoc/classes/bytestreamreader.html
+++ b/docs/reference/api/typedoc/classes/bytestreamreader.html
@@ -119,7 +119,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L43">rpc_server.ts:43</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L43">rpc_server.ts:43</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -141,7 +141,7 @@
<div class="tsd-signature tsd-kind-icon">bytes<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">Uint8Array</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L43">rpc_server.ts:43</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L43">rpc_server.ts:43</a></li>
</ul>
</aside>
</section>
@@ -151,7 +151,7 @@
<div class="tsd-signature tsd-kind-icon">offset<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span><span class="tsd-signature-symbol"> = 0</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L42">rpc_server.ts:42</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L42">rpc_server.ts:42</a></li>
</ul>
</aside>
</section>
@@ -168,7 +168,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L63">rpc_server.ts:63</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L63">rpc_server.ts:63</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">Uint8Array</span></h4>
@@ -185,7 +185,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L49">rpc_server.ts:49</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L49">rpc_server.ts:49</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">number</span></h4>
@@ -202,7 +202,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/rpc_server.ts#L57">rpc_server.ts:57</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/rpc_server.ts#L57">rpc_server.ts:57</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">number</span></h4>
diff --git a/docs/reference/api/typedoc/classes/cachedcallstack.html b/docs/reference/api/typedoc/classes/cachedcallstack.html
index 3e40bc49a9..383d940b01 100644
--- a/docs/reference/api/typedoc/classes/cachedcallstack.html
+++ b/docs/reference/api/typedoc/classes/cachedcallstack.html
@@ -144,7 +144,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L223">memory.ts:223</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L223">memory.ts:223</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -172,7 +172,7 @@
<div class="tsd-signature tsd-kind-icon">temp<wbr>Args<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">Array</span><span class="tsd-signature-symbol"><</span><a href="../interfaces/disposable.html" class="tsd-signature-type">Disposable</a><span class="tsd-signature-symbol">></span><span class="tsd-signature-symbol"> = []</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L208">memory.ts:208</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L208">memory.ts:208</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -194,7 +194,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L312">memory.ts:312</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L312">memory.ts:312</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -226,7 +226,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L284">memory.ts:284</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L284">memory.ts:284</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -262,7 +262,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L388">memory.ts:388</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L388">memory.ts:388</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -300,7 +300,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L376">memory.ts:376</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L376">memory.ts:376</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -340,7 +340,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L267">memory.ts:267</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L267">memory.ts:267</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -373,7 +373,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L243">memory.ts:243</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L243">memory.ts:243</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">void</span></h4>
@@ -390,7 +390,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L321">memory.ts:321</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L321">memory.ts:321</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -422,7 +422,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L252">memory.ts:252</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L252">memory.ts:252</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -444,7 +444,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L359">memory.ts:359</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L359">memory.ts:359</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -470,7 +470,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L342">memory.ts:342</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L342">memory.ts:342</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -496,7 +496,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L350">memory.ts:350</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L350">memory.ts:350</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -522,7 +522,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L326">memory.ts:326</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L326">memory.ts:326</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -548,7 +548,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L363">memory.ts:363</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L363">memory.ts:363</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -574,7 +574,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L346">memory.ts:346</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L346">memory.ts:346</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -600,7 +600,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/memory.ts#L334">memory.ts:334</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/memory.ts#L334">memory.ts:334</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
diff --git a/docs/reference/api/typedoc/classes/dldatatype.html b/docs/reference/api/typedoc/classes/dldatatype.html
index d7869512dd..f9d7ded6bc 100644
--- a/docs/reference/api/typedoc/classes/dldatatype.html
+++ b/docs/reference/api/typedoc/classes/dldatatype.html
@@ -119,7 +119,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L262">runtime.ts:262</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L262">runtime.ts:262</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -147,7 +147,7 @@
<div class="tsd-signature tsd-kind-icon">bits<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L260">runtime.ts:260</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L260">runtime.ts:260</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -162,7 +162,7 @@
<div class="tsd-signature tsd-kind-icon">code<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L258">runtime.ts:258</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L258">runtime.ts:258</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -177,7 +177,7 @@
<div class="tsd-signature tsd-kind-icon">lanes<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L262">runtime.ts:262</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L262">runtime.ts:262</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -199,7 +199,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L279">runtime.ts:279</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L279">runtime.ts:279</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">number</span></h4>
@@ -216,7 +216,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L270">runtime.ts:270</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L270">runtime.ts:270</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">string</span></h4>
diff --git a/docs/reference/api/typedoc/classes/dldevice.html b/docs/reference/api/typedoc/classes/dldevice.html
index 288b52016a..d632623e97 100644
--- a/docs/reference/api/typedoc/classes/dldevice.html
+++ b/docs/reference/api/typedoc/classes/dldevice.html
@@ -118,7 +118,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L202">runtime.ts:202</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L202">runtime.ts:202</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -146,7 +146,7 @@
<div class="tsd-signature tsd-kind-icon">device<wbr>Id<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L200">runtime.ts:200</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L200">runtime.ts:200</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -161,7 +161,7 @@
<div class="tsd-signature tsd-kind-icon">device<wbr>Type<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">number</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L198">runtime.ts:198</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L198">runtime.ts:198</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -183,7 +183,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L223">runtime.ts:223</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L223">runtime.ts:223</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -205,7 +205,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L230">runtime.ts:230</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L230">runtime.ts:230</a></li>
</ul>
</aside>
<h4 class="tsd-returns-title">Returns <span class="tsd-signature-type">string</span></h4>
diff --git a/docs/reference/api/typedoc/classes/environment.html b/docs/reference/api/typedoc/classes/environment.html
index d9b12736c7..d41277b15a 100644
--- a/docs/reference/api/typedoc/classes/environment.html
+++ b/docs/reference/api/typedoc/classes/environment.html
@@ -125,7 +125,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L86">environment.ts:86</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L86">environment.ts:86</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -169,7 +169,7 @@
<aside class="tsd-sources">
<p>Implementation of <a href="../interfaces/libraryprovider.html">LibraryProvider</a>.<a href="../interfaces/libraryprovider.html#imports">imports</a></p>
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L70">environment.ts:70</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L70">environment.ts:70</a></li>
</ul>
</aside>
</section>
@@ -179,7 +179,7 @@
<div class="tsd-signature tsd-kind-icon">logger<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-symbol">(</span>msg<span class="tsd-signature-symbol">: </span><span class="tsd-signature-type">string</span><span class="tsd-signature-symbol">)</span><span class="tsd-signature-symbol"> => </span><span class="tsd-signature-type">void</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L69">environment.ts:69</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L69">environment.ts:69</a></li>
</ul>
</aside>
<div class="tsd-type-declaration">
@@ -210,7 +210,7 @@
<div class="tsd-signature tsd-kind-icon">packedCFunc<wbr>Table<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">Array</span><span class="tsd-signature-symbol"><</span><span class="tsd-signature-type">ctypes.FTVMWasmPackedCFunc</span><span class="tsd-signature-symbol"> | </span><span class="tsd-signature-type">undefined</span><span class="tsd-signature-symbol">></span><span class="tsd-signature-symbol"> = [undefined,]</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L78">environment.ts:78</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L78">environment.ts:78</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -228,7 +228,7 @@
<div class="tsd-signature tsd-kind-icon">packedCFunc<wbr>Table<wbr>Free<wbr>Id<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">Array</span><span class="tsd-signature-symbol"><</span><span class="tsd-signature-type">number</span><span class="tsd-signature-symbol">></span><span class="tsd-signature-symbol"> = []</span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L84">environment.ts:84</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L84">environment.ts:84</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
@@ -250,7 +250,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/environment.ts#L105">environment.ts:105</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/environment.ts#L105">environment.ts:105</a></li>
</ul>
</aside>
<div class="tsd-comment tsd-typography">
diff --git a/docs/reference/api/typedoc/classes/ffilibrary.html b/docs/reference/api/typedoc/classes/ffilibrary.html
index b25b403b2b..c9a4184c1c 100644
--- a/docs/reference/api/typedoc/classes/ffilibrary.html
+++ b/docs/reference/api/typedoc/classes/ffilibrary.html
@@ -131,7 +131,7 @@
<li class="tsd-description">
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L49">runtime.ts:49</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L49">runtime.ts:49</a></li>
</ul>
</aside>
<h4 class="tsd-parameters-title">Parameters</h4>
@@ -156,7 +156,7 @@
<div class="tsd-signature tsd-kind-icon">exports<span class="tsd-signature-symbol">:</span> <span class="tsd-signature-type">Record</span><span class="tsd-signature-symbol"><</span><span class="tsd-signature-type">string</span><span class="tsd-signature-symbol">, </span><span class="tsd-signature-type">Function</span><span class="tsd-signature-symbol">></span></div>
<aside class="tsd-sources">
<ul>
- <li>Defined in <a href="https://github.com/apache/tvm/blob/efa5a388c/web/src/runtime.ts#L46">runtime.ts:46</a></li>
+ <li>Defined in <a href="https://github.com/apache/tvm/blob/d907de3b3/web/src/runtime.ts#L46">runtime.ts:46</a></li>
</ul>
</aside>
</section>
@@ -166,7 +166,7 @@
... 2529 lines suppressed ...