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 2023/01/17 23:54:57 UTC

[tvm-site] branch asf-site updated: deploying docs (apache/tvm@7e3dc45fed385e12f8cfd9c9284b5551ada82f95)

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 52f50e5476 deploying docs (apache/tvm@7e3dc45fed385e12f8cfd9c9284b5551ada82f95)
52f50e5476 is described below

commit 52f50e5476e731d2eee95e819c1be97e1a9a4422
Author: tvm-bot <95...@users.noreply.github.com>
AuthorDate: Tue Jan 17 23:54:46 2023 +0000

    deploying docs (apache/tvm@7e3dc45fed385e12f8cfd9c9284b5551ada82f95)
---
 docs/_images/sphx_glr_micro_train_001.png          |   Bin 324216 -> 335230 bytes
 docs/_images/sphx_glr_micro_train_thumb.png        |   Bin 23634 -> 23974 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   |     7 +-
 .../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 +-
 .../extend_tvm/low_level_custom_pass.rst.txt       |    52 +-
 .../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 |   272 +-
 .../how_to/optimize_operators/opt_gemm.rst.txt     |   399 +-
 .../optimize_operators/sg_execution_times.rst.txt  |     8 +-
 .../sg_execution_times.rst.txt                     |    14 +-
 .../tune_conv2d_layer_cuda.rst.txt                 |  1339 ++-
 .../tune_network_cuda.rst.txt                      |     4 +-
 .../tune_network_x86.rst.txt                       |     4 +-
 .../tune_sparse_x86.rst.txt                        |    70 +-
 .../tune_with_autotvm/sg_execution_times.rst.txt   |     6 +-
 .../tune_with_autotvm/tune_conv2d_cuda.rst.txt     |   612 +-
 .../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  |    16 +-
 .../work_with_relay/sg_execution_times.rst.txt     |     8 +-
 .../how_to/work_with_schedules/intrin_math.rst.txt |     2 +-
 .../how_to/work_with_schedules/reduction.rst.txt   |   214 +-
 .../how_to/work_with_schedules/scan.rst.txt        |   159 +-
 .../schedule_primitives.rst.txt                    |   387 +-
 .../work_with_schedules/sg_execution_times.rst.txt |    16 +-
 .../how_to/work_with_schedules/tensorize.rst.txt   |   121 +-
 .../work_with_schedules/tuple_inputs.rst.txt       |   136 +-
 .../tutorials/autotvm/sg_execution_times.rst.txt   |     6 +-
 .../frontend/deploy_classification.rst.txt         |     2 +-
 .../tutorials/frontend/deploy_detection.rst.txt    |     2 +-
 .../tutorials/frontend/sg_execution_times.rst.txt  |     6 +-
 .../topic/vta/tutorials/matrix_multiply.rst.txt    |   227 +-
 .../vta/tutorials/optimize/convolution_opt.rst.txt |   525 +-
 .../tutorials/optimize/matrix_multiply_opt.rst.txt |   350 +-
 .../tutorials/optimize/sg_execution_times.rst.txt  |     6 +-
 .../topic/vta/tutorials/sg_execution_times.rst.txt |     6 +-
 .../topic/vta/tutorials/vta_get_started.rst.txt    |   109 +-
 .../tutorial/auto_scheduler_matmul_x86.rst.txt     |    93 +-
 docs/_sources/tutorial/autotvm_matmul_x86.rst.txt  |    20 +-
 docs/_sources/tutorial/autotvm_relay_x86.rst.txt   |    59 +-
 .../tutorial/cross_compilation_and_rpc.rst.txt     |     2 +-
 docs/_sources/tutorial/intro_topi.rst.txt          |   607 +-
 docs/_sources/tutorial/sg_execution_times.rst.txt  |    22 +-
 .../tutorial/tensor_expr_get_started.rst.txt       |   489 +-
 .../tutorial/tensor_ir_blitz_course.rst.txt        |   147 +-
 docs/commit_hash                                   |     2 +-
 docs/genindex.html                                 |    10 +-
 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       |    14 +-
 docs/how_to/compile_models/from_pytorch.html       |     9 +-
 docs/how_to/compile_models/from_tensorflow.html    |     2 +-
 docs/how_to/compile_models/sg_execution_times.html |    22 +-
 .../deploy_models/deploy_model_on_adreno.html      |     3 +-
 .../deploy_models/deploy_model_on_android.html     |     2 +-
 .../deploy_object_detection_pytorch.html           |    33 +-
 docs/how_to/deploy_models/deploy_prequantized.html |     7 +-
 .../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  |    37 +-
 docs/how_to/deploy_models/sg_execution_times.html  |    20 +-
 .../extend_tvm/bring_your_own_datatypes.html       |     2 +-
 docs/how_to/extend_tvm/low_level_custom_pass.html  |    48 +-
 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    |   268 +-
 docs/how_to/optimize_operators/opt_gemm.html       |   385 +-
 .../optimize_operators/sg_execution_times.html     |     8 +-
 .../sg_execution_times.html                        |    14 +-
 .../tune_conv2d_layer_cuda.html                    |  1337 ++-
 .../tune_with_autoscheduler/tune_network_cuda.html |     4 +-
 .../tune_with_autoscheduler/tune_network_x86.html  |     4 +-
 .../tune_with_autoscheduler/tune_sparse_x86.html   |    68 +-
 .../tune_with_autotvm/sg_execution_times.html      |     6 +-
 .../how_to/tune_with_autotvm/tune_conv2d_cuda.html |   612 +-
 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     |    16 +-
 .../how_to/work_with_relay/sg_execution_times.html |     8 +-
 docs/how_to/work_with_schedules/intrin_math.html   |     2 +-
 docs/how_to/work_with_schedules/reduction.html     |   204 +-
 docs/how_to/work_with_schedules/scan.html          |   153 +-
 .../work_with_schedules/schedule_primitives.html   |   365 +-
 .../work_with_schedules/sg_execution_times.html    |    16 +-
 docs/how_to/work_with_schedules/tensorize.html     |   113 +-
 docs/how_to/work_with_schedules/tuple_inputs.html  |   130 +-
 docs/objects.inv                                   |   Bin 24151 -> 24170 bytes
 docs/reference/api/doxygen/affine__type_8h.html    |     2 +-
 .../api/doxygen/affine__type_8h__incl.svg          |   932 +-
 .../api/doxygen/affine__type_8h_source.html        |     2 +-
 docs/reference/api/doxygen/algorithm_8h.html       |     2 +-
 docs/reference/api/doxygen/algorithm_8h__incl.svg  |  1376 +--
 docs/reference/api/doxygen/algorithms_8h.html      |     2 +-
 docs/reference/api/doxygen/algorithms_8h__incl.svg |  1098 +-
 docs/reference/api/doxygen/analyzer_8h.html        |     2 +-
 docs/reference/api/doxygen/analyzer_8h__incl.svg   |  1056 +-
 docs/reference/api/doxygen/analyzer_8h_source.html |     4 +-
 docs/reference/api/doxygen/annotated.html          |   125 +-
 docs/reference/api/doxygen/annotation_8h.html      |     2 +-
 docs/reference/api/doxygen/annotation_8h__incl.svg |   986 +-
 docs/reference/api/doxygen/arg__info_8h.html       |     2 +-
 docs/reference/api/doxygen/arg__info_8h__incl.svg  |  1322 +--
 .../reference/api/doxygen/arg__info_8h_source.html |     4 +-
 docs/reference/api/doxygen/array_8h__dep__incl.svg |    96 +-
 docs/reference/api/doxygen/array__utils_8h.html    |     2 +-
 .../api/doxygen/array__utils_8h__incl.svg          |  1250 +--
 .../api/doxygen/attr__registry__map_8h.html        |     2 +-
 .../doxygen/attr__registry__map_8h__dep__incl.svg  |  1194 ++-
 docs/reference/api/doxygen/auto__schedule_8h.html  |     2 +-
 .../api/doxygen/auto__schedule_8h__incl.svg        |  1160 +-
 .../doxygen/auto__scheduler_2cost__model_8h.html   |     2 +-
 .../auto__scheduler_2cost__model_8h__incl.svg      |  1076 +-
 docs/reference/api/doxygen/autodiff_8h.html        |     2 +-
 docs/reference/api/doxygen/autodiff_8h__incl.svg   |  1174 +-
 docs/reference/api/doxygen/bitserial_8h.html       |     2 +-
 docs/reference/api/doxygen/bitserial_8h__incl.svg  |  1152 +-
 .../reference/api/doxygen/bitserial_8h_source.html |     2 +-
 docs/reference/api/doxygen/block__scope_8h.html    |     2 +-
 .../api/doxygen/block__scope_8h__incl.svg          |  1088 +-
 docs/reference/api/doxygen/bound_8h.html           |     2 +-
 docs/reference/api/doxygen/bound_8h__incl.svg      |  1084 +-
 docs/reference/api/doxygen/broadcast_8h.html       |     4 +-
 docs/reference/api/doxygen/buffer_8h.html          |     2 +-
 .../reference/api/doxygen/buffer_8h__dep__incl.svg |    12 +-
 docs/reference/api/doxygen/buffer_8h__incl.svg     |   968 +-
 docs/reference/api/doxygen/buffer_8h_source.html   |     6 +-
 docs/reference/api/doxygen/builder_8h_source.html  |     2 +-
 docs/reference/api/doxygen/builtin_8h.html         |     2 +-
 docs/reference/api/doxygen/builtin_8h__incl.svg    |  1218 +--
 docs/reference/api/doxygen/bytecode_8h_source.html |     2 +-
 .../api/doxygen/c__backend__api_8h_source.html     |     4 +-
 .../api/doxygen/c__runtime__api_8h_source.html     |     4 +-
 docs/reference/api/doxygen/call_8h.html            |     2 +-
 docs/reference/api/doxygen/call_8h__incl.svg       |   986 +-
 docs/reference/api/doxygen/classes.html            |   458 +-
 .../classtvm_1_1BaseExprNode__inherit__graph.svg   |    25 +-
 .../api/doxygen/classtvm_1_1BaseFuncNode.html      |     2 +-
 .../classtvm_1_1BaseFuncNode__inherit__graph.svg   |   219 +-
 .../doxygen/classtvm_1_1FloatImmNode-members.html  |    29 +-
 .../api/doxygen/classtvm_1_1FloatImmNode.html      |     7 +-
 .../classtvm_1_1FloatImmNode__coll__graph.svg      |   275 +-
 .../classtvm_1_1FloatImmNode__inherit__graph.svg   |   127 +-
 .../doxygen/classtvm_1_1IRModuleNode-members.html  |    48 +-
 .../api/doxygen/classtvm_1_1IRModuleNode.html      |   157 +-
 .../classtvm_1_1IRModuleNode__coll__graph.svg      |   510 +-
 .../classtvm_1_1IRModuleNode__inherit__graph.svg   |   104 +-
 .../doxygen/classtvm_1_1IntImmNode-members.html    |    29 +-
 .../api/doxygen/classtvm_1_1IntImmNode.html        |     7 +-
 .../classtvm_1_1IntImmNode__coll__graph.svg        |   275 +-
 .../classtvm_1_1IntImmNode__inherit__graph.svg     |   127 +-
 .../doxygen/classtvm_1_1PrimExprNode-members.html  |    19 +-
 .../api/doxygen/classtvm_1_1PrimExprNode.html      |    58 +-
 .../classtvm_1_1PrimExprNode__coll__graph.svg      |   271 +-
 .../classtvm_1_1PrimExprNode__inherit__graph.svg   |   271 +-
 .../api/doxygen/classtvm_1_1RelayExprNode.html     |     2 +-
 .../classtvm_1_1RelayExprNode__inherit__graph.svg  |   647 +-
 .../classtvm_1_1ReprLegacyPrinter-members.html     |    87 +
 .../api/doxygen/classtvm_1_1ReprLegacyPrinter.html |   273 +
 .../classtvm_1_1ReprLegacyPrinter__coll__graph.svg |    27 +
 ...asstvm_1_1arith_1_1IterMapExprNode-members.html |    23 +-
 .../classtvm_1_1arith_1_1IterMapExprNode.html      |     7 +-
 ...vm_1_1arith_1_1IterMapExprNode__coll__graph.svg |   275 +-
 ...1_1arith_1_1IterMapExprNode__inherit__graph.svg |   127 +-
 ...stvm_1_1arith_1_1IterSplitExprNode-members.html |    31 +-
 .../classtvm_1_1arith_1_1IterSplitExprNode.html    |     7 +-
 ..._1_1arith_1_1IterSplitExprNode__coll__graph.svg |   353 +-
 ...1arith_1_1IterSplitExprNode__inherit__graph.svg |   127 +-
 ...asstvm_1_1arith_1_1IterSumExprNode-members.html |    29 +-
 .../classtvm_1_1arith_1_1IterSumExprNode.html      |     7 +-
 ...vm_1_1arith_1_1IterSumExprNode__coll__graph.svg |   347 +-
 ...1_1arith_1_1IterSumExprNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1AddNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1AddNode.html    |     7 +-
 .../classtvm_1_1tir_1_1AddNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1AddNode__inherit__graph.svg |   127 +-
 ...asstvm_1_1tir_1_1AllocateConstNode-members.html |    29 +-
 .../classtvm_1_1tir_1_1AllocateConstNode.html      |     7 +-
 ...vm_1_1tir_1_1AllocateConstNode__coll__graph.svg |   337 +-
 ...1_1tir_1_1AllocateConstNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1AllocateNode-members.html   |    29 +-
 .../doxygen/classtvm_1_1tir_1_1AllocateNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1AllocateNode__coll__graph.svg |   307 +-
 ...stvm_1_1tir_1_1AllocateNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1AndNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1AndNode.html    |     7 +-
 .../classtvm_1_1tir_1_1AndNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1AndNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1AnyNode-members.html        |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1AnyNode.html    |     7 +-
 .../classtvm_1_1tir_1_1AnyNode__coll__graph.svg    |   275 +-
 .../classtvm_1_1tir_1_1AnyNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1AssertStmtNode-members.html |    29 +-
 .../doxygen/classtvm_1_1tir_1_1AssertStmtNode.html |     7 +-
 ...sstvm_1_1tir_1_1AssertStmtNode__coll__graph.svg |   279 +-
 ...vm_1_1tir_1_1AssertStmtNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1AttrStmtNode-members.html   |    31 +-
 .../doxygen/classtvm_1_1tir_1_1AttrStmtNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1AttrStmtNode__coll__graph.svg |   373 +-
 ...stvm_1_1tir_1_1AttrStmtNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1BinaryOpNode-members.html   |    27 +-
 .../doxygen/classtvm_1_1tir_1_1BinaryOpNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1BinaryOpNode__coll__graph.svg |   323 +-
 ...stvm_1_1tir_1_1BinaryOpNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1BlockNode-members.html      |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1BlockNode.html  |     5 +-
 .../classtvm_1_1tir_1_1BlockNode__coll__graph.svg  |    37 +-
 ...lasstvm_1_1tir_1_1BlockNode__inherit__graph.svg |   109 +-
 ...lasstvm_1_1tir_1_1BlockRealizeNode-members.html |    29 +-
 .../classtvm_1_1tir_1_1BlockRealizeNode.html       |     7 +-
 ...tvm_1_1tir_1_1BlockRealizeNode__coll__graph.svg |   329 +-
 ..._1_1tir_1_1BlockRealizeNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1BroadcastNode-members.html  |    29 +-
 .../doxygen/classtvm_1_1tir_1_1BroadcastNode.html  |     7 +-
 ...asstvm_1_1tir_1_1BroadcastNode__coll__graph.svg |   321 +-
 ...tvm_1_1tir_1_1BroadcastNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1BufferLoadNode-members.html |    31 +-
 .../doxygen/classtvm_1_1tir_1_1BufferLoadNode.html |     7 +-
 ...sstvm_1_1tir_1_1BufferLoadNode__coll__graph.svg |   349 +-
 ...vm_1_1tir_1_1BufferLoadNode__inherit__graph.svg |   127 +-
 ...asstvm_1_1tir_1_1BufferRealizeNode-members.html |    29 +-
 .../classtvm_1_1tir_1_1BufferRealizeNode.html      |     7 +-
 ...vm_1_1tir_1_1BufferRealizeNode__coll__graph.svg |   351 +-
 ...1_1tir_1_1BufferRealizeNode__inherit__graph.svg |   109 +-
 ...classtvm_1_1tir_1_1BufferStoreNode-members.html |    31 +-
 .../classtvm_1_1tir_1_1BufferStoreNode.html        |     7 +-
 ...stvm_1_1tir_1_1BufferStoreNode__coll__graph.svg |   327 +-
 ...m_1_1tir_1_1BufferStoreNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1CallNode-members.html       |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1CallNode.html   |     7 +-
 .../classtvm_1_1tir_1_1CallNode__coll__graph.svg   |   345 +-
 ...classtvm_1_1tir_1_1CallNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1CastNode-members.html       |    29 +-
 .../api/doxygen/classtvm_1_1tir_1_1CastNode.html   |     7 +-
 .../classtvm_1_1tir_1_1CastNode__coll__graph.svg   |   321 +-
 ...classtvm_1_1tir_1_1CastNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1CmpOpNode-members.html      |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1CmpOpNode.html  |     7 +-
 .../classtvm_1_1tir_1_1CmpOpNode__coll__graph.svg  |   323 +-
 ...lasstvm_1_1tir_1_1CmpOpNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1DeclBufferNode-members.html |    29 +-
 .../doxygen/classtvm_1_1tir_1_1DeclBufferNode.html |     7 +-
 ...sstvm_1_1tir_1_1DeclBufferNode__coll__graph.svg |   279 +-
 ...vm_1_1tir_1_1DeclBufferNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1DivNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1DivNode.html    |     7 +-
 .../classtvm_1_1tir_1_1DivNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1DivNode__inherit__graph.svg |   127 +-
 .../doxygen/classtvm_1_1tir_1_1EQNode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1EQNode.html     |     7 +-
 .../classtvm_1_1tir_1_1EQNode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1EQNode__inherit__graph.svg  |   127 +-
 .../classtvm_1_1tir_1_1EvaluateNode-members.html   |    31 +-
 .../doxygen/classtvm_1_1tir_1_1EvaluateNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1EvaluateNode__coll__graph.svg |   255 +-
 ...stvm_1_1tir_1_1EvaluateNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1FloorDivNode-members.html   |    27 +-
 .../doxygen/classtvm_1_1tir_1_1FloorDivNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1FloorDivNode__coll__graph.svg |   323 +-
 ...stvm_1_1tir_1_1FloorDivNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1FloorModNode-members.html   |    27 +-
 .../doxygen/classtvm_1_1tir_1_1FloorModNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1FloorModNode__coll__graph.svg |   323 +-
 ...stvm_1_1tir_1_1FloorModNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1ForNode-members.html        |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1ForNode.html    |     7 +-
 .../classtvm_1_1tir_1_1ForNode__coll__graph.svg    |   307 +-
 .../classtvm_1_1tir_1_1ForNode__inherit__graph.svg |   109 +-
 .../doxygen/classtvm_1_1tir_1_1GENode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1GENode.html     |     7 +-
 .../classtvm_1_1tir_1_1GENode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1GENode__inherit__graph.svg  |   127 +-
 .../doxygen/classtvm_1_1tir_1_1GTNode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1GTNode.html     |     7 +-
 .../classtvm_1_1tir_1_1GTNode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1GTNode__inherit__graph.svg  |   127 +-
 .../classtvm_1_1tir_1_1IfThenElseNode-members.html |    31 +-
 .../doxygen/classtvm_1_1tir_1_1IfThenElseNode.html |     7 +-
 ...sstvm_1_1tir_1_1IfThenElseNode__coll__graph.svg |   305 +-
 ...vm_1_1tir_1_1IfThenElseNode__inherit__graph.svg |   109 +-
 .../doxygen/classtvm_1_1tir_1_1LTNode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1LTNode.html     |     7 +-
 .../classtvm_1_1tir_1_1LTNode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1LTNode__inherit__graph.svg  |   127 +-
 .../classtvm_1_1tir_1_1LetNode-members.html        |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1LetNode.html    |     5 +-
 .../classtvm_1_1tir_1_1LetNode__coll__graph.svg    |    25 +-
 .../classtvm_1_1tir_1_1LetNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1LetStmtNode-members.html    |    33 +-
 .../doxygen/classtvm_1_1tir_1_1LetStmtNode.html    |     7 +-
 ...classtvm_1_1tir_1_1LetStmtNode__coll__graph.svg |   283 +-
 ...sstvm_1_1tir_1_1LetStmtNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1LoadNode-members.html       |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1LoadNode.html   |     5 +-
 .../classtvm_1_1tir_1_1LoadNode__coll__graph.svg   |    25 +-
 ...classtvm_1_1tir_1_1LoadNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1MaxNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1MaxNode.html    |     7 +-
 .../classtvm_1_1tir_1_1MaxNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1MaxNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1MinNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1MinNode.html    |     7 +-
 .../classtvm_1_1tir_1_1MinNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1MinNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1ModNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1ModNode.html    |     7 +-
 .../classtvm_1_1tir_1_1ModNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1ModNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1MulNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1MulNode.html    |     7 +-
 .../classtvm_1_1tir_1_1MulNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1MulNode__inherit__graph.svg |   127 +-
 .../doxygen/classtvm_1_1tir_1_1NENode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1NENode.html     |     7 +-
 .../classtvm_1_1tir_1_1NENode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1NENode__inherit__graph.svg  |   127 +-
 .../classtvm_1_1tir_1_1NotNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1NotNode.html    |     7 +-
 .../classtvm_1_1tir_1_1NotNode__coll__graph.svg    |   321 +-
 .../classtvm_1_1tir_1_1NotNode__inherit__graph.svg |   127 +-
 .../doxygen/classtvm_1_1tir_1_1OrNode-members.html |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1OrNode.html     |     7 +-
 .../classtvm_1_1tir_1_1OrNode__coll__graph.svg     |   323 +-
 .../classtvm_1_1tir_1_1OrNode__inherit__graph.svg  |   127 +-
 .../classtvm_1_1tir_1_1PrefetchNode-members.html   |    29 +-
 .../doxygen/classtvm_1_1tir_1_1PrefetchNode.html   |     7 +-
 ...lasstvm_1_1tir_1_1PrefetchNode__coll__graph.svg |   283 +-
 ...stvm_1_1tir_1_1PrefetchNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1PrimFuncNode-members.html   |    35 +-
 .../doxygen/classtvm_1_1tir_1_1PrimFuncNode.html   |    58 +-
 ...lasstvm_1_1tir_1_1PrimFuncNode__coll__graph.svg |   445 +-
 ...stvm_1_1tir_1_1PrimFuncNode__inherit__graph.svg |   187 +-
 ...lasstvm_1_1tir_1_1ProducerLoadNode-members.html |    27 +-
 .../classtvm_1_1tir_1_1ProducerLoadNode.html       |     7 +-
 ...tvm_1_1tir_1_1ProducerLoadNode__coll__graph.svg |   327 +-
 ..._1_1tir_1_1ProducerLoadNode__inherit__graph.svg |   127 +-
 ...stvm_1_1tir_1_1ProducerRealizeNode-members.html |    31 +-
 .../classtvm_1_1tir_1_1ProducerRealizeNode.html    |     5 +-
 ..._1_1tir_1_1ProducerRealizeNode__coll__graph.svg |    37 +-
 ...1tir_1_1ProducerRealizeNode__inherit__graph.svg |   109 +-
 ...asstvm_1_1tir_1_1ProducerStoreNode-members.html |    31 +-
 .../classtvm_1_1tir_1_1ProducerStoreNode.html      |     7 +-
 ...vm_1_1tir_1_1ProducerStoreNode__coll__graph.svg |   307 +-
 ...1_1tir_1_1ProducerStoreNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1RampNode-members.html       |    29 +-
 .../api/doxygen/classtvm_1_1tir_1_1RampNode.html   |     7 +-
 .../classtvm_1_1tir_1_1RampNode__coll__graph.svg   |   323 +-
 ...classtvm_1_1tir_1_1RampNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1ReduceNode-members.html     |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1ReduceNode.html |     7 +-
 .../classtvm_1_1tir_1_1ReduceNode__coll__graph.svg |   405 +-
 ...asstvm_1_1tir_1_1ReduceNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1SelectNode-members.html     |    29 +-
 .../api/doxygen/classtvm_1_1tir_1_1SelectNode.html |     7 +-
 .../classtvm_1_1tir_1_1SelectNode__coll__graph.svg |   325 +-
 ...asstvm_1_1tir_1_1SelectNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1SeqStmtNode-members.html    |    33 +-
 .../doxygen/classtvm_1_1tir_1_1SeqStmtNode.html    |     7 +-
 ...classtvm_1_1tir_1_1SeqStmtNode__coll__graph.svg |   237 +-
 ...sstvm_1_1tir_1_1SeqStmtNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1ShuffleNode-members.html    |    29 +-
 .../doxygen/classtvm_1_1tir_1_1ShuffleNode.html    |     7 +-
 ...classtvm_1_1tir_1_1ShuffleNode__coll__graph.svg |   305 +-
 ...sstvm_1_1tir_1_1ShuffleNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1SizeVarNode-members.html    |    31 +-
 .../doxygen/classtvm_1_1tir_1_1SizeVarNode.html    |     7 +-
 ...classtvm_1_1tir_1_1SizeVarNode__coll__graph.svg |   333 +-
 ...sstvm_1_1tir_1_1SizeVarNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1StmtNode-members.html       |    21 +-
 .../api/doxygen/classtvm_1_1tir_1_1StmtNode.html   |    58 +-
 .../classtvm_1_1tir_1_1StmtNode__coll__graph.svg   |   205 +-
 ...classtvm_1_1tir_1_1StmtNode__inherit__graph.svg |   181 +-
 .../doxygen/classtvm_1_1tir_1_1StmtSRefNode.html   |     2 +-
 ...lasstvm_1_1tir_1_1StmtSRefNode__coll__graph.svg |   215 +-
 .../classtvm_1_1tir_1_1StoreNode-members.html      |    31 +-
 .../api/doxygen/classtvm_1_1tir_1_1StoreNode.html  |     7 +-
 .../classtvm_1_1tir_1_1StoreNode__coll__graph.svg  |   259 +-
 ...lasstvm_1_1tir_1_1StoreNode__inherit__graph.svg |   109 +-
 .../classtvm_1_1tir_1_1StringImmNode-members.html  |    29 +-
 .../doxygen/classtvm_1_1tir_1_1StringImmNode.html  |     7 +-
 ...asstvm_1_1tir_1_1StringImmNode__coll__graph.svg |   309 +-
 ...tvm_1_1tir_1_1StringImmNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1SubNode-members.html        |    27 +-
 .../api/doxygen/classtvm_1_1tir_1_1SubNode.html    |     7 +-
 .../classtvm_1_1tir_1_1SubNode__coll__graph.svg    |   323 +-
 .../classtvm_1_1tir_1_1SubNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1VarNode-members.html        |    29 +-
 .../api/doxygen/classtvm_1_1tir_1_1VarNode.html    |     7 +-
 .../classtvm_1_1tir_1_1VarNode__coll__graph.svg    |   333 +-
 .../classtvm_1_1tir_1_1VarNode__inherit__graph.svg |   127 +-
 .../classtvm_1_1tir_1_1WhileNode-members.html      |    29 +-
 .../api/doxygen/classtvm_1_1tir_1_1WhileNode.html  |     7 +-
 .../classtvm_1_1tir_1_1WhileNode__coll__graph.svg  |   279 +-
 ...lasstvm_1_1tir_1_1WhileNode__inherit__graph.svg |   109 +-
 docs/reference/api/doxygen/codegen_8h_source.html  |     2 +-
 .../api/doxygen/compilation__config_8h.html        |     2 +-
 .../api/doxygen/compilation__config_8h__incl.svg   |  1222 +--
 .../api/doxygen/compilation__config_8h_source.html |     2 +-
 docs/reference/api/doxygen/compute__dag_8h.html    |     2 +-
 .../api/doxygen/compute__dag_8h__incl.svg          |  1234 +--
 .../api/doxygen/constant__utils_8h_source.html     |     2 +-
 .../api/doxygen/cuda_2dense_8h_source.html         |     2 +-
 .../api/doxygen/cuda_2injective_8h_source.html     |     6 +-
 .../api/doxygen/cuda_2pooling_8h_source.html       |     6 +-
 .../api/doxygen/cuda_2reduction_8h_source.html     |     6 +-
 .../api/doxygen/cuda_2softmax_8h_source.html       |     2 +-
 docs/reference/api/doxygen/data__layout_8h.html    |     2 +-
 .../api/doxygen/data__layout_8h__incl.svg          |  1154 +-
 .../api/doxygen/data__type_8h__dep__incl.svg       |    76 +-
 .../api/doxygen/data__type_8h_source.html          |     2 +-
 .../api/doxygen/data__type__rewriter_8h.html       |     2 +-
 .../api/doxygen/data__type__rewriter_8h__incl.svg  |  1148 +-
 .../doxygen/data__type__rewriter_8h_source.html    |    28 +-
 docs/reference/api/doxygen/database_8h_source.html |     2 +-
 .../api/doxygen/dataflow__matcher_8h_source.html   |     4 +-
 .../api/doxygen/dataflow__pattern_8h_source.html   |     2 +-
 .../api/doxygen/detail_2broadcast_8h_source.html   |     2 +-
 .../api/doxygen/device__api_8h_source.html         |     4 +-
 docs/reference/api/doxygen/device__copy_8h.html    |     2 +-
 .../api/doxygen/device__copy_8h__incl.svg          |  1214 +--
 docs/reference/api/doxygen/diagnostic_8h.html      |     4 +-
 .../api/doxygen/diagnostic_8h__dep__incl.svg       |   892 +-
 docs/reference/api/doxygen/diagnostic_8h__incl.svg |  1150 +-
 .../api/doxygen/diagnostic_8h_source.html          |     2 +-
 docs/reference/api/doxygen/dilate_8h.html          |     2 +-
 docs/reference/api/doxygen/dilate_8h__incl.svg     |  1278 +--
 docs/reference/api/doxygen/dilate_8h_source.html   |     2 +-
 docs/reference/api/doxygen/dir_000007_000005.html  |    73 -
 .../dir_63946bee875c6d52bce55e72a67a86ad.html      |     2 +-
 .../dir_63946bee875c6d52bce55e72a67a86ad_dep.svg   |   232 +-
 .../dir_b4c7d8e826c599ba55146c099a14beb5.html      |     2 +-
 .../dir_b4c7d8e826c599ba55146c099a14beb5_dep.svg   |   564 +-
 .../dir_dc867ff9a37cad1764f1670dc7eba6c1.html      |     2 +-
 .../dir_dc867ff9a37cad1764f1670dc7eba6c1_dep.svg   |   255 +-
 docs/reference/api/doxygen/doc_8h.html             |     2 +-
 docs/reference/api/doxygen/doc_8h__incl.svg        |   938 +-
 docs/reference/api/doxygen/doc_8h_source.html      |     4 +-
 .../api/doxygen/driver__api_8h_source.html         |     2 +-
 docs/reference/api/doxygen/elemwise_8h.html        |     2 +-
 docs/reference/api/doxygen/elemwise_8h__incl.svg   |  1160 +-
 docs/reference/api/doxygen/elemwise_8h_source.html |     2 +-
 docs/reference/api/doxygen/env__func_8h.html       |     2 +-
 .../api/doxygen/env__func_8h__dep__incl.svg        |  1254 ++-
 docs/reference/api/doxygen/error_8h.html           |     4 +-
 docs/reference/api/doxygen/error_8h__dep__incl.svg |   901 +-
 docs/reference/api/doxygen/error_8h__incl.svg      |  1162 +-
 docs/reference/api/doxygen/error_8h_source.html    |     4 +-
 docs/reference/api/doxygen/executor_8h.html        |     2 +-
 docs/reference/api/doxygen/executor_8h__incl.svg   |  1192 +--
 docs/reference/api/doxygen/executor_8h_source.html |     2 +-
 .../api/doxygen/extracted__task_8h_source.html     |     4 +-
 docs/reference/api/doxygen/flatten_8h_source.html  |     2 +-
 docs/reference/api/doxygen/functions_c.html        |     9 +-
 docs/reference/api/doxygen/functions_f.html        |     5 +-
 docs/reference/api/doxygen/functions_func_p.html   |     6 +-
 docs/reference/api/doxygen/functions_func_r.html   |     7 +-
 docs/reference/api/doxygen/functions_func_s.html   |    28 +-
 docs/reference/api/doxygen/functions_func_t.html   |     4 +-
 docs/reference/api/doxygen/functions_func_v.html   |    35 +-
 docs/reference/api/doxygen/functions_g.html        |     8 +-
 docs/reference/api/doxygen/functions_i.html        |    12 +-
 docs/reference/api/doxygen/functions_m.html        |     2 +-
 docs/reference/api/doxygen/functions_p.html        |    10 +-
 docs/reference/api/doxygen/functions_r.html        |     7 +-
 docs/reference/api/doxygen/functions_s.html        |    26 +-
 docs/reference/api/doxygen/functions_t.html        |     4 +-
 docs/reference/api/doxygen/functions_type.html     |     3 +-
 docs/reference/api/doxygen/functions_u.html        |     2 +-
 docs/reference/api/doxygen/functions_v.html        |    39 +-
 docs/reference/api/doxygen/functions_vars_c.html   |     3 +
 docs/reference/api/doxygen/functions_vars_g.html   |     6 +
 docs/reference/api/doxygen/functions_vars_i.html   |     6 +-
 docs/reference/api/doxygen/fuse_8h.html            |     2 +-
 docs/reference/api/doxygen/fuse_8h__incl.svg       |  1250 +--
 docs/reference/api/doxygen/generic__func_8h.html   |     2 +-
 .../api/doxygen/generic__func_8h__incl.svg         |  1180 +-
 .../api/doxygen/global__var__supply_8h.html        |     2 +-
 .../api/doxygen/global__var__supply_8h__incl.svg   |  1188 +--
 .../api/doxygen/global__var__supply_8h_source.html |     4 +-
 docs/reference/api/doxygen/hierarchy.html          |   251 +-
 docs/reference/api/doxygen/image_8h.html           |     2 +-
 docs/reference/api/doxygen/image_8h__incl.svg      |  1152 +-
 docs/reference/api/doxygen/index__map_8h.html      |     2 +-
 docs/reference/api/doxygen/index__map_8h__incl.svg |   970 +-
 docs/reference/api/doxygen/inherit_graph_100.svg   |    18 +-
 docs/reference/api/doxygen/inherit_graph_101.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_102.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_103.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_104.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_105.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_106.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_107.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_108.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_109.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_110.svg   |    27 +-
 docs/reference/api/doxygen/inherit_graph_111.svg   | 10626 +------------------
 docs/reference/api/doxygen/inherit_graph_112.svg   | 10626 ++++++++++++++++++-
 docs/reference/api/doxygen/inherit_graph_113.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_114.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_115.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_116.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_117.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_118.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_119.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_12.svg    |    16 +-
 docs/reference/api/doxygen/inherit_graph_120.svg   |  7518 +------------
 docs/reference/api/doxygen/inherit_graph_121.svg   |  7519 ++++++++++++-
 docs/reference/api/doxygen/inherit_graph_122.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_123.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_124.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_125.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_126.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_127.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_128.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_129.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_130.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_131.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_132.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_133.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_134.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_135.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_136.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_137.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_138.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_139.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_140.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_141.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_142.svg   |    61 +-
 docs/reference/api/doxygen/inherit_graph_143.svg   |    62 +-
 docs/reference/api/doxygen/inherit_graph_144.svg   |    21 +-
 docs/reference/api/doxygen/inherit_graph_145.svg   |    21 +-
 docs/reference/api/doxygen/inherit_graph_146.svg   |    20 +-
 docs/reference/api/doxygen/inherit_graph_147.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_148.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_149.svg   |    20 +-
 docs/reference/api/doxygen/inherit_graph_150.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_151.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_152.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_153.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_154.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_155.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_156.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_157.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_158.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_159.svg   |    18 +-
 docs/reference/api/doxygen/inherit_graph_160.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_161.svg   |     4 +-
 docs/reference/api/doxygen/inherit_graph_162.svg   |    21 +-
 docs/reference/api/doxygen/inherit_graph_163.svg   |    21 +-
 docs/reference/api/doxygen/inherit_graph_164.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_165.svg   |    18 +-
 docs/reference/api/doxygen/inherit_graph_166.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_167.svg   |    23 +-
 docs/reference/api/doxygen/inherit_graph_168.svg   |    28 +-
 docs/reference/api/doxygen/inherit_graph_169.svg   |    23 +-
 docs/reference/api/doxygen/inherit_graph_170.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_171.svg   |    18 +-
 docs/reference/api/doxygen/inherit_graph_172.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_173.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_174.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_175.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_176.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_177.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_178.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_179.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_180.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_181.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_182.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_183.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_184.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_185.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_186.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_187.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_188.svg   |    30 +-
 docs/reference/api/doxygen/inherit_graph_189.svg   |    30 +-
 docs/reference/api/doxygen/inherit_graph_190.svg   |    30 +-
 docs/reference/api/doxygen/inherit_graph_191.svg   |    29 +-
 docs/reference/api/doxygen/inherit_graph_192.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_193.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_194.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_195.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_196.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_197.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_198.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_199.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_200.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_201.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_202.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_203.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_204.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_205.svg   |    14 +-
 docs/reference/api/doxygen/inherit_graph_206.svg   |    17 +-
 docs/reference/api/doxygen/inherit_graph_207.svg   |   125 +-
 docs/reference/api/doxygen/inherit_graph_208.svg   |   123 +-
 docs/reference/api/doxygen/inherit_graph_209.svg   |    79 +-
 docs/reference/api/doxygen/inherit_graph_210.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_211.svg   |    18 +-
 docs/reference/api/doxygen/inherit_graph_212.svg   |    19 +-
 docs/reference/api/doxygen/inherit_graph_213.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_214.svg   |    15 +-
 docs/reference/api/doxygen/inherit_graph_215.svg   |    29 +-
 docs/reference/api/doxygen/inherit_graph_216.svg   |    24 +-
 docs/reference/api/doxygen/inherit_graph_217.svg   |    30 +-
 docs/reference/api/doxygen/inherit_graph_218.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_219.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_220.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_221.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_222.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_223.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_224.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_225.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_226.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_227.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_228.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_229.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_230.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_231.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_232.svg   |    12 +-
 docs/reference/api/doxygen/inherit_graph_233.svg   |    12 +-
 ...inherit_graph_233.svg => inherit_graph_234.svg} |     0
 docs/reference/api/doxygen/inherit_graph_41.svg    |    16 +-
 docs/reference/api/doxygen/inherit_graph_45.svg    |     8 +-
 docs/reference/api/doxygen/inherit_graph_89.svg    |    12 +-
 docs/reference/api/doxygen/inherit_graph_90.svg    |    15 +-
 docs/reference/api/doxygen/inherit_graph_91.svg    |    15 +-
 docs/reference/api/doxygen/inherit_graph_92.svg    |    12 +-
 docs/reference/api/doxygen/inherit_graph_93.svg    |    16 +-
 docs/reference/api/doxygen/inherit_graph_94.svg    |    46 +-
 docs/reference/api/doxygen/inherit_graph_95.svg    |    47 +-
 docs/reference/api/doxygen/inherit_graph_96.svg    |    17 +-
 docs/reference/api/doxygen/inherit_graph_97.svg    |    17 +-
 docs/reference/api/doxygen/inherit_graph_98.svg    |    14 +-
 docs/reference/api/doxygen/inherit_graph_99.svg    |    18 +-
 docs/reference/api/doxygen/inherits.html           |   278 +-
 docs/reference/api/doxygen/instrument_8h.html      |     2 +-
 .../api/doxygen/instrument_8h__dep__incl.svg       |   801 +-
 .../api/doxygen/instrument_8h_source.html          |     2 +-
 docs/reference/api/doxygen/int__set_8h.html        |     2 +-
 docs/reference/api/doxygen/int__set_8h__incl.svg   |  1066 +-
 docs/reference/api/doxygen/int__set_8h_source.html |     4 +-
 docs/reference/api/doxygen/int__solver_8h.html     |     2 +-
 .../reference/api/doxygen/int__solver_8h__incl.svg |  1188 +--
 .../api/doxygen/int__solver_8h_source.html         |     4 +-
 .../api/doxygen/interpreter_8h_source.html         |     4 +-
 docs/reference/api/doxygen/ir_2adt_8h.html         |     4 +-
 .../api/doxygen/ir_2adt_8h__dep__incl.svg          |  1096 +-
 docs/reference/api/doxygen/ir_2adt_8h__incl.svg    |   982 +-
 docs/reference/api/doxygen/ir_2adt_8h_source.html  |     8 +-
 docs/reference/api/doxygen/ir_2attrs_8h.html       |     4 +-
 .../api/doxygen/ir_2attrs_8h__dep__incl.svg        |  1080 +-
 docs/reference/api/doxygen/ir_2attrs_8h__incl.svg  |   974 +-
 .../reference/api/doxygen/ir_2attrs_8h_source.html |     6 +-
 docs/reference/api/doxygen/ir_2expr_8h.html        |     2 +-
 .../api/doxygen/ir_2expr_8h__dep__incl.svg         |   100 +-
 docs/reference/api/doxygen/ir_2expr_8h__incl.svg   |   916 +-
 docs/reference/api/doxygen/ir_2expr_8h_source.html |   108 +-
 docs/reference/api/doxygen/ir_2frame_8h.html       |     2 +-
 docs/reference/api/doxygen/ir_2frame_8h__incl.svg  |  1056 +-
 docs/reference/api/doxygen/ir_2function_8h.html    |     4 +-
 .../api/doxygen/ir_2function_8h__dep__incl.svg     |   992 +-
 .../api/doxygen/ir_2function_8h__incl.svg          |  1014 +-
 .../api/doxygen/ir_2function_8h_source.html        |     4 +-
 docs/reference/api/doxygen/ir_2ir_8h.html          |     2 +-
 docs/reference/api/doxygen/ir_2ir_8h__incl.svg     |  1054 +-
 docs/reference/api/doxygen/ir_2module_8h.html      |     4 +-
 .../api/doxygen/ir_2module_8h__dep__incl.svg       |  1098 +-
 docs/reference/api/doxygen/ir_2module_8h__incl.svg |  1164 +-
 .../api/doxygen/ir_2module_8h_source.html          |    56 +-
 docs/reference/api/doxygen/ir_2op_8h.html          |     4 +-
 .../reference/api/doxygen/ir_2op_8h__dep__incl.svg |  1357 +--
 docs/reference/api/doxygen/ir_2op_8h__incl.svg     |  1186 ++-
 docs/reference/api/doxygen/ir_2op_8h_source.html   |     8 +-
 docs/reference/api/doxygen/ir_2span_8h.html        |     2 +-
 .../api/doxygen/ir_2span_8h__dep__incl.svg         |   428 +-
 docs/reference/api/doxygen/ir_2span_8h__incl.svg   |   820 +-
 docs/reference/api/doxygen/ir_2transform_8h.html   |     2 +-
 .../api/doxygen/ir_2transform_8h__dep__incl.svg    |   897 +-
 .../api/doxygen/ir_2transform_8h_source.html       |     2 +-
 docs/reference/api/doxygen/ir_2type_8h.html        |     5 +-
 .../api/doxygen/ir_2type_8h__dep__incl.svg         |   120 +-
 docs/reference/api/doxygen/ir_2type_8h__incl.svg   |   858 +-
 docs/reference/api/doxygen/ir_2type_8h_source.html |    97 +-
 docs/reference/api/doxygen/ir__docsifier_8h.html   |     2 +-
 .../api/doxygen/ir__docsifier_8h__incl.svg         |  1244 +--
 .../api/doxygen/ir__docsifier__functor_8h.html     |     2 +-
 .../doxygen/ir__docsifier__functor_8h__incl.svg    |   854 +-
 .../doxygen/ir__docsifier__functor_8h_source.html  |     8 +-
 .../api/doxygen/iter__affine__map_8h.html          |     2 +-
 .../api/doxygen/iter__affine__map_8h__incl.svg     |  1214 +--
 .../api/doxygen/iter__affine__map_8h_source.html   |     2 +-
 docs/reference/api/doxygen/layer__norm_8h.html     |     2 +-
 .../reference/api/doxygen/layer__norm_8h__incl.svg |  1272 +--
 .../api/doxygen/layer__norm_8h_source.html         |     2 +-
 .../api/doxygen/local__response__norm_8h.html      |     2 +-
 .../api/doxygen/local__response__norm_8h__incl.svg |  1272 +--
 .../doxygen/local__response__norm_8h_source.html   |     4 +-
 docs/reference/api/doxygen/loop__state_8h.html     |     2 +-
 .../reference/api/doxygen/loop__state_8h__incl.svg |  1150 +-
 docs/reference/api/doxygen/map_8h__dep__incl.svg   |    76 +-
 docs/reference/api/doxygen/mapping_8h.html         |     2 +-
 docs/reference/api/doxygen/mapping_8h__incl.svg    |  1272 +--
 docs/reference/api/doxygen/measure_8h.html         |     2 +-
 docs/reference/api/doxygen/measure_8h__incl.svg    |  1064 +-
 .../api/doxygen/measure__candidate_8h.html         |     2 +-
 .../api/doxygen/measure__candidate_8h__incl.svg    |  1264 +--
 docs/reference/api/doxygen/measure__record_8h.html |     2 +-
 .../api/doxygen/measure__record_8h__incl.svg       |  1070 +-
 docs/reference/api/doxygen/memory__pools_8h.html   |     2 +-
 .../api/doxygen/memory__pools_8h__incl.svg         |  1170 +-
 .../api/doxygen/memory__pools_8h_source.html       |     4 +-
 docs/reference/api/doxygen/metadata_8h.html        |     2 +-
 docs/reference/api/doxygen/metadata_8h__incl.svg   |   940 +-
 docs/reference/api/doxygen/metadata__base_8h.html  |     2 +-
 .../api/doxygen/metadata__base_8h__incl.svg        |   956 +-
 docs/reference/api/doxygen/name__supply_8h.html    |     2 +-
 .../api/doxygen/name__supply_8h__incl.svg          |   938 +-
 docs/reference/api/doxygen/namespacemembers.html   |     3 +
 .../api/doxygen/namespacemembers_func.html         |     3 +
 .../api/doxygen/namespacemembers_func_l.html       |    13 +-
 .../api/doxygen/namespacemembers_func_s.html       |    11 +-
 .../api/doxygen/namespacemembers_func_t.html       |     3 +
 docs/reference/api/doxygen/namespacemembers_l.html |    15 +-
 docs/reference/api/doxygen/namespacemembers_s.html |    11 +-
 docs/reference/api/doxygen/namespacemembers_t.html |     3 +
 docs/reference/api/doxygen/namespacetvm.html       |   548 +-
 .../api/doxygen/namespacetvm_1_1runtime.html       |    28 +
 .../doxygen/namespacetvm_1_1script_1_1printer.html |    62 -
 .../reference/api/doxygen/namespacetvm_1_1tir.html |    23 -
 .../api/doxygen/ndarray_8h__dep__incl.svg          |    28 +-
 docs/reference/api/doxygen/nn_2bnn_8h.html         |     2 +-
 docs/reference/api/doxygen/nn_2bnn_8h__incl.svg    |  1404 +--
 docs/reference/api/doxygen/nn_2bnn_8h_source.html  |     2 +-
 docs/reference/api/doxygen/nn_2dense_8h.html       |     2 +-
 docs/reference/api/doxygen/nn_2dense_8h__incl.svg  |  1272 +--
 .../reference/api/doxygen/nn_2dense_8h_source.html |     2 +-
 .../api/doxygen/nn_2pooling_8h_source.html         |     4 +-
 .../api/doxygen/nn_2softmax_8h_source.html         |     6 +-
 docs/reference/api/doxygen/node_8h.html            |     2 +-
 docs/reference/api/doxygen/node_8h__dep__incl.svg  |    56 +-
 docs/reference/api/doxygen/node_8h__incl.svg       |   800 +-
 .../reference/api/doxygen/object_8h__dep__incl.svg |    28 +-
 docs/reference/api/doxygen/object_8h_source.html   |     2 +-
 .../api/doxygen/object__path_8h__dep__incl.svg     |   114 +-
 docs/reference/api/doxygen/on__device_8h.html      |     2 +-
 docs/reference/api/doxygen/on__device_8h__incl.svg |  1214 +--
 .../api/doxygen/op__strategy_8h_source.html        |     2 +-
 docs/reference/api/doxygen/operation_8h.html       |     2 +-
 docs/reference/api/doxygen/operation_8h__incl.svg  |  1272 +--
 .../reference/api/doxygen/operation_8h_source.html |     6 +-
 .../api/doxygen/optional_8h__dep__incl.svg         |    20 +-
 .../api/doxygen/packed__func_8h__dep__incl.svg     |    48 +-
 docs/reference/api/doxygen/pad__utils_8h.html      |     2 +-
 docs/reference/api/doxygen/pad__utils_8h__incl.svg |  1276 +--
 docs/reference/api/doxygen/parser_8h_source.html   |     2 +-
 docs/reference/api/doxygen/pattern_8h.html         |     2 +-
 docs/reference/api/doxygen/pattern_8h__incl.svg    |  1064 +-
 docs/reference/api/doxygen/printer_8h.html         |     5 +-
 docs/reference/api/doxygen/printer_8h__incl.svg    |  1188 +--
 docs/reference/api/doxygen/printer_8h_source.html  |     4 +-
 .../reference/api/doxygen/profiling_8h_source.html |     4 +-
 docs/reference/api/doxygen/random_8h.html          |     2 +-
 docs/reference/api/doxygen/random_8h__incl.svg     |   976 +-
 docs/reference/api/doxygen/random_8h_source.html   |     2 +-
 docs/reference/api/doxygen/ravel__unravel_8h.html  |     2 +-
 .../api/doxygen/ravel__unravel_8h__incl.svg        |  1256 +--
 docs/reference/api/doxygen/reduce_8h.html          |     2 +-
 docs/reference/api/doxygen/reduce_8h__incl.svg     |   986 +-
 .../reference/api/doxygen/reduction_8h_source.html |     4 +-
 docs/reference/api/doxygen/reflection_8h.html      |     2 +-
 .../api/doxygen/reflection_8h__dep__incl.svg       |   895 +-
 docs/reference/api/doxygen/registry_8h.html        |     2 +-
 .../api/doxygen/registry_8h__dep__incl.svg         |  1237 +--
 docs/reference/api/doxygen/relay_2adt_8h.html      |     4 +-
 .../api/doxygen/relay_2adt_8h__dep__incl.svg       |    60 +-
 docs/reference/api/doxygen/relay_2adt_8h__incl.svg |  1316 +--
 .../api/doxygen/relay_2adt_8h_source.html          |     4 +-
 .../api/doxygen/relay_2attrs_2debug_8h.html        |     2 +-
 .../api/doxygen/relay_2attrs_2debug_8h__incl.svg   |  1006 +-
 .../reference/api/doxygen/relay_2attrs_2nn_8h.html |     2 +-
 .../api/doxygen/relay_2attrs_2nn_8h__incl.svg      |  1150 +-
 .../api/doxygen/relay_2attrs_2nn_8h_source.html    |     2 +-
 .../api/doxygen/relay_2attrs_2transform_8h.html    |     2 +-
 .../doxygen/relay_2attrs_2transform_8h__incl.svg   |  1406 +--
 .../doxygen/relay_2attrs_2transform_8h_source.html |     6 +-
 .../reference/api/doxygen/relay_2attrs_2vm_8h.html |     2 +-
 .../api/doxygen/relay_2attrs_2vm_8h__incl.svg      |   974 +-
 docs/reference/api/doxygen/relay_2base_8h.html     |     4 +-
 .../api/doxygen/relay_2base_8h__dep__incl.svg      |   644 +-
 .../reference/api/doxygen/relay_2base_8h__incl.svg |  1084 +-
 .../api/doxygen/relay_2base_8h_source.html         |     2 +-
 docs/reference/api/doxygen/relay_2expr_8h.html     |     2 +-
 .../api/doxygen/relay_2expr_8h__dep__incl.svg      |   430 +-
 .../api/doxygen/relay_2expr_8h_source.html         |    14 +-
 .../doxygen/relay_2expr__functor_8h_source.html    |     4 +-
 .../api/doxygen/relay_2feature_8h_source.html      |     6 +-
 .../api/doxygen/relay_2function_8h_source.html     |     8 +-
 .../doxygen/relay_2op__attr__types_8h_source.html  |     2 +-
 .../api/doxygen/relay_2qnn_2attrs_8h.html          |     2 +-
 .../api/doxygen/relay_2qnn_2attrs_8h__incl.svg     |   986 +-
 .../api/doxygen/relay_2transform_8h_source.html    |     4 +-
 docs/reference/api/doxygen/relay_2type_8h.html     |     4 +-
 .../api/doxygen/relay_2type_8h__dep__incl.svg      |   574 +-
 .../reference/api/doxygen/relay_2type_8h__incl.svg |  1330 +--
 .../api/doxygen/relay_2type_8h_source.html         |    28 +-
 docs/reference/api/doxygen/repr__printer_8h.html   |     8 +-
 .../api/doxygen/repr__printer_8h__dep__incl.svg    |    56 +-
 .../api/doxygen/repr__printer_8h__incl.svg         |   148 +-
 .../api/doxygen/repr__printer_8h_source.html       |    14 +-
 docs/reference/api/doxygen/runner_8h.html          |     2 +-
 docs/reference/api/doxygen/runner_8h__incl.svg     |  1304 +--
 .../api/doxygen/runtime_2container_2adt_8h.html    |     2 +-
 .../runtime_2container_2adt_8h__dep__incl.svg      |  1014 +-
 .../runtime_2container_2base_8h__dep__incl.svg     |    16 +-
 .../api/doxygen/runtime_2module_8h__dep__incl.svg  |    48 +-
 .../api/doxygen/runtime_2module_8h_source.html     |     2 +-
 .../api/doxygen/runtime_2vm_2vm_8h_source.html     |     2 +-
 docs/reference/api/doxygen/runtime_8h.html         |     2 +-
 docs/reference/api/doxygen/runtime_8h__incl.svg    |  1192 +--
 docs/reference/api/doxygen/schedule__pass_8h.html  |     2 +-
 .../api/doxygen/schedule__pass_8h__incl.svg        |  1286 +--
 docs/reference/api/doxygen/schedule__rule_8h.html  |     2 +-
 .../api/doxygen/schedule__rule_8h__incl.svg        |  1010 +-
 .../api/doxygen/schedule__rule_8h_source.html      |     2 +-
 .../api/doxygen/script_2ir__builder_2base_8h.html  |     2 +-
 .../doxygen/script_2ir__builder_2base_8h__incl.svg |  1032 +-
 docs/reference/api/doxygen/search/all_10.js        |     2 +-
 docs/reference/api/doxygen/search/all_11.js        |     8 +-
 docs/reference/api/doxygen/search/all_13.js        |     9 +-
 docs/reference/api/doxygen/search/all_14.js        |    18 +-
 docs/reference/api/doxygen/search/all_15.js        |    13 +-
 docs/reference/api/doxygen/search/all_16.js        |     2 +-
 docs/reference/api/doxygen/search/all_17.js        |     4 +-
 docs/reference/api/doxygen/search/all_2.js         |     1 +
 docs/reference/api/doxygen/search/all_4.js         |     1 +
 docs/reference/api/doxygen/search/all_7.js         |     2 +-
 docs/reference/api/doxygen/search/all_8.js         |     2 +
 docs/reference/api/doxygen/search/all_9.js         |     2 +-
 docs/reference/api/doxygen/search/all_a.js         |     5 +-
 docs/reference/api/doxygen/search/all_d.js         |     1 -
 docs/reference/api/doxygen/search/all_e.js         |     6 +-
 docs/reference/api/doxygen/search/classes_10.js    |     2 +-
 docs/reference/api/doxygen/search/classes_11.js    |     6 +-
 docs/reference/api/doxygen/search/classes_7.js     |     2 +-
 docs/reference/api/doxygen/search/classes_f.js     |     3 +-
 docs/reference/api/doxygen/search/functions_1.js   |     1 +
 docs/reference/api/doxygen/search/functions_10.js  |     6 +-
 docs/reference/api/doxygen/search/functions_12.js  |     5 +-
 docs/reference/api/doxygen/search/functions_13.js  |    11 +-
 docs/reference/api/doxygen/search/functions_14.js  |     3 +-
 docs/reference/api/doxygen/search/functions_16.js  |     2 +-
 docs/reference/api/doxygen/search/functions_9.js   |     2 +-
 docs/reference/api/doxygen/search/functions_c.js   |     1 -
 docs/reference/api/doxygen/search/functions_d.js   |     4 +-
 docs/reference/api/doxygen/search/typedefs_5.js    |     2 +-
 docs/reference/api/doxygen/search/variables_3.js   |     1 +
 docs/reference/api/doxygen/search/variables_7.js   |     2 +
 docs/reference/api/doxygen/search/variables_9.js   |     3 +-
 docs/reference/api/doxygen/search__policy_8h.html  |     2 +-
 .../api/doxygen/search__policy_8h__incl.svg        |  1178 +-
 .../reference/api/doxygen/search__strategy_8h.html |     2 +-
 .../api/doxygen/search__strategy_8h__incl.svg      |  1394 +--
 .../api/doxygen/serializer_8h__dep__incl.svg       |    28 +-
 .../api/doxygen/serializer_8h_source.html          |     2 +-
 .../api/doxygen/shape__tuple_8h__dep__incl.svg     |    28 +-
 docs/reference/api/doxygen/source__map_8h.html     |     4 +-
 .../api/doxygen/source__map_8h__dep__incl.svg      |  1062 +-
 .../reference/api/doxygen/source__map_8h__incl.svg |   904 +-
 .../api/doxygen/space__generator_8h_source.html    |     2 +-
 docs/reference/api/doxygen/state_8h_source.html    |     4 +-
 docs/reference/api/doxygen/stmt_8h.html            |     2 +-
 docs/reference/api/doxygen/stmt_8h__incl.svg       |  1074 +-
 docs/reference/api/doxygen/stmt_8h_source.html     |   541 +-
 docs/reference/api/doxygen/stmt__functor_8h.html   |     2 +-
 .../api/doxygen/stmt__functor_8h__incl.svg         |  1152 +-
 .../api/doxygen/stmt__functor_8h_source.html       |    44 +-
 docs/reference/api/doxygen/strided__slice_8h.html  |     2 +-
 .../api/doxygen/strided__slice_8h__incl.svg        |  1178 +-
 .../api/doxygen/strided__slice_8h_source.html      |     2 +-
 .../reference/api/doxygen/string_8h__dep__incl.svg |    72 +-
 docs/reference/api/doxygen/string_8h_source.html   |     2 +-
 .../structtvm_1_1tir_1_1LENode-members.html        |    27 +-
 .../api/doxygen/structtvm_1_1tir_1_1LENode.html    |     7 +-
 .../structtvm_1_1tir_1_1LENode__coll__graph.svg    |   323 +-
 .../structtvm_1_1tir_1_1LENode__inherit__graph.svg |   127 +-
 .../doxygen/structural__equal_8h__dep__incl.svg    |   118 +-
 .../api/doxygen/structural__hash_8h__dep__incl.svg |   118 +-
 docs/reference/api/doxygen/tag_8h.html             |     2 +-
 docs/reference/api/doxygen/tag_8h__incl.svg        |  1094 +-
 docs/reference/api/doxygen/target_8h.html          |     2 +-
 .../reference/api/doxygen/target_8h__dep__incl.svg |    20 +-
 docs/reference/api/doxygen/target_8h__incl.svg     |  1248 +--
 docs/reference/api/doxygen/target__info_8h.html    |     2 +-
 .../api/doxygen/target__info_8h__incl.svg          |   932 +-
 .../api/doxygen/target__info_8h_source.html        |     2 +-
 docs/reference/api/doxygen/target__kind_8h.html    |     2 +-
 .../api/doxygen/target__kind_8h__dep__incl.svg     |    20 +-
 .../api/doxygen/target__kind_8h__incl.svg          |  1184 ++-
 .../api/doxygen/target__kind_8h_source.html        |     2 +-
 docs/reference/api/doxygen/te_2schedule_8h.html    |     2 +-
 .../api/doxygen/te_2schedule_8h__incl.svg          |  1164 +-
 .../api/doxygen/te_2schedule_8h_source.html        |     4 +-
 docs/reference/api/doxygen/tensor_8h.html          |     2 +-
 docs/reference/api/doxygen/tensor_8h__incl.svg     |  1216 +--
 docs/reference/api/doxygen/tensor_8h_source.html   |     2 +-
 docs/reference/api/doxygen/tensor__intrin_8h.html  |     2 +-
 .../api/doxygen/tensor__intrin_8h__incl.svg        |  1204 +--
 .../api/doxygen/tensor__intrin_8h_source.html      |     2 +-
 docs/reference/api/doxygen/tensor__type_8h.html    |     4 +-
 .../api/doxygen/tensor__type_8h__dep__incl.svg     |   398 +-
 .../api/doxygen/tensor__type_8h__incl.svg          |   932 +-
 .../api/doxygen/tensor__type_8h_source.html        |     2 +-
 docs/reference/api/doxygen/tensor__utils_8h.html   |     2 +-
 .../api/doxygen/tensor__utils_8h__incl.svg         |  1256 +--
 docs/reference/api/doxygen/thread__bind_8h.html    |     2 +-
 .../api/doxygen/thread__bind_8h__incl.svg          |  1192 +--
 .../api/doxygen/tir_2analysis_8h_source.html       |    16 +-
 docs/reference/api/doxygen/tir_2expr_8h.html       |     5 +-
 .../api/doxygen/tir_2expr_8h__dep__incl.svg        |   292 +-
 docs/reference/api/doxygen/tir_2expr_8h__incl.svg  |  1060 +-
 .../reference/api/doxygen/tir_2expr_8h_source.html |    13 +-
 .../api/doxygen/tir_2expr__functor_8h.html         |     2 +-
 .../api/doxygen/tir_2expr__functor_8h__incl.svg    |  1068 +-
 .../api/doxygen/tir_2expr__functor_8h_source.html  |     6 +-
 docs/reference/api/doxygen/tir_2frame_8h.html      |     2 +-
 docs/reference/api/doxygen/tir_2frame_8h__incl.svg |  1184 ++-
 .../api/doxygen/tir_2frame_8h_source.html          |     6 +-
 .../api/doxygen/tir_2function_8h_source.html       |    36 +-
 docs/reference/api/doxygen/tir_2ir_8h.html         |     2 +-
 docs/reference/api/doxygen/tir_2ir_8h__incl.svg    |  1246 +--
 docs/reference/api/doxygen/tir_2ir_8h_source.html  |     8 +-
 docs/reference/api/doxygen/tir_2op_8h.html         |     2 +-
 docs/reference/api/doxygen/tir_2op_8h__incl.svg    |  1282 +--
 docs/reference/api/doxygen/tir_2op_8h_source.html  |    18 +-
 .../api/doxygen/tir_2op__attr__types_8h.html       |     2 +-
 .../api/doxygen/tir_2op__attr__types_8h__incl.svg  |   940 +-
 .../doxygen/tir_2op__attr__types_8h_source.html    |     2 +-
 .../api/doxygen/tir_2schedule_2schedule_8h.html    |     2 +-
 .../doxygen/tir_2schedule_2schedule_8h__incl.svg   |  1196 +--
 .../doxygen/tir_2schedule_2schedule_8h_source.html |    12 +-
 .../api/doxygen/tir_2usmp_2transform_8h.html       |     2 +-
 .../api/doxygen/tir_2usmp_2transform_8h__incl.svg  |  1098 +-
 .../doxygen/tir_2usmp_2transform_8h_source.html    |     2 +-
 .../reference/api/doxygen/tir_2usmp_2utils_8h.html |     2 +-
 .../api/doxygen/tir_2usmp_2utils_8h__incl.svg      |  1112 +-
 .../api/doxygen/tir_2usmp_2utils_8h_source.html    |     8 +-
 docs/reference/api/doxygen/topi_2nn_8h_source.html |     6 +-
 .../api/doxygen/topi_2transform_8h_source.html     |    14 +-
 docs/reference/api/doxygen/topi_2utils_8h.html     |     2 +-
 .../reference/api/doxygen/topi_2utils_8h__incl.svg |   930 +-
 .../api/doxygen/topi_2utils_8h_source.html         |     2 +-
 docs/reference/api/doxygen/transform__step_8h.html |     2 +-
 .../api/doxygen/transform__step_8h__incl.svg       |  1140 +-
 .../api/doxygen/transform__step_8h_source.html     |     2 +-
 docs/reference/api/doxygen/type__functor_8h.html   |     8 +-
 .../api/doxygen/type__functor_8h__incl.svg         |  2631 +++--
 .../api/doxygen/type__functor_8h_source.html       |    20 +-
 docs/reference/api/doxygen/type__relation_8h.html  |     2 +-
 .../api/doxygen/type__relation_8h__dep__incl.svg   |  1286 ++-
 .../api/doxygen/type__relation_8h_source.html      |     8 +-
 docs/reference/api/doxygen/var_8h.html             |     2 +-
 docs/reference/api/doxygen/var_8h__dep__incl.svg   |   212 +-
 docs/reference/api/doxygen/var_8h__incl.svg        |   942 +-
 docs/reference/api/doxygen/var_8h_source.html      |     4 +-
 docs/reference/api/doxygen/virtual__device_8h.html |     4 +-
 .../api/doxygen/virtual__device_8h__dep__incl.svg  |   512 +-
 .../api/doxygen/virtual__device_8h__incl.svg       |  1216 +--
 .../api/doxygen/virtual__device_8h_source.html     |     2 +-
 docs/reference/api/doxygen/vision_8h.html          |     2 +-
 docs/reference/api/doxygen/vision_8h__incl.svg     |  1152 +-
 docs/reference/api/doxygen/vision_8h_source.html   |     2 +-
 docs/reference/api/doxygen/winograd_8h.html        |     2 +-
 docs/reference/api/doxygen/winograd_8h__incl.svg   |  1196 +--
 docs/reference/api/doxygen/with_8h__dep__incl.svg  |    44 +-
 docs/reference/api/python/auto_scheduler.html      |     4 +-
 docs/reference/api/python/ir.html                  |  2141 ++--
 docs/reference/api/python/tir.html                 |   246 +-
 .../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  |     6 +-
 .../tutorials/frontend/deploy_classification.html  |     2 +-
 .../vta/tutorials/frontend/deploy_detection.html   |     2 +-
 .../vta/tutorials/frontend/sg_execution_times.html |     6 +-
 docs/topic/vta/tutorials/matrix_multiply.html      |   221 +-
 .../vta/tutorials/optimize/convolution_opt.html    |   519 +-
 .../tutorials/optimize/matrix_multiply_opt.html    |   344 +-
 .../vta/tutorials/optimize/sg_execution_times.html |     6 +-
 docs/topic/vta/tutorials/sg_execution_times.html   |     6 +-
 docs/topic/vta/tutorials/vta_get_started.html      |   105 +-
 docs/tutorial/auto_scheduler_matmul_x86.html       |    87 +-
 docs/tutorial/autotvm_matmul_x86.html              |    20 +-
 docs/tutorial/autotvm_relay_x86.html               |   273 +-
 docs/tutorial/cross_compilation_and_rpc.html       |     2 +-
 docs/tutorial/intro_topi.html                      |   597 +-
 docs/tutorial/sg_execution_times.html              |    22 +-
 docs/tutorial/tensor_expr_get_started.html         |   471 +-
 docs/tutorial/tensor_ir_blitz_course.html          |   137 +-
 1038 files changed, 122092 insertions(+), 121052 deletions(-)

diff --git a/docs/_images/sphx_glr_micro_train_001.png b/docs/_images/sphx_glr_micro_train_001.png
index 58230570fb..4730ebaecb 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 c7f45c5bc4..4f63c99e35 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 117035f891..d1a6ab7235 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  18.798 seconds)
+   **Total running time of the script:** ( 1 minutes  13.059 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 25f0bb76f6..84bf98fa93 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 1000ms/step
+
    1/1 [==============================] - ETA: 0s
    1/1 [==============================] - 1s 952ms/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 567ee1c539..9d6bad40bb 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.zipfaabf753-aa08-47b4-a3a6-91f5abd6fdd6 from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
+    Downloading /workspace/.mxnet/models/resnet18_v1-a0666292.zipf850238c-fdb0-49a7-83f7-d5e376bb9e0e 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 2dbe62b04f..203268401f 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, 43.6MB/s]
     39%|###8      | 16.0M/41.5M [00:00<00:00, 56.4MB/s]
     54%|#####3    | 22.3M/41.5M [00:00<00:00, 59.9MB/s]
     68%|######8   | 28.3M/41.5M [00:00<00:00, 56.5MB/s]
     82%|########1 | 33.8M/41.5M [00:00<00:00, 55.8MB/s]
     95%|#########4| 39.2M/41.5M [00:00<00:00, 52.5MB/s]
    100%|##########| 41.5M/41.5M [00:00<00:00, 52.8MB/s]
+
      0%|          | 0.00/41.5M [00:00<?, ?B/s]
     15%|#5        | 6.33M/41.5M [00:00<00:00, 61.2MB/s]
     29%|##9       | 12.2M/41.5M [00:00<00:00, 52.0MB/s]
     48%|####7     | 19.8M/41.5M [00:00<00:00, 58.6MB/s]
     61%|######1   | 25.4M/41.5M [00:00<00:00, 49.5MB/s]
     77%|#######7  | 32.0M/41.5M [00:00<00:00, 48.2MB/s]
     96%|#########6| 40.0M/41.5M [00:00<00:00, 56.5MB/s]
    100%|##########| 41.5M/41.5M [00:00<00:00, 55.5MB/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 969ae7f6cd..e7b9b7f898 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]
     21%|##        | 9.23M/44.7M [00:00<00:00, 96.8MB/s]
     41%|####1     | 18.5M/44.7M [00:00<00:00, 53.0MB/s]
     69%|######9   | 31.0M/44.7M [00:00<00:00, 77.8MB/s]
     90%|########9 | 40.1M/44.7M [00:00<00:00, 78.7MB/s]
    100%|##########| 44.7M/44.7M [00:00<00:00, 77.3MB/s]
+
      0%|          | 0.00/44.7M [00:00<?, ?B/s]
     34%|###4      | 15.4M/44.7M [00:00<00:00, 161MB/s]
     69%|######8   | 30.7M/44.7M [00:00<00:00, 119MB/s]
     96%|#########5| 42.7M/44.7M [00:00<00:00, 111MB/s]
    100%|##########| 44.7M/44.7M [00:00<00:00, 115MB/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 59aef21764..0024cf8b24 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  20.954 seconds)
+   **Total running time of the script:** ( 1 minutes  18.793 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 452f7df655..5cae5cb422 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
 =================
-**06:23.938** total execution time for **how_to_compile_models** files:
+**06:06.722** total execution time for **how_to_compile_models** files:
 
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_tensorflow.py` (``from_tensorflow.py``) | 01:20.954 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_tensorflow.py` (``from_tensorflow.py``) | 01:18.793 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_darknet.py` (``from_darknet.py``)       | 01:18.798 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_darknet.py` (``from_darknet.py``)       | 01:13.059 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_paddle.py` (``from_paddle.py``)         | 00:52.499 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_paddle.py` (``from_paddle.py``)         | 00:50.030 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_oneflow.py` (``from_oneflow.py``)       | 00:35.739 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_oneflow.py` (``from_oneflow.py``)       | 00:34.474 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_mxnet.py` (``from_mxnet.py``)           | 00:31.009 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_mxnet.py` (``from_mxnet.py``)           | 00:30.104 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_coreml.py` (``from_coreml.py``)         | 00:30.661 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_coreml.py` (``from_coreml.py``)         | 00:29.098 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_tflite.py` (``from_tflite.py``)         | 00:26.645 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_tflite.py` (``from_tflite.py``)         | 00:25.686 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_pytorch.py` (``from_pytorch.py``)       | 00:24.472 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_pytorch.py` (``from_pytorch.py``)       | 00:23.616 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_keras.py` (``from_keras.py``)           | 00:20.480 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_keras.py` (``from_keras.py``)           | 00:19.290 | 0.0 MB |
 +-----------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_compile_models_from_onnx.py` (``from_onnx.py``)             | 00:02.680 | 0.0 MB |
+| :ref:`sphx_glr_how_to_compile_models_from_onnx.py` (``from_onnx.py``)             | 00:02.573 | 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 92603ea1c0..9868c24cde 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
@@ -727,13 +727,18 @@ 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)  
-     2685.5380    2685.2784    2688.3172    2683.2740      1.5475   
+     3303.7416    3294.8620    3332.7372    3291.4654     14.2290   
                
 
 
 
 
 
+.. rst-class:: sphx-glr-timing
+
+   **Total running time of the script:** ( 1 minutes  1.415 seconds)
+
+
 .. _sphx_glr_download_how_to_deploy_models_deploy_model_on_adreno.py:
 
 .. only:: html
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 039906f8b9..3a14ec65b6 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)  
-      16.6962      16.7289      17.6200      15.6778       0.5175   
+      15.7950      15.6632      16.5296      15.5237       0.3098   
                
 
 
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 8efefa6f34..c30d80b021 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]
      2%|2         | 3.48M/170M [00:00<00:04, 36.5MB/s]
      5%|4         | 7.99M/170M [00:00<00:04, 41.6MB/s]
     11%|#         | 18.6M/170M [00:00<00:02, 54.3MB/s]
     14%|#4        | 24.0M/170M [00:00<00:03, 46.8MB/s]
     19%|#8        | 32.0M/170M [00:00<00:02, 55.2MB/s]
     26%|##5       | 43.6M/170M [00:00<00:01, 74.0MB/s]
     33%|###2      | 56.0M/170M [00:00<00:01, 81.6MB/s]
     42%|####2     | 72.0M/170M [00:00<00:00, 103MB/s] 
     48%|####8     | 82.2M/170M [00:01<00:00, 99.9MB/s]
     57%|#####6    | 96.0M/170M [00:01<00:00, 98.7MB/s]
     62%|######2   | 106M/170M [00:01<00:00, 79.6MB/s] 
     71%|#######   | 120M/170M [00:01<00:00, 86.6MB/s]
     79%|#######8  | 134M/170M [00:01<00:00, 99.7MB/s]
     85%|########4 | 144M/170M [00:01<00:00, 87.0MB/s]
     90%|######### | 153M/170M [00:02<00:00, 77.9MB/s]
     95%|#########4| 161M/170M [00:02<00:00, 79.7MB/s]
    100%|##########| 170M/170M [00:02<00:00, 80.1MB/s]
+
      0%|          | 0.00/170M [00:00<?, ?B/s]
      7%|6         | 11.2M/170M [00:00<00:01, 117MB/s]
     13%|#3        | 22.4M/170M [00:00<00:01, 105MB/s]
     19%|#9        | 32.5M/170M [00:00<00:01, 88.2MB/s]
     24%|##4       | 41.1M/170M [00:00<00:02, 67.1MB/s]
     33%|###2      | 55.3M/170M [00:00<00:01, 88.9MB/s]
     42%|####2     | 72.0M/170M [00:00<00:01, 101MB/s] 
     49%|####9     | 83.9M/170M [00:00<00:00, 108MB/s]
     57%|#####6    | 96.1M/170M [00:01<00:00, 110MB/s]
     63%|######3   | 107M/170M [00:01<00:00, 80.0MB/s]
     68%|######8   | 116M/170M [00:01<00:00, 68.6MB/s]
     76%|#######6  | 129M/170M [00:01<00:00, 83.5MB/s]
     86%|########6 | 147M/170M [00:01<00:00, 105MB/s] 
     93%|#########3| 158M/170M [00:01<00:00, 98.8MB/s]
    100%|##########| 170M/170M [00:01<00:00, 95.1MB/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  24.271 seconds)
+   **Total running time of the script:** ( 3 minutes  23.093 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 b62300fb70..b9ca038cc6 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]
    100%|##########| 13.6M/13.6M [00:00<00:00, 163MB/s]
+
      0%|          | 0.00/13.6M [00:00<?, ?B/s]
     59%|#####8    | 7.99M/13.6M [00:00<00:00, 65.0MB/s]
    100%|##########| 13.6M/13.6M [00:00<00:00, 87.9MB/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)  
-      90.2511      90.1482      94.6582      90.0187       0.4841   
+      90.4932      90.2412      107.9217     90.0188       1.7672   
                
 
 
@@ -458,7 +458,7 @@ TODO
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 1 minutes  11.765 seconds)
+   **Total running time of the script:** ( 1 minutes  11.605 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 aa028b952a..f7b418a7bb 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)  
-      120.6249     120.5219     127.1944     119.6546      0.9120   
+      115.6808     115.1310     137.7318     114.1419      2.4827   
                
 
 
@@ -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  26.571 seconds)
+   **Total running time of the script:** ( 2 minutes  25.778 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 4d92f06f17..816d309f91 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  37.738 seconds)
+   **Total running time of the script:** ( 1 minutes  33.040 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 187d4e57cb..0ac0bd6e91 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         | 6337/132723 [00:00<00:01, 63357.52KB/s]
     11%|#1        | 14856/132723 [00:00<00:01, 76193.71KB/s]
     18%|#7        | 23420/132723 [00:00<00:01, 80502.38KB/s]
     24%|##4       | 31993/132723 [00:00<00:01, 82561.00KB/s]
     31%|###       | 40595/132723 [00:00<00:01, 83805.51KB/s]
     37%|###7      | 49207/132723 [00:00<00:00, 84590.00KB/s]
     44%|####3     | 57821/132723 [00:00<00:00, 85095.20KB/s]
     50%|#####     | 66608/132723 [00:00<00:00, 85976.45KB/s]
     57%|#####6    | 75323/132723 [00:00<00:00, 86340.10KB/s]
     63%|######3   | 84000/132723 [00:01<00:00, 86470.42KB/s]
     70%|######9   | 92744/132723 [00:01<00:00, 86764.89KB/s]
     76%|#######6  | 101511/132723 [00:01<00:00, 87037.90KB/s]
     83%|########3 | 110234/132723 [00:01<00:00, 87093.79KB/s]
     90%|########9 | 118962/132723 [00:01<00:00, 87147.48KB/s]
     96%|#########6| 127677/132723 [00:01<00:00, 86950.69KB/s]
    100%|#######
 ###| 132723/132723 [00:01<00:00, 85021.87KB/s]
+
      0%|          | 0/132723 [00:00<?, ?KB/s]
      3%|3         | 4073/132723 [00:00<00:03, 40719.46KB/s]
      6%|6         | 8145/132723 [00:00<00:03, 39282.65KB/s]
      9%|9         | 12077/132723 [00:00<00:04, 27859.60KB/s]
     14%|#3        | 18379/132723 [00:00<00:02, 38827.23KB/s]
     20%|#9        | 26346/132723 [00:00<00:02, 51416.18KB/s]
     26%|##5       | 34373/132723 [00:00<00:01, 60233.61KB/s]
     32%|###1      | 42447/132723 [00:00<00:01, 66465.90KB/s]
     38%|###8      | 50483/132723 [00:00<00:01, 70670.72KB/s]
     44%|####4     | 58456/132723 [00:00<00:01, 73403.55KB/s]
     50%|#####     | 66501/132723 [00:01<00:00, 75524.61KB/s]
     56%|#####6    | 74502/132723 [00:01<00:00, 76873.15KB/s]
     62%|######2   | 82509/132723 [00:01<00:00, 77832.90KB/s]
     68%|######8   | 90497/132723 [00:01<00:00, 78446.62KB/s]
     74%|#######4  | 98389/132723 [00:01<00:00, 77460.44KB/s]
     80%|########  | 106341/132723 [00:01<00:00, 78066.73KB/s]
     86%|########6 |
  114285/132723 [00:01<00:00, 78474.33KB/s]
     92%|#########2| 122226/132723 [00:01<00:00, 78750.32KB/s]
     98%|#########8| 130146/132723 [00:01<00:00, 78880.08KB/s]
    100%|##########| 132723/132723 [00:01<00:00, 68565.13KB/s]
 
 
 
@@ -246,7 +246,7 @@ Display result
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 3 minutes  32.737 seconds)
+   **Total running time of the script:** ( 3 minutes  25.032 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 c12188b9cc..a2f3ff2a99 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
 =================
-**14:42.795** total execution time for **how_to_deploy_models** files:
+**14:31.509** total execution time for **how_to_deploy_models** files:
 
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``)                           | 03:32.737 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``)                           | 03:25.032 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_object_detection_pytorch.py` (``deploy_object_detection_pytorch.py``) | 03:24.271 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_object_detection_pytorch.py` (``deploy_object_detection_pytorch.py``) | 03:23.093 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized_tflite.py` (``deploy_prequantized_tflite.py``)           | 02:26.571 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized_tflite.py` (``deploy_prequantized_tflite.py``)           | 02:25.778 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_quantized.py` (``deploy_quantized.py``)                               | 01:37.738 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_quantized.py` (``deploy_quantized.py``)                               | 01:33.040 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized.py` (``deploy_prequantized.py``)                         | 01:11.765 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_prequantized.py` (``deploy_prequantized.py``)                         | 01:11.605 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_adreno.py` (``deploy_model_on_adreno.py``)                   | 00:55.410 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_adreno.py` (``deploy_model_on_adreno.py``)                   | 01:01.415 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_android.py` (``deploy_model_on_android.py``)                 | 00:40.553 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_android.py` (``deploy_model_on_android.py``)                 | 00:38.403 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_nano.py` (``deploy_model_on_nano.py``)                       | 00:27.139 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_nano.py` (``deploy_model_on_nano.py``)                       | 00:26.594 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``)                       | 00:26.605 | 0.0 MB |
+| :ref:`sphx_glr_how_to_deploy_models_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``)                       | 00:26.543 | 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 dc5fc13175..63cbedcf64 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.zipfa2710d0-5013-4fc8-b61b-36167b142c18 from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
+    Downloading /workspace/.mxnet/models/mobilenet0.25-9f83e440.zip7164de1d-d2c1-4340-a025-1b0504e11918 from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
 
 
 
diff --git a/docs/_sources/how_to/extend_tvm/low_level_custom_pass.rst.txt b/docs/_sources/how_to/extend_tvm/low_level_custom_pass.rst.txt
index a18f9cf4fa..35427c1612 100644
--- a/docs/_sources/how_to/extend_tvm/low_level_custom_pass.rst.txt
+++ b/docs/_sources/how_to/extend_tvm/low_level_custom_pass.rst.txt
@@ -90,18 +90,19 @@ our customized lowering pass to manipulate the IR directly instead of using sche
 
  .. code-block:: none
 
-    @main = primfn(a_1: handle, b_1: handle, c_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {a: Buffer(a_2: Pointer(float32), float32, [128], []),
-                 b: Buffer(b_2: Pointer(float32), float32, [128], []),
-                 c: Buffer(c_2: Pointer(float32), float32, [128], [])}
-      buffer_map = {a_1: a, b_1: b, c_1: c} {
-      for (i: int32, 0, 128) {
-        c_3: Buffer(c_2, float32, [128], [])[i] = (a_3: Buffer(a_2, float32, [128], [])[i] + b_3: Buffer(b_2, float32, [128], [])[i])
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(a: T.handle, b: T.handle, c: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            a_1 = T.match_buffer(a, (128,))
+            b_1 = T.match_buffer(b, (128,))
+            c_1 = T.match_buffer(c, (128,))
+            for i in range(128):
+                c_2 = T.buffer_decl((128,), data=c_1.data)
+                a_2 = T.buffer_decl((128,), data=a_1.data)
+                b_2 = T.buffer_decl((128,), data=b_1.data)
+                c_2[i] = a_2[i] + b_2[i]
 
 
 
@@ -249,19 +250,20 @@ Thus, a good place to put this transformation pass is just after Phase 1.
 
  .. code-block:: none
 
-    @main = primfn(a_1: handle, b_1: handle, c_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {a: Buffer(a_2: Pointer(float32), float32, [128], []),
-                 b: Buffer(b_2: Pointer(float32), float32, [128], []),
-                 c: Buffer(c_2: Pointer(float32), float32, [128], [])}
-      buffer_map = {a_1: a, b_1: b, c_1: c} {
-      for (i.outer: int32, 0, 16) {
-        let cse_var_1: int32 = (i.outer*8)
-        c_3: Buffer(c_2, float32, [128], [])[ramp(cse_var_1, 1, 8)] = (a_3: Buffer(a_2, float32, [128], [])[ramp(cse_var_1, 1, 8)] + b_3: Buffer(b_2, float32, [128], [])[ramp(cse_var_1, 1, 8)])
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(a: T.handle, b: T.handle, c: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            a_1 = T.match_buffer(a, (128,))
+            b_1 = T.match_buffer(b, (128,))
+            c_1 = T.match_buffer(c, (128,))
+            for i_outer in range(16):
+                cse_var_1: T.int32 = i_outer * 8
+                c_2 = T.buffer_decl((128,), data=c_1.data)
+                a_2 = T.buffer_decl((128,), data=a_1.data)
+                b_2 = T.buffer_decl((128,), data=b_1.data)
+                c_2[cse_var_1:cse_var_1 + 8] = a_2[cse_var_1:cse_var_1 + 8] + b_2[cse_var_1:cse_var_1 + 8]
 
 
 
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 78a2ccbc6e..bc34284b76 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
 =================
-**00:52.331** total execution time for **how_to_extend_tvm** files:
+**00:50.766** 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``) | 00:48.649 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_bring_your_own_datatypes.py` (``bring_your_own_datatypes.py``) | 00:47.175 | 0.0 MB |
 +-------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_extend_tvm_use_pass_instrument.py` (``use_pass_instrument.py``)           | 00:02.615 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_use_pass_instrument.py` (``use_pass_instrument.py``)           | 00:02.561 | 0.0 MB |
 +-------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_extend_tvm_use_pass_infra.py` (``use_pass_infra.py``)                     | 00:01.059 | 0.0 MB |
+| :ref:`sphx_glr_how_to_extend_tvm_use_pass_infra.py` (``use_pass_infra.py``)                     | 00:01.024 | 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 45354b4c6c..235f35225e 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: 17743us [17743us] (48.67%; 48.67%)
-    FoldScaleAxis: 18715us [7us] (51.33%; 51.33%)
-            FoldConstant: 18709us [1698us] (51.32%; 99.96%)
-                    InferType: 17011us [17011us] (46.66%; 90.93%)
+    InferType: 17873us [17873us] (48.73%; 48.73%)
+    FoldScaleAxis: 18801us [6us] (51.27%; 51.27%)
+            FoldConstant: 18794us [1685us] (51.25%; 99.97%)
+                    InferType: 17110us [17110us] (46.65%; 91.04%)
 
 
 
@@ -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: 17048us [17048us] (47.94%; 47.94%)
-    FoldScaleAxis: 18513us [5us] (52.06%; 52.06%)
-            FoldConstant: 18507us [1659us] (52.04%; 99.97%)
-                    InferType: 16848us [16848us] (47.38%; 91.04%)
+    InferType: 17834us [17834us] (48.85%; 48.85%)
+    FoldScaleAxis: 18671us [5us] (51.15%; 51.15%)
+            FoldConstant: 18667us [1666us] (51.13%; 99.98%)
+                    InferType: 17000us [17000us] (46.57%; 91.07%)
 
 
 
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 b88fb9964f..82add9b9ac 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: 54.142974 ms
+    Convolution: 54.136863 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 84e31b24c6..11daccc10b 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
@@ -432,101 +432,66 @@ one time.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, W_1: handle, Conv_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float16), float16, [16, 14, 14, 16, 16, 16], []),
-                 W: Buffer(W_2: Pointer(float16), float16, [3, 3, 16, 32, 16, 16], []),
-                 Conv: Buffer(Conv_2: Pointer(float32), float32, [16, 14, 14, 32, 16, 16], [])}
-      buffer_map = {A_1: A, W_1: W, Conv_1: Conv} {
-      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 196;
-      allocate(Conv.wmma.accumulator: Pointer(wmma.accumulator float32), float32, [2048]), storage_scope = wmma.accumulator;
-      allocate(Apad.shared: Pointer(shared float16), float16, [12288]), storage_scope = shared;
-      allocate(W.shared: Pointer(shared float16), float16, [12288]), storage_scope = shared;
-      allocate(Apad.shared.wmma.matrix_a: Pointer(wmma.matrix_a float16), float16, [512]), storage_scope = wmma.matrix_a;
-      allocate(W.shared.wmma.matrix_b: Pointer(wmma.matrix_b float16), float16, [1024]), storage_scope = wmma.matrix_b;
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
-      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 4;
-      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 4;
-      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 2 {
-        for (n.c.init: int32, 0, 2) {
-          for (o.c.init: int32, 0, 4) {
-            for (nn.c.init: int32, 0, 16) {
-              for (oo.c.init: int32, 0, 16) {
-                Conv.wmma.accumulator_1: Buffer(Conv.wmma.accumulator, float32, [2048], [], scope="wmma.accumulator")[((((n.c.init*1024) + (o.c.init*256)) + (nn.c.init*16)) + oo.c.init)] = 0f32
-              }
-            }
-          }
-        }
-        for (ic.outer: int32, 0, 8) {
-          for (kh: int32, 0, 3) {
-            for (ax2: int32, 0, 3) {
-              for (ax3: int32, 0, 2) {
-                for (ax4.ax5.fused.outer: int32, 0, 8) {
-                  let cse_var_2: int32 = (ax3*256)
-                  let cse_var_1: int32 = (ax4.ax5.fused.outer*32)
-                  attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
-                  Apad.shared_1: Buffer(Apad.shared, float16, [12288], [], scope="shared")[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + cse_var_2) + cse_var_1) + threadIdx.x)] = @tir.if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), A_3: Buffer(A_2, float16, [12845056], [])[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threa [...]
-                }
-              }
-            }
-            for (ax1: int32, 0, 3) {
-              for (ax2_1: int32, 0, 2) {
-                attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
-                W.shared_1: Buffer(W.shared, float16, [12288], [], scope="shared")[ramp((((((ax1*4096) + (ax2_1*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = W_3: Buffer(W_2, float16, [1179648], [])[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2_1*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)]
-              }
-            }
-            for (ic.inner: int32, 0, 2) {
-              for (kw: int32, 0, 3) {
-                for (ax0: int32, 0, 2) {
-                  for (ax4: int32, 0, 16) {
-                    for (ax5: int32, 0, 16) {
-                      let cse_var_3: int32 = (ax4*16)
-                      Apad.shared.wmma.matrix_a_1: Buffer(Apad.shared.wmma.matrix_a, float16, [512], [], scope="wmma.matrix_a")[(((ax0*256) + cse_var_3) + ax5)] = Apad.shared_1[((((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)) + cse_var_3) + ax5)]
-                    }
-                  }
-                }
-                for (ax3_1: int32, 0, 4) {
-                  for (ax4_1: int32, 0, 16) {
-                    for (ax5_1: int32, 0, 16) {
-                      let cse_var_5: int32 = (ax3_1*256)
-                      let cse_var_4: int32 = (ax4_1*16)
-                      W.shared.wmma.matrix_b_1: Buffer(W.shared.wmma.matrix_b, float16, [1024], [], scope="wmma.matrix_b")[((cse_var_5 + cse_var_4) + ax5_1)] = W.shared_1[((((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + cse_var_5) + cse_var_4) + ax5_1)]
-                    }
-                  }
-                }
-                for (n.c: int32, 0, 2) {
-                  for (o.c: int32, 0, 4) {
-                    for (nn.c: int32, 0, 16) {
-                      for (oo.c: int32, 0, 16) {
-                        for (ii: int32, 0, 16) {
-                          let cse_var_8: int32 = (o.c*256)
-                          let cse_var_7: int32 = (nn.c*16)
-                          let cse_var_6: int32 = ((((n.c*1024) + cse_var_8) + cse_var_7) + oo.c)
-                          Conv.wmma.accumulator_1[cse_var_6] = (Conv.wmma.accumulator_1[cse_var_6] + (cast(float32, Apad.shared.wmma.matrix_a_1[(((n.c*256) + cse_var_7) + ii)])*cast(float32, W.shared.wmma.matrix_b_1[((cse_var_8 + (ii*16)) + oo.c)])))
-                        }
-                      }
-                    }
-                  }
-                }
-              }
-            }
-          }
-        }
-        for (n.inner: int32, 0, 2) {
-          for (o.inner: int32, 0, 4) {
-            for (nn: int32, 0, 16) {
-              for (oo: int32, 0, 16) {
-                let cse_var_10: int32 = (o.inner*256)
-                let cse_var_9: int32 = (nn*16)
-                Conv_3: Buffer(Conv_2, float32, [25690112], [])[(((((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + cse_var_10) + cse_var_9) + oo)] = Conv.wmma.accumulator_1[((((n.inner*1024) + cse_var_10) + cse_var_9) + oo)]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, W: T.handle, Conv: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (16, 14, 14, 16, 16, 16), "float16")
+            W_1 = T.match_buffer(W, (3, 3, 16, 32, 16, 16), "float16")
+            Conv_1 = T.match_buffer(Conv, (16, 14, 14, 32, 16, 16))
+            blockIdx_z = T.env_thread("blockIdx.z")
+            T.launch_thread(blockIdx_z, 196)
+            Conv_wmma_accumulator = T.allocate([2048], "float32", "wmma.accumulator")
+            Apad_shared = T.allocate([12288], "float16", "shared")
+            W_shared = T.allocate([12288], "float16", "shared")
+            Apad_shared_wmma_matrix_a = T.allocate([512], "float16", "wmma.matrix_a")
+            W_shared_wmma_matrix_b = T.allocate([1024], "float16", "wmma.matrix_b")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, 2)
+            blockIdx_y = T.env_thread("blockIdx.y")
+            T.launch_thread(blockIdx_y, 4)
+            threadIdx_y = T.env_thread("threadIdx.y")
+            T.launch_thread(threadIdx_y, 4)
+            threadIdx_z = T.env_thread("threadIdx.z")
+            T.launch_thread(threadIdx_z, 2)
+            Conv_wmma_accumulator_1 = T.buffer_decl((2048,), data=Conv_wmma_accumulator, scope="wmma.accumulator")
+            for n_c_init, o_c_init, nn_c_init, oo_c_init in T.grid(2, 4, 16, 16):
+                Conv_wmma_accumulator_1[n_c_init * 1024 + o_c_init * 256 + nn_c_init * 16 + oo_c_init] = T.float32(0)
+            for ic_outer, kh in T.grid(8, 3):
+                threadIdx_x = T.env_thread("threadIdx.x")
+                Apad_shared_1 = T.buffer_decl((12288,), "float16", data=Apad_shared, scope="shared")
+                for ax2, ax3, ax4_ax5_fused_outer in T.grid(3, 2, 8):
+                    cse_var_2: T.int32 = ax3 * 256
+                    cse_var_1: T.int32 = ax4_ax5_fused_outer * 32
+                    T.launch_thread(threadIdx_x, 32)
+                    A_2 = T.buffer_decl((12845056,), "float16", data=A_1.data)
+                    Apad_shared_1[threadIdx_y * 3072 + threadIdx_z * 1536 + ax2 * 512 + cse_var_2 + cse_var_1 + threadIdx_x] = T.if_then_else(1 <= blockIdx_z // 14 + kh and blockIdx_z // 14 + kh < 15 and 1 <= ax2 + blockIdx_z % 14 and ax2 + blockIdx_z % 14 < 15, A_2[blockIdx_x * 6422528 + threadIdx_y * 1605632 + threadIdx_z * 802816 + kh * 57344 + blockIdx_z * 4096 + ax2 * 4096 + ic_outer * 512 + cse_var_2 + cse_var_1 + threadIdx_x - 61440], T.float16(0))
+                W_shared_1 = T.buffer_decl((12288,), "float16", data=W_shared, scope="shared")
+                for ax1, ax2 in T.grid(3, 2):
+                    T.launch_thread(threadIdx_x, 32)
+                    W_2 = T.buffer_decl((1179648,), "float16", data=W_1.data)
+                    W_shared_1[ax1 * 4096 + ax2 * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8:ax1 * 4096 + ax2 * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8 + 8] = W_2[kh * 393216 + ax1 * 131072 + ic_outer * 16384 + ax2 * 8192 + blockIdx_y * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8:kh * 393216 + ax1 * 131072 + ic_outer * 16384 + ax2 * 8192 + blockIdx_y * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8 + 8]
+                for ic_inner, kw in T.grid(2, 3):
+                    Apad_shared_wmma_matrix_a_1 = T.buffer_decl((512,), "float16", data=Apad_shared_wmma_matrix_a, scope="wmma.matrix_a")
+                    for ax0, ax4, ax5 in T.grid(2, 16, 16):
+                        cse_var_3: T.int32 = ax4 * 16
+                        Apad_shared_wmma_matrix_a_1[ax0 * 256 + cse_var_3 + ax5] = Apad_shared_1[threadIdx_y * 3072 + ax0 * 1536 + kw * 512 + ic_inner * 256 + cse_var_3 + ax5]
+                    W_shared_wmma_matrix_b_1 = T.buffer_decl((1024,), "float16", data=W_shared_wmma_matrix_b, scope="wmma.matrix_b")
+                    for ax3, ax4, ax5 in T.grid(4, 16, 16):
+                        cse_var_5: T.int32 = ax3 * 256
+                        cse_var_4: T.int32 = ax4 * 16
+                        W_shared_wmma_matrix_b_1[cse_var_5 + cse_var_4 + ax5] = W_shared_1[kw * 4096 + ic_inner * 2048 + threadIdx_z * 1024 + cse_var_5 + cse_var_4 + ax5]
+                    for n_c, o_c, nn_c, oo_c, ii in T.grid(2, 4, 16, 16, 16):
+                        cse_var_8: T.int32 = o_c * 256
+                        cse_var_7: T.int32 = nn_c * 16
+                        cse_var_6: T.int32 = n_c * 1024 + cse_var_8 + cse_var_7 + oo_c
+                        Conv_wmma_accumulator_1[cse_var_6] = Conv_wmma_accumulator_1[cse_var_6] + T.Cast("float32", Apad_shared_wmma_matrix_a_1[n_c * 256 + cse_var_7 + ii]) * T.Cast("float32", W_shared_wmma_matrix_b_1[cse_var_8 + ii * 16 + oo_c])
+            for n_inner, o_inner, nn, oo in T.grid(2, 4, 16, 16):
+                cse_var_10: T.int32 = o_inner * 256
+                cse_var_9: T.int32 = nn * 16
+                Conv_2 = T.buffer_decl((25690112,), data=Conv_1.data)
+                Conv_2[blockIdx_x * 12845056 + threadIdx_y * 3211264 + n_inner * 1605632 + blockIdx_z * 8192 + blockIdx_y * 2048 + threadIdx_z * 1024 + cse_var_10 + cse_var_9 + oo] = Conv_wmma_accumulator_1[n_inner * 1024 + cse_var_10 + cse_var_9 + oo]
 
 
 
@@ -557,72 +522,55 @@ by mapping the 2D convolution to tensor intrinsics
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, W_1: handle, Conv_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float16), float16, [16, 14, 14, 16, 16, 16], []),
-                 W: Buffer(W_2: Pointer(float16), float16, [3, 3, 16, 32, 16, 16], []),
-                 Conv: Buffer(Conv_2: Pointer(float32), float32, [16, 14, 14, 32, 16, 16], [])}
-      buffer_map = {A_1: A, W_1: W, Conv_1: Conv} {
-      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 196;
-      allocate(Conv.wmma.accumulator: Pointer(wmma.accumulator float32), float32, [2048]), storage_scope = wmma.accumulator;
-      allocate(Apad.shared: Pointer(shared float16), float16, [12288]), storage_scope = shared;
-      allocate(W.shared: Pointer(shared float16), float16, [12288]), storage_scope = shared;
-      allocate(Apad.shared.wmma.matrix_a: Pointer(wmma.matrix_a float16), float16, [512]), storage_scope = wmma.matrix_a;
-      allocate(W.shared.wmma.matrix_b: Pointer(wmma.matrix_b float16), float16, [1024]), storage_scope = wmma.matrix_b;
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
-      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 4;
-      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 4;
-      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 2 {
-        for (n.c.init: int32, 0, 2) {
-          for (o.c.init: int32, 0, 4) {
-            @tir.tvm_fill_fragment(Conv.wmma.accumulator, 16, 16, 16, ((n.c.init*4) + o.c.init), 0f32, dtype=handle)
-          }
-        }
-        for (ic.outer: int32, 0, 8) {
-          for (kh: int32, 0, 3) {
-            for (ax2: int32, 0, 3) {
-              for (ax3: int32, 0, 2) {
-                for (ax4.ax5.fused.outer: int32, 0, 8) {
-                  let cse_var_2: int32 = (ax3*256)
-                  let cse_var_1: int32 = (ax4.ax5.fused.outer*32)
-                  attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
-                  Apad.shared_1: Buffer(Apad.shared, float16, [12288], [], scope="shared")[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + cse_var_2) + cse_var_1) + threadIdx.x)] = @tir.if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), A_3: Buffer(A_2, float16, [12845056], [])[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threa [...]
-                }
-              }
-            }
-            for (ax1: int32, 0, 3) {
-              for (ax2_1: int32, 0, 2) {
-                attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
-                W.shared_1: Buffer(W.shared, float16, [12288], [], scope="shared")[ramp((((((ax1*4096) + (ax2_1*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = W_3: Buffer(W_2, float16, [1179648], [])[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2_1*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)]
-              }
-            }
-            for (ic.inner: int32, 0, 2) {
-              for (kw: int32, 0, 3) {
-                for (ax0: int32, 0, 2) {
-                  @tir.tvm_load_matrix_sync(Apad.shared.wmma.matrix_a, 16, 16, 16, ax0, @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float16), Apad.shared, ((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)), 256, 1, dtype=handle), 16, "row_major", dtype=handle)
-                }
-                for (ax3_1: int32, 0, 4) {
-                  @tir.tvm_load_matrix_sync(W.shared.wmma.matrix_b, 16, 16, 16, ax3_1, @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float16), W.shared, ((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + (ax3_1*256)), 256, 1, dtype=handle), 16, "row_major", dtype=handle)
-                }
-                for (n.c: int32, 0, 2) {
-                  for (o.c: int32, 0, 4) {
-                    let cse_var_3: int32 = ((n.c*4) + o.c)
-                    @tir.tvm_mma_sync(Conv.wmma.accumulator, cse_var_3, Apad.shared.wmma.matrix_a, n.c, W.shared.wmma.matrix_b, o.c, Conv.wmma.accumulator, cse_var_3, dtype=handle)
-                  }
-                }
-              }
-            }
-          }
-        }
-        for (n.inner: int32, 0, 2) {
-          for (o.inner: int32, 0, 4) {
-            @tir.tvm_store_matrix_sync(Conv.wmma.accumulator, 16, 16, 16, ((n.inner*4) + o.inner), @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), Conv_2, (((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + (o.inner*256)), 256, 2, dtype=handle), 16, "row_major", dtype=handle)
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, W: T.handle, Conv: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (16, 14, 14, 16, 16, 16), "float16")
+            W_1 = T.match_buffer(W, (3, 3, 16, 32, 16, 16), "float16")
+            Conv_1 = T.match_buffer(Conv, (16, 14, 14, 32, 16, 16))
+            blockIdx_z = T.env_thread("blockIdx.z")
+            T.launch_thread(blockIdx_z, 196)
+            Conv_wmma_accumulator = T.allocate([2048], "float32", "wmma.accumulator")
+            Apad_shared = T.allocate([12288], "float16", "shared")
+            W_shared = T.allocate([12288], "float16", "shared")
+            Apad_shared_wmma_matrix_a = T.allocate([512], "float16", "wmma.matrix_a")
+            W_shared_wmma_matrix_b = T.allocate([1024], "float16", "wmma.matrix_b")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, 2)
+            blockIdx_y = T.env_thread("blockIdx.y")
+            T.launch_thread(blockIdx_y, 4)
+            threadIdx_y = T.env_thread("threadIdx.y")
+            T.launch_thread(threadIdx_y, 4)
+            threadIdx_z = T.env_thread("threadIdx.z")
+            T.launch_thread(threadIdx_z, 2)
+            for n_c_init, o_c_init in T.grid(2, 4):
+                T.tvm_fill_fragment(Conv_wmma_accumulator, 16, 16, 16, n_c_init * 4 + o_c_init, T.float32(0))
+            for ic_outer, kh in T.grid(8, 3):
+                threadIdx_x = T.env_thread("threadIdx.x")
+                for ax2, ax3, ax4_ax5_fused_outer in T.grid(3, 2, 8):
+                    cse_var_2: T.int32 = ax3 * 256
+                    cse_var_1: T.int32 = ax4_ax5_fused_outer * 32
+                    T.launch_thread(threadIdx_x, 32)
+                    Apad_shared_1 = T.buffer_decl((12288,), "float16", data=Apad_shared, scope="shared")
+                    A_2 = T.buffer_decl((12845056,), "float16", data=A_1.data)
+                    Apad_shared_1[threadIdx_y * 3072 + threadIdx_z * 1536 + ax2 * 512 + cse_var_2 + cse_var_1 + threadIdx_x] = T.if_then_else(1 <= blockIdx_z // 14 + kh and blockIdx_z // 14 + kh < 15 and 1 <= ax2 + blockIdx_z % 14 and ax2 + blockIdx_z % 14 < 15, A_2[blockIdx_x * 6422528 + threadIdx_y * 1605632 + threadIdx_z * 802816 + kh * 57344 + blockIdx_z * 4096 + ax2 * 4096 + ic_outer * 512 + cse_var_2 + cse_var_1 + threadIdx_x - 61440], T.float16(0))
+                for ax1, ax2 in T.grid(3, 2):
+                    T.launch_thread(threadIdx_x, 32)
+                    W_shared_1 = T.buffer_decl((12288,), "float16", data=W_shared, scope="shared")
+                    W_2 = T.buffer_decl((1179648,), "float16", data=W_1.data)
+                    W_shared_1[ax1 * 4096 + ax2 * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8:ax1 * 4096 + ax2 * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8 + 8] = W_2[kh * 393216 + ax1 * 131072 + ic_outer * 16384 + ax2 * 8192 + blockIdx_y * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8:kh * 393216 + ax1 * 131072 + ic_outer * 16384 + ax2 * 8192 + blockIdx_y * 2048 + threadIdx_y * 512 + threadIdx_z * 256 + threadIdx_x * 8 + 8]
+                for ic_inner, kw in T.grid(2, 3):
+                    for ax0 in range(2):
+                        T.tvm_load_matrix_sync(Apad_shared_wmma_matrix_a, 16, 16, 16, ax0, T.tvm_access_ptr(T.type_annotation("float16"), Apad_shared, threadIdx_y * 3072 + ax0 * 1536 + kw * 512 + ic_inner * 256, 256, 1), 16, "row_major")
+                    for ax3 in range(4):
+                        T.tvm_load_matrix_sync(W_shared_wmma_matrix_b, 16, 16, 16, ax3, T.tvm_access_ptr(T.type_annotation("float16"), W_shared, kw * 4096 + ic_inner * 2048 + threadIdx_z * 1024 + ax3 * 256, 256, 1), 16, "row_major")
+                    for n_c, o_c in T.grid(2, 4):
+                        cse_var_3: T.int32 = n_c * 4 + o_c
+                        T.tvm_mma_sync(Conv_wmma_accumulator, cse_var_3, Apad_shared_wmma_matrix_a, n_c, W_shared_wmma_matrix_b, o_c, Conv_wmma_accumulator, cse_var_3)
+            for n_inner, o_inner in T.grid(2, 4):
+                T.tvm_store_matrix_sync(Conv_wmma_accumulator, 16, 16, 16, n_inner * 4 + o_inner, T.tvm_access_ptr(T.type_annotation("float32"), Conv_1.data, blockIdx_x * 12845056 + threadIdx_y * 3211264 + n_inner * 1605632 + blockIdx_z * 8192 + blockIdx_y * 2048 + threadIdx_z * 1024 + o_inner * 256, 256, 2), 16, "row_major")
 
 
 
@@ -660,7 +608,7 @@ be able to run on our build server
 
  .. code-block:: none
 
-    conv2d with tensor core: 7.462708 ms
+    conv2d with tensor core: 6.841651 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 3341b21aee..835a54c44b 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: 0.017961
-    Baseline: 3.258530
+    Numpy running time: 0.018406
+    Baseline: 3.286550
 
 
 
@@ -160,25 +160,23 @@ Here is the generated IR using our baseline schedule.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (m: int32, 0, 1024) {
-        for (n: int32, 0, 1024) {
-          C_3: Buffer(C_2, float32, [1048576], [])[((m*1024) + n)] = 0f32
-          for (k: int32, 0, 1024) {
-            let cse_var_2: int32 = (m*1024)
-            let cse_var_1: int32 = (cse_var_2 + n)
-            C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [1048576], [])[(cse_var_2 + k)]*B_3: Buffer(B_2, float32, [1048576], [])[((k*1024) + n)]))
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for m, n in T.grid(1024, 1024):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                C_2[m * 1024 + n] = T.float32(0)
+                for k in range(1024):
+                    cse_var_2: T.int32 = m * 1024
+                    cse_var_1: T.int32 = cse_var_2 + n
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[cse_var_2 + k] * B_2[k * 1024 + n]
 
 
 
@@ -229,7 +227,7 @@ fill 32 * 32 * sizeof(float) which is 4KB in the cache whose total size is 32KB
 
  .. code-block:: none
 
-    Opt1: 0.299520
+    Opt1: 0.298701
 
 
 
@@ -253,36 +251,25 @@ Here is the generated IR after blocking.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (m.outer: int32, 0, 32) {
-        for (n.outer: int32, 0, 32) {
-          for (m.inner.init: int32, 0, 32) {
-            for (n.inner.init: int32, 0, 32) {
-              C_3: Buffer(C_2, float32, [1048576], [])[((((m.outer*32768) + (m.inner.init*1024)) + (n.outer*32)) + n.inner.init)] = 0f32
-            }
-          }
-          for (k.outer: int32, 0, 256) {
-            for (k.inner: int32, 0, 4) {
-              for (m.inner: int32, 0, 32) {
-                for (n.inner: int32, 0, 32) {
-                  let cse_var_3: int32 = (n.outer*32)
-                  let cse_var_2: int32 = ((m.outer*32768) + (m.inner*1024))
-                  let cse_var_1: int32 = ((cse_var_2 + cse_var_3) + n.inner)
-                  C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)]*B_3: Buffer(B_2, float32, [1048576], [])[((((k.outer*4096) + (k.inner*1024)) + cse_var_3) + n.inner)]))
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for m_outer, n_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for m_inner_init, n_inner_init in T.grid(32, 32):
+                    C_2[m_outer * 32768 + m_inner_init * 1024 + n_outer * 32 + n_inner_init] = T.float32(0)
+                for k_outer, k_inner, m_inner, n_inner in T.grid(256, 4, 32, 32):
+                    cse_var_3: T.int32 = n_outer * 32
+                    cse_var_2: T.int32 = m_outer * 32768 + m_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3 + n_inner
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[cse_var_2 + k_outer * 4 + k_inner] * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3 + n_inner]
 
 
 
@@ -331,7 +318,7 @@ In this tutorial, we chose to vectorize the inner loop row data since it is cach
 
  .. code-block:: none
 
-    Opt2: 0.336287
+    Opt2: 0.334528
 
 
 
@@ -355,32 +342,25 @@ Here is the generated IR after vectorization.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (m.outer: int32, 0, 32) {
-        for (n.outer: int32, 0, 32) {
-          for (m.inner.init: int32, 0, 32) {
-            C_3: Buffer(C_2, float32, [1048576], [])[ramp((((m.outer*32768) + (m.inner.init*1024)) + (n.outer*32)), 1, 32)] = broadcast(0f32, 32)
-          }
-          for (k.outer: int32, 0, 256) {
-            for (k.inner: int32, 0, 4) {
-              for (m.inner: int32, 0, 32) {
-                let cse_var_3: int32 = (n.outer*32)
-                let cse_var_2: int32 = ((m.outer*32768) + (m.inner*1024))
-                let cse_var_1: int32 = (cse_var_2 + cse_var_3)
-                C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)], 32)*B_3: Buffer(B_2, float32, [1048576], [])[ramp((((k.outer*4096) + (k.inner*1024)) + cse_var_3), 1, 32)]))
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for m_outer, n_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for m_inner_init in range(32):
+                    C_2[m_outer * 32768 + m_inner_init * 1024 + n_outer * 32:m_outer * 32768 + m_inner_init * 1024 + n_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, k_inner, m_inner in T.grid(256, 4, 32):
+                    cse_var_3: T.int32 = n_outer * 32
+                    cse_var_2: T.int32 = m_outer * 32768 + m_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_2 + k_outer * 4 + k_inner], 32) * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3:k_outer * 4096 + k_inner * 1024 + cse_var_3 + 32]
 
 
 
@@ -426,7 +406,7 @@ the access pattern for A matrix is more cache friendly.
 
  .. code-block:: none
 
-    Opt3: 0.116361
+    Opt3: 0.112976
 
 
 
@@ -450,32 +430,25 @@ Here is the generated IR after loop permutation.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (m.outer: int32, 0, 32) {
-        for (n.outer: int32, 0, 32) {
-          for (m.inner.init: int32, 0, 32) {
-            C_3: Buffer(C_2, float32, [1048576], [])[ramp((((m.outer*32768) + (m.inner.init*1024)) + (n.outer*32)), 1, 32)] = broadcast(0f32, 32)
-          }
-          for (k.outer: int32, 0, 256) {
-            for (m.inner: int32, 0, 32) {
-              for (k.inner: int32, 0, 4) {
-                let cse_var_3: int32 = (n.outer*32)
-                let cse_var_2: int32 = ((m.outer*32768) + (m.inner*1024))
-                let cse_var_1: int32 = (cse_var_2 + cse_var_3)
-                C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)], 32)*B_3: Buffer(B_2, float32, [1048576], [])[ramp((((k.outer*4096) + (k.inner*1024)) + cse_var_3), 1, 32)]))
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for m_outer, n_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for m_inner_init in range(32):
+                    C_2[m_outer * 32768 + m_inner_init * 1024 + n_outer * 32:m_outer * 32768 + m_inner_init * 1024 + n_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, m_inner, k_inner in T.grid(256, 32, 4):
+                    cse_var_3: T.int32 = n_outer * 32
+                    cse_var_2: T.int32 = m_outer * 32768 + m_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_2 + k_outer * 4 + k_inner], 32) * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3:k_outer * 4096 + k_inner * 1024 + cse_var_3 + 32]
 
 
 
@@ -550,7 +523,7 @@ flattening.
 
  .. code-block:: none
 
-    Opt4: 0.109872
+    Opt4: 0.108978
 
 
 
@@ -574,39 +547,30 @@ Here is the generated IR after array packing.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global {
-        for (bigN: int32, 0, 32) "parallel" {
-          for (k: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((bigN*1024) + k)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((k*1024) + (bigN*32)), 1, 32)]
-          }
-        }
-        for (m.outer: int32, 0, 32) {
-          for (n.outer: int32, 0, 32) {
-            for (m.inner.init: int32, 0, 32) {
-              C_3: Buffer(C_2, float32, [1048576], [])[ramp((((m.outer*32768) + (m.inner.init*1024)) + (n.outer*32)), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (m.inner: int32, 0, 32) {
-                for (k.inner: int32, 0, 4) {
-                  let cse_var_3: int32 = ((m.outer*32768) + (m.inner*1024))
-                  let cse_var_2: int32 = (k.outer*4)
-                  let cse_var_1: int32 = (cse_var_3 + (n.outer*32))
-                  C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_3 + cse_var_2) + k.inner)], 32)*packedB_1[(((n.outer*1024) + cse_var_2) + k.inner)]))
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for bigN in T.parallel(32):
+                for k in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[bigN * 1024 + k] = B_2[k * 1024 + bigN * 32:k * 1024 + bigN * 32 + 32]
+            for m_outer, n_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for m_inner_init in range(32):
+                    C_2[m_outer * 32768 + m_inner_init * 1024 + n_outer * 32:m_outer * 32768 + m_inner_init * 1024 + n_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, m_inner, k_inner in T.grid(256, 32, 4):
+                    cse_var_3: T.int32 = m_outer * 32768 + m_inner * 1024
+                    cse_var_2: T.int32 = k_outer * 4
+                    cse_var_1: T.int32 = cse_var_3 + n_outer * 32
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_3 + cse_var_2 + k_inner], 32) * packedB_1[n_outer * 1024 + cse_var_2 + k_inner]
 
 
 
@@ -671,7 +635,7 @@ write to C when all the block results are ready.
 
  .. code-block:: none
 
-    Opt5: 0.110848
+    Opt5: 0.111004
 
 
 
@@ -695,49 +659,38 @@ Here is the generated IR after blocking.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global;
-      allocate(C.global: Pointer(global float32), float32, [1024]), storage_scope = global {
-        for (bigN: int32, 0, 32) "parallel" {
-          for (k: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((bigN*1024) + k)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((k*1024) + (bigN*32)), 1, 32)]
-          }
-        }
-        for (m.outer: int32, 0, 32) {
-          for (n.outer: int32, 0, 32) {
-            for (m.c.init: int32, 0, 32) {
-              C.global_1: Buffer(C.global, float32, [1024], [])[ramp((m.c.init*32), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (m.c: int32, 0, 32) {
-                let cse_var_4: int32 = (k.outer*4)
-                let cse_var_3: int32 = (m.c*32)
-                let cse_var_2: int32 = ((n.outer*1024) + cse_var_4)
-                let cse_var_1: int32 = (((m.outer*32768) + (m.c*1024)) + cse_var_4)
-                 {
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[cse_var_1], 32)*packedB_1[cse_var_2]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 1)], 32)*packedB_1[(cse_var_2 + 1)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 2)], 32)*packedB_1[(cse_var_2 + 2)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 3)], 32)*packedB_1[(cse_var_2 + 3)]))
-                }
-              }
-            }
-            for (m.inner: int32, 0, 32) {
-              for (n.inner: int32, 0, 32) {
-                C_3: Buffer(C_2, float32, [1048576], [])[((((m.outer*32768) + (m.inner*1024)) + (n.outer*32)) + n.inner)] = C.global_1[((m.inner*32) + n.inner)]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            C_global = T.allocate([1024], "float32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for bigN in T.parallel(32):
+                for k in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[bigN * 1024 + k] = B_2[k * 1024 + bigN * 32:k * 1024 + bigN * 32 + 32]
+            for m_outer, n_outer in T.grid(32, 32):
+                C_global_1 = T.buffer_decl((1024,), data=C_global)
+                for m_c_init in range(32):
+                    C_global_1[m_c_init * 32:m_c_init * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, m_c in T.grid(256, 32):
+                    cse_var_4: T.int32 = k_outer * 4
+                    cse_var_3: T.int32 = m_c * 32
+                    cse_var_2: T.int32 = n_outer * 1024 + cse_var_4
+                    cse_var_1: T.int32 = m_outer * 32768 + m_c * 1024 + cse_var_4
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1], 32) * packedB_1[cse_var_2]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 1], 32) * packedB_1[cse_var_2 + 1]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 2], 32) * packedB_1[cse_var_2 + 2]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 3], 32) * packedB_1[cse_var_2 + 3]
+                for m_inner, n_inner in T.grid(32, 32):
+                    C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                    C_2[m_outer * 32768 + m_inner * 1024 + n_outer * 32 + n_inner] = C_global_1[m_inner * 32 + n_inner]
 
 
 
@@ -795,7 +748,7 @@ Furthermore, we can also utilize multi-core processors to do the thread-level pa
 
  .. code-block:: none
 
-    Opt6: 0.146714
+    Opt6: 0.146827
 
 
 
@@ -819,49 +772,39 @@ Here is the generated IR after parallelization.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global {
-        for (bigN: int32, 0, 32) "parallel" {
-          for (k: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((bigN*1024) + k)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((k*1024) + (bigN*32)), 1, 32)]
-          }
-        }
-        for (m.outer: int32, 0, 32) "parallel" {
-          allocate(C.global: Pointer(global float32), float32, [1024]), storage_scope = global;
-          for (n.outer: int32, 0, 32) {
-            for (m.c.init: int32, 0, 32) {
-              C.global_1: Buffer(C.global, float32, [1024], [])[ramp((m.c.init*32), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (m.c: int32, 0, 32) {
-                let cse_var_4: int32 = (k.outer*4)
-                let cse_var_3: int32 = (m.c*32)
-                let cse_var_2: int32 = ((n.outer*1024) + cse_var_4)
-                let cse_var_1: int32 = (((m.outer*32768) + (m.c*1024)) + cse_var_4)
-                 {
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[cse_var_1], 32)*packedB_1[cse_var_2]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 1)], 32)*packedB_1[(cse_var_2 + 1)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 2)], 32)*packedB_1[(cse_var_2 + 2)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 3)], 32)*packedB_1[(cse_var_2 + 3)]))
-                }
-              }
-            }
-            for (m.inner: int32, 0, 32) {
-              for (n.inner: int32, 0, 32) {
-                C_3: Buffer(C_2, float32, [1048576], [])[((((m.outer*32768) + (m.inner*1024)) + (n.outer*32)) + n.inner)] = C.global_1[((m.inner*32) + n.inner)]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for bigN in T.parallel(32):
+                for k in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[bigN * 1024 + k] = B_2[k * 1024 + bigN * 32:k * 1024 + bigN * 32 + 32]
+            for m_outer in T.parallel(32):
+                C_global = T.allocate([1024], "float32", "global")
+                for n_outer in range(32):
+                    C_global_1 = T.buffer_decl((1024,), data=C_global)
+                    for m_c_init in range(32):
+                        C_global_1[m_c_init * 32:m_c_init * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                    for k_outer, m_c in T.grid(256, 32):
+                        cse_var_4: T.int32 = k_outer * 4
+                        cse_var_3: T.int32 = m_c * 32
+                        cse_var_2: T.int32 = n_outer * 1024 + cse_var_4
+                        cse_var_1: T.int32 = m_outer * 32768 + m_c * 1024 + cse_var_4
+                        A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1], 32) * packedB_1[cse_var_2]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 1], 32) * packedB_1[cse_var_2 + 1]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 2], 32) * packedB_1[cse_var_2 + 2]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 3], 32) * packedB_1[cse_var_2 + 3]
+                    for m_inner, n_inner in T.grid(32, 32):
+                        C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                        C_2[m_outer * 32768 + m_inner * 1024 + n_outer * 32 + n_inner] = C_global_1[m_inner * 32 + n_inner]
 
 
 
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 8c1d3c32a2..7d6dc6b6bf 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
 =================
-**00:34.295** total execution time for **how_to_optimize_operators** files:
+**00:34.492** total execution time for **how_to_optimize_operators** files:
 
 +-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_gemm.py` (``opt_gemm.py``)                       | 00:31.770 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_gemm.py` (``opt_gemm.py``)                       | 00:31.893 | 0.0 MB |
 +-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``) | 00:01.438 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``) | 00:01.454 | 0.0 MB |
 +-----------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_cuda.py` (``opt_conv_cuda.py``)             | 00:01.087 | 0.0 MB |
+| :ref:`sphx_glr_how_to_optimize_operators_opt_conv_cuda.py` (``opt_conv_cuda.py``)             | 00:01.145 | 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 53ef86cc02..352b5ad9fd 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
 =================
-**09:09.515** total execution time for **how_to_tune_with_autoscheduler** files:
+**09:27.678** 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``) | 05:33.867 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_conv2d_layer_cuda.py` (``tune_conv2d_layer_cuda.py``) | 05:38.572 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_x86.py` (``tune_network_x86.py``)             | 01:37.080 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_x86.py` (``tune_network_x86.py``)             | 01:36.142 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_cuda.py` (``tune_network_cuda.py``)           | 01:04.265 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_cuda.py` (``tune_network_cuda.py``)           | 01:05.307 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_sparse_x86.py` (``tune_sparse_x86.py``)               | 00:28.205 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_sparse_x86.py` (``tune_sparse_x86.py``)               | 00:41.139 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_arm.py` (``tune_network_arm.py``)             | 00:13.619 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_arm.py` (``tune_network_arm.py``)             | 00:13.861 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_mali.py` (``tune_network_mali.py``)           | 00:12.479 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autoscheduler_tune_network_mali.py` (``tune_network_mali.py``)           | 00:12.656 | 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 73e37b8a79..1bcf9043e0 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
@@ -235,232 +235,491 @@ cooperative fetching, unrolling and operator fusion.
  .. code-block:: none
 
     Lowered TIR:
-    @main = primfn(data_1: handle, kernel_1: handle, bias_1: handle, compute_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(float32), float32, [1, 512, 7, 7], []),
-                 kernel: Buffer(kernel_2: Pointer(float32), float32, [512, 512, 3, 3], []),
-                 bias: Buffer(bias_2: Pointer(float32), float32, [1, 512, 1, 1], []),
-                 compute: Buffer(compute_2: Pointer(float32), float32, [1, 512, 7, 7], [])}
-      buffer_map = {data_1: data, kernel_1: kernel, bias_1: bias, compute_1: compute} {
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 32;
-      allocate(conv2d_nchw: Pointer(local float32), float32, [7]), storage_scope = local;
-      allocate(pad_temp.shared: Pointer(shared float32), float32, [504]), storage_scope = shared;
-      allocate(kernel.shared: Pointer(shared float32), float32, [384]), storage_scope = shared;
-      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112 {
-        conv2d_nchw_1: Buffer(conv2d_nchw, float32, [7], [], scope="local", align=16)[0] = 0f32
-        conv2d_nchw_1[1] = 0f32
-        conv2d_nchw_1[2] = 0f32
-        conv2d_nchw_1[3] = 0f32
-        conv2d_nchw_1[4] = 0f32
-        conv2d_nchw_1[5] = 0f32
-        conv2d_nchw_1[6] = 0f32
-        for (rc.outer.outer: int32, 0, 64) {
-          for (ry.outer.outer: int32, 0, 3) {
-            let cse_var_4: int32 = (rc.outer.outer*392)
-            let cse_var_3: int32 = (ry.outer.outer*7)
-            let cse_var_2: int32 = (rc.outer.outer*72)
-            let cse_var_1: int32 = (ry.outer.outer*3)
-             {
-              attr [IterVar(threadIdx.x_1: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              pad_temp.shared_1: Buffer(pad_temp.shared, float32, [504], [], scope="shared")[threadIdx.x_1] = @tir.if_then_else(((((1 <= (floordiv(floormod(threadIdx.x_1, 63), 9) + ry.outer.outer)) && ((floordiv(floormod(threadIdx.x_1, 63), 9) + ry.outer.outer) < 8)) && (1 <= floormod(threadIdx.x_1, 9))) && (floormod(threadIdx.x_1, 9) < 8)), data_3: Buffer(data_2, float32, [25088], [])[((((cse_var_4 + (floordiv(threadIdx.x_1, 9)*7)) + cse_var_3) + floormod(threadIdx.x_1, 9)) - 8)], 0f32, [...]
-              attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              pad_temp.shared_1[(threadIdx.x_1 + 112)] = @tir.if_then_else(((((1 <= (floordiv(floormod((threadIdx.x_1 + 49), 63), 9) + ry.outer.outer)) && ((floordiv(floormod((threadIdx.x_1 + 49), 63), 9) + ry.outer.outer) < 8)) && (1 <= floormod((threadIdx.x_1 + 4), 9))) && (floormod((threadIdx.x_1 + 4), 9) < 8)), data_3[((((cse_var_4 + (floordiv((threadIdx.x_1 + 112), 9)*7)) + cse_var_3) + floormod((threadIdx.x_1 + 4), 9)) - 8)], 0f32, dtype=float32)
-              attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              pad_temp.shared_1[(threadIdx.x_1 + 224)] = @tir.if_then_else(((((1 <= (floordiv(floormod((threadIdx.x_1 + 35), 63), 9) + ry.outer.outer)) && ((floordiv(floormod((threadIdx.x_1 + 35), 63), 9) + ry.outer.outer) < 8)) && (1 <= floormod((threadIdx.x_1 + 8), 9))) && (floormod((threadIdx.x_1 + 8), 9) < 8)), data_3[((((cse_var_4 + (floordiv((threadIdx.x_1 + 224), 9)*7)) + cse_var_3) + floormod((threadIdx.x_1 + 8), 9)) - 8)], 0f32, dtype=float32)
-              attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              pad_temp.shared_1[(threadIdx.x_1 + 336)] = @tir.if_then_else(((((1 <= (floordiv(floormod((threadIdx.x_1 + 21), 63), 9) + ry.outer.outer)) && ((floordiv(floormod((threadIdx.x_1 + 21), 63), 9) + ry.outer.outer) < 8)) && (1 <= floormod((threadIdx.x_1 + 3), 9))) && (floormod((threadIdx.x_1 + 3), 9) < 8)), data_3[((((cse_var_4 + (floordiv((threadIdx.x_1 + 336), 9)*7)) + cse_var_3) + floormod((threadIdx.x_1 + 3), 9)) - 8)], 0f32, dtype=float32)
-              attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              if @tir.likely((threadIdx.x_1 < 56), dtype=bool) {
-                pad_temp.shared_1[(threadIdx.x_1 + 448)] = @tir.if_then_else(((((1 <= (floordiv(floormod((threadIdx.x_1 + 7), 63), 9) + ry.outer.outer)) && ((floordiv(floormod((threadIdx.x_1 + 7), 63), 9) + ry.outer.outer) < 8)) && (1 <= floormod((threadIdx.x_1 + 7), 9))) && (floormod((threadIdx.x_1 + 7), 9) < 8)), data_3[((((cse_var_4 + (floordiv((threadIdx.x_1 + 448), 9)*7)) + cse_var_3) + floormod((threadIdx.x_1 + 7), 9)) - 8)], 0f32, dtype=float32)
-              }
-              attr [IterVar(threadIdx.x_2: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              kernel.shared_1: Buffer(kernel.shared, float32, [384], [], scope="shared")[threadIdx.x_2] = kernel_3: Buffer(kernel_2, float32, [2359296], [])[((((((blockIdx.x*73728) + (floordiv(threadIdx.x_2, 24)*4608)) + cse_var_2) + (floordiv(floormod(threadIdx.x_2, 24), 3)*9)) + cse_var_1) + floormod(threadIdx.x_2, 3))]
-              attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              kernel.shared_1[(threadIdx.x_2 + 112)] = kernel_3[((((((blockIdx.x*73728) + (floordiv((threadIdx.x_2 + 112), 24)*4608)) + cse_var_2) + (floordiv(floormod((threadIdx.x_2 + 16), 24), 3)*9)) + cse_var_1) + floormod((threadIdx.x_2 + 1), 3))]
-              attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              kernel.shared_1[(threadIdx.x_2 + 224)] = kernel_3[((((((blockIdx.x*73728) + (floordiv((threadIdx.x_2 + 224), 24)*4608)) + cse_var_2) + (floordiv(floormod((threadIdx.x_2 + 8), 24), 3)*9)) + cse_var_1) + floormod((threadIdx.x_2 + 2), 3))]
-              attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 112;
-              if @tir.likely((threadIdx.x_2 < 48), dtype=bool) {
-                kernel.shared_1[(threadIdx.x_2 + 336)] = kernel_3[(((((((blockIdx.x*73728) + (floordiv(threadIdx.x_2, 24)*4608)) + cse_var_2) + (floordiv(floormod(threadIdx.x_2, 24), 3)*9)) + cse_var_1) + floormod(threadIdx.x_2, 3)) + 64512)]
-              }
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[floormod(threadIdx.x, 7)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 9)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 18)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 27)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 36)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 45)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 54)]*kernel.shared_1[(floordiv(threadIdx.x, 7)*24)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 1)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 10)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 19)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 28)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 37)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 46)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 55)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 1)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 2)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 11)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 20)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 29)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 38)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 47)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 56)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 2)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 63)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 72)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 81)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 90)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 99)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 108)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 117)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 3)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 64)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 73)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 82)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 91)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 100)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 109)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 118)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 4)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 65)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 74)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 83)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 92)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 101)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 110)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 119)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 5)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 126)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 135)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 144)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 153)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 162)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 171)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 180)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 6)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 127)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 136)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 145)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 154)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 163)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 172)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 181)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 7)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 128)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 137)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 146)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 155)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 164)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 173)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 182)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 8)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 189)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 198)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 207)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 216)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 225)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 234)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 243)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 9)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 190)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 199)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 208)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 217)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 226)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 235)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 244)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 10)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 191)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 200)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 209)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 218)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 227)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 236)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 245)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 11)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 252)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 261)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 270)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 279)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 288)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 297)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 306)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 12)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 253)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 262)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 271)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 280)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 289)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 298)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 307)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 13)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 254)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 263)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 272)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 281)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 290)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 299)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 308)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 14)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 315)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 324)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 333)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 342)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 351)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 360)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 369)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 15)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 316)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 325)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 334)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 343)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 352)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 361)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 370)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 16)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 317)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 326)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 335)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 344)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 353)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 362)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 371)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 17)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 378)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 387)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 396)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 405)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 414)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 423)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 432)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 18)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 379)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 388)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 397)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 406)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 415)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 424)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 433)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 19)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 380)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 389)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 398)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 407)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 416)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 425)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 434)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 20)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 441)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 450)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 459)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 468)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 477)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 486)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 495)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 21)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 442)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 451)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 460)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 469)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 478)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 487)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 496)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 22)]))
-              conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 443)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 452)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 461)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 470)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 479)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 488)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-              conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(floormod(threadIdx.x, 7) + 497)]*kernel.shared_1[((floordiv(threadIdx.x, 7)*24) + 23)]))
-            }
-          }
-        }
-        for (i2.inner: int32, 0, 7) {
-          compute_3: Buffer(compute_2, float32, [25088], [])[((((blockIdx.x*784) + (floordiv(threadIdx.x, 7)*49)) + (i2.inner*7)) + floormod(threadIdx.x, 7))] = max((conv2d_nchw_1[i2.inner] + bias_3: Buffer(bias_2, float32, [512], [])[((blockIdx.x*16) + floordiv(threadIdx.x, 7))]), 0f32)
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, kernel: T.handle, bias: T.handle, compute: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 512, 7, 7))
+            kernel_1 = T.match_buffer(kernel, (512, 512, 3, 3))
+            bias_1 = T.match_buffer(bias, (1, 512, 1, 1))
+            compute_1 = T.match_buffer(compute, (1, 512, 7, 7))
+            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")
+            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)
+            conv2d_nchw_1[0] = T.float32(0)
+            conv2d_nchw_1[1] = T.float32(0)
+            conv2d_nchw_1[2] = T.float32(0)
+            conv2d_nchw_1[3] = T.float32(0)
+            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
+                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_2 = T.buffer_decl((25088,), data=data_1.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_2[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_2[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_2[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_2[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))
+                threadIdx_x_2 = T.env_thread("threadIdx.x")
+                kernel_shared_1 = T.buffer_decl((3072,), data=kernel_shared, scope="shared")
+                kernel_2 = T.buffer_decl((2359296,), data=kernel_1.data)
+                with T.launch_thread(threadIdx_x_2, 64):
+                    kernel_shared_1[threadIdx_x_2] = kernel_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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_2[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):
+                compute_2 = T.buffer_decl((25088,), data=compute_1.data)
+                bias_2 = T.buffer_decl((512,), data=bias_1.data)
+                compute_2[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_2[blockIdx_x // 7 * 128 + threadIdx_x * 2 + i1_inner], T.float32(0))
 
 
 
@@ -510,7 +769,7 @@ We build the binary and check its correctness and performance.
 
  .. code-block:: none
 
-    Execution time of this operator: 0.289 ms
+    Execution time of this operator: 0.360 ms
 
 
 
@@ -559,35 +818,35 @@ 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=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=16)
+    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_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=7)
+    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_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=1)
-    conv2d_nchw_xx_o_o_o_i, conv2d_nchw_xx_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_i, factor=7)
+    conv2d_nchw_xx_o_o_i, conv2d_nchw_xx_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_i, factor=7)
+    conv2d_nchw_xx_o_o_o_i, conv2d_nchw_xx_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_i, factor=1)
     conv2d_nchw_xx_o_o_o_o, conv2d_nchw_xx_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_o_i, factor=1)
     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=3)
-    conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_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)
     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=1)
-    compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=16)
+    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_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=7)
+    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_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=1)
-    compute_i3_o_o_i, compute_i3_o_i = s[compute].split(compute_i3_o_i, factor=7)
+    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)
     compute_i3_o_o_o, compute_i3_o_o_i = s[compute].split(compute_i3_o_o_i, factor=1)
     s[compute].reorder(compute_i0_o_o_o, compute_i1_o_o_o, compute_i2_o_o_o, compute_i3_o_o_o, compute_i0_o_o_i, compute_i1_o_o_i, compute_i2_o_o_i, compute_i3_o_o_i, compute_i0_o_i, compute_i1_o_i, compute_i2_o_i, compute_i3_o_i, compute_i0_i, compute_i1_i, compute_i2_i, compute_i3_i)
     s[conv2d_nchw].compute_at(s[compute], compute_i3_o_i)
@@ -607,12 +866,12 @@ 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=112)
+    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)
     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=1)
+    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)
     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=112)
+    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)
     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, "unroll_explicit", True)
@@ -632,10 +891,10 @@ 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__(112) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
-      float conv2d_nchw[7];
-      __shared__ float pad_temp_shared[504];
-      __shared__ float kernel_shared[384];
+    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];
       conv2d_nchw[0] = 0.000000e+00f;
       conv2d_nchw[1] = 0.000000e+00f;
       conv2d_nchw[2] = 0.000000e+00f;
@@ -643,195 +902,419 @@ 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();
-          pad_temp_shared[((int)threadIdx.x)] = (((((1 <= (((((int)threadIdx.x) % 63) / 9) + ry_outer_outer)) && ((((((int)threadIdx.x) % 63) / 9) + ry_outer_outer) < 8)) && (1 <= (((int)threadIdx.x) % 9))) && ((((int)threadIdx.x) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + ((((int)threadIdx.x) / 9) * 7)) + (ry_outer_outer * 7)) + (((int)threadIdx.x) % 9)) - 8)] : 0.000000e+00f);
-          pad_temp_shared[(((int)threadIdx.x) + 112)] = (((((1 <= ((((((int)threadIdx.x) + 49) % 63) / 9) + ry_outer_outer)) && (((((((int)threadIdx.x) + 49) % 63) / 9) + ry_outer_outer) < 8)) && (1 <= ((((int)threadIdx.x) + 4) % 9))) && (((((int)threadIdx.x) + 4) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 112) / 9) * 7)) + (ry_outer_outer * 7)) + ((((int)threadIdx.x) + 4) % 9)) - 8)] : 0.000000e+00f);
-          pad_temp_shared[(((int)threadIdx.x) + 224)] = (((((1 <= ((((((int)threadIdx.x) + 35) % 63) / 9) + ry_outer_outer)) && (((((((int)threadIdx.x) + 35) % 63) / 9) + ry_outer_outer) < 8)) && (1 <= ((((int)threadIdx.x) + 8) % 9))) && (((((int)threadIdx.x) + 8) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 224) / 9) * 7)) + (ry_outer_outer * 7)) + ((((int)threadIdx.x) + 8) % 9)) - 8)] : 0.000000e+00f);
-          pad_temp_shared[(((int)threadIdx.x) + 336)] = (((((1 <= ((((((int)threadIdx.x) + 21) % 63) / 9) + ry_outer_outer)) && (((((((int)threadIdx.x) + 21) % 63) / 9) + ry_outer_outer) < 8)) && (1 <= ((((int)threadIdx.x) + 3) % 9))) && (((((int)threadIdx.x) + 3) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 336) / 9) * 7)) + (ry_outer_outer * 7)) + ((((int)threadIdx.x) + 3) % 9)) - 8)] : 0.000000e+00f);
-          if (((int)threadIdx.x) < 56) {
-            pad_temp_shared[(((int)threadIdx.x) + 448)] = (((((1 <= (((((int)threadIdx.x) + 7) / 9) + ry_outer_outer)) && ((((((int)threadIdx.x) + 7) / 9) + ry_outer_outer) < 8)) && (1 <= ((((int)threadIdx.x) + 7) % 9))) && (((((int)threadIdx.x) + 7) % 9) < 8)) ? data[(((((rc_outer_outer * 392) + (((((int)threadIdx.x) + 448) / 9) * 7)) + (ry_outer_outer * 7)) + ((((int)threadIdx.x) + 7) % 9)) - 8)] : 0.000000e+00f);
+          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);
           }
-          kernel_shared[((int)threadIdx.x)] = kernel[((((((((int)blockIdx.x) * 73728) + ((((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) + 112)] = kernel[((((((((int)blockIdx.x) * 73728) + (((((int)threadIdx.x) + 112) / 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) + 224)] = kernel[((((((((int)blockIdx.x) * 73728) + (((((int)threadIdx.x) + 224) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
-          if (((int)threadIdx.x) < 48) {
-            kernel_shared[(((int)threadIdx.x) + 336)] = kernel[(((((((((int)blockIdx.x) * 73728) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 64512)];
+          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[(((int)threadIdx.x) % 7)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 9)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[((((int)threadIdx.x) / 7) * 24)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 1)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 1)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 2)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 63)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 72)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 81)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 3)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 64)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 73)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 4)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 65)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 74)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 5)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 144)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 153)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 162)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 171)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 180)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 6)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 145)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 154)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 163)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 172)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 181)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 7)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 146)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 155)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 164)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 173)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 182)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 8)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 189)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 198)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 207)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 216)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 225)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 234)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 243)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 9)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 190)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 199)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 208)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 217)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 226)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 235)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 244)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 10)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 191)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 200)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 209)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 218)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 227)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 236)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 245)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 11)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 252)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 261)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 270)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 279)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 288)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 297)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 306)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 12)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 253)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 262)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 271)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 280)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 289)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 298)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 307)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 13)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 254)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 263)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 272)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 281)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 290)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 299)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 308)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 14)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 315)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 324)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 333)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 342)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 351)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 360)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 369)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 15)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 316)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 325)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 334)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 343)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 352)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 361)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 370)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 16)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 317)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 326)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 335)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 344)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 353)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 362)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 371)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 17)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 378)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 387)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 396)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 405)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 414)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 423)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 432)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 18)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 379)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 388)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 397)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 406)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 415)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 424)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 433)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 19)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 380)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 389)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 398)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 407)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 416)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 425)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 434)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 20)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 441)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 450)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 459)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 468)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 477)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 486)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 495)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 21)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 442)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 451)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 460)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 469)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 478)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 487)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 496)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 22)]));
-          conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 443)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 452)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 461)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 470)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 479)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 488)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
-          conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 497)] * kernel_shared[(((((int)threadIdx.x) / 7) * 24) + 23)]));
+          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)]));
         }
       }
-      for (int i2_inner = 0; i2_inner < 7; ++i2_inner) {
-        compute[((((((int)blockIdx.x) * 784) + ((((int)threadIdx.x) / 7) * 49)) + (i2_inner * 7)) + (((int)threadIdx.x) % 7))] = max((conv2d_nchw[i2_inner] + bias[((((int)blockIdx.x) * 16) + (((int)threadIdx.x) / 7))]), 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);
+        }
       }
     }
 
@@ -893,7 +1376,7 @@ 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  33.867 seconds)
+   **Total running time of the script:** ( 5 minutes  38.572 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 03eae9f39f..8031439775 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)  
-       7.8752       7.8711       7.8906       7.8638       0.0113   
+       7.8969       7.9011       7.9067       7.8829       0.0102   
                
 
 
@@ -675,7 +675,7 @@ Other Tips
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 1 minutes  4.265 seconds)
+   **Total running time of the script:** ( 1 minutes  5.307 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 4a37ba6d02..7d1f9b21bb 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)  
-      755.9901     756.4164     757.1877     754.3664      1.1906   
+      747.4873     750.1255     750.1615     742.1749      3.7564   
                
 
 
@@ -694,7 +694,7 @@ Other Tips
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 1 minutes  37.080 seconds)
+   **Total running time of the script:** ( 1 minutes  36.142 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 1813cfe062..e7e98bd5f9 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
@@ -381,44 +381,36 @@ layout transformation, parallelization, vectorization, unrolling, and operator f
  .. code-block:: none
 
     Lowered TIR:
-    @main = primfn(placeholder_5: handle, placeholder_6: handle, placeholder_7: handle, placeholder_8: handle, placeholder_9: handle, compute_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {placeholder: Buffer(placeholder_10: Pointer(float32), float32, [128, 256], []),
-                 placeholder_1: Buffer(placeholder_11: Pointer(float32), float32, [4916, 16, 1], []),
-                 placeholder_2: Buffer(placeholder_12: Pointer(int32), int32, [4916], []),
-                 placeholder_3: Buffer(placeholder_13: Pointer(int32), int32, [33], []),
-                 placeholder_4: Buffer(placeholder_14: Pointer(float32), float32, [128, 512], []),
-                 compute: Buffer(compute_2: Pointer(float32), float32, [128, 512], [])}
-      buffer_map = {placeholder_5: placeholder, placeholder_6: placeholder_1, placeholder_7: placeholder_2, placeholder_8: placeholder_3, placeholder_9: placeholder_4, compute_1: compute} {
-      for (i0.outer.i1.outer.fused: int32, 0, 128) "parallel" {
-        allocate(compute_3: Pointer(global float32), float32, [1024]), storage_scope = global {
-          for (i.outer.inner: int32, 0, 16) {
-            for (i.inner.init: int32, 0, 4) {
-              for (j.init: int32, 0, 16) {
-                compute_4: Buffer(compute_3, float32, [1024], [])[(((i.outer.inner*64) + (i.inner.init*16)) + j.init)] = 0f32
-              }
-            }
-            for (elem_idx: int32, 0, let cse_var_1: int32 = floordiv(floormod(i0.outer.i1.outer.fused, 64), 2) in (placeholder_15: Buffer(placeholder_13, int32, [33], [])[(cse_var_1 + 1)] - placeholder_15[cse_var_1])) {
-              for (i.inner: int32, 0, 4) {
-                for (j: int32, 0, 16) {
-                  let cse_var_3: int32 = floordiv(floormod(i0.outer.i1.outer.fused, 64), 2)
-                  let cse_var_2: int32 = (((i.outer.inner*64) + (i.inner*16)) + j)
-                  compute_4[cse_var_2] = (compute_4[cse_var_2] + (placeholder_16: Buffer(placeholder_11, float32, [78656], [])[(((placeholder_15[cse_var_3]*16) + (elem_idx*16)) + j)]*max(placeholder_17: Buffer(placeholder_10, float32, [32768], [])[((((floordiv(i0.outer.i1.outer.fused, 64)*16384) + (i.outer.inner*1024)) + (i.inner*256)) + placeholder_18: Buffer(placeholder_12, int32, [4916], [])[(placeholder_15[cse_var_3] + elem_idx)])], 0f32)))
-                }
-              }
-            }
-          }
-          for (i0.inner: int32, 0, 64) {
-            let cse_var_5: int32 = floormod(i0.outer.i1.outer.fused, 64)
-            let cse_var_6: int32 = (cse_var_5*8)
-            let cse_var_4: int32 = (((floordiv(i0.outer.i1.outer.fused, 64)*32768) + (i0.inner*512)) + cse_var_6)
-            compute_5: Buffer(compute_2, float32, [65536], [])[ramp(cse_var_4, 1, 8)] = max((compute_4[ramp((((i0.inner*16) + cse_var_6) - (floordiv(cse_var_5, 2)*16)), 1, 8)] + placeholder_19: Buffer(placeholder_14, float32, [65536], [])[ramp(cse_var_4, 1, 8)]), broadcast(0f32, 8))
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(placeholder: T.handle, placeholder_1: T.handle, placeholder_2: T.handle, placeholder_3: T.handle, placeholder_4: T.handle, compute: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            placeholder_5 = T.match_buffer(placeholder, (128, 256))
+            placeholder_6 = T.match_buffer(placeholder_1, (4916, 16, 1))
+            placeholder_7 = T.match_buffer(placeholder_2, (4916,), "int32")
+            placeholder_8 = T.match_buffer(placeholder_3, (33,), "int32")
+            placeholder_9 = T.match_buffer(placeholder_4, (128, 512))
+            compute_1 = T.match_buffer(compute, (128, 512))
+            for i0_outer_i1_outer_fused in T.parallel(32):
+                compute_2 = T.allocate([2048], "float32", "global")
+                compute_3 = T.buffer_decl((2048,), data=compute_2)
+                for i_outer_inner in range(4):
+                    for i_inner_init, j_init in T.grid(32, 16):
+                        compute_3[i_outer_inner * 512 + i_inner_init * 16 + j_init] = T.float32(0)
+                    for elem_idx, i_inner, j in T.grid(placeholder_10[i0_outer_i1_outer_fused + 1] - placeholder_10[i0_outer_i1_outer_fused], 32, 16):
+                        placeholder_10 = T.buffer_decl((33,), "int32", data=placeholder_8.data)
+                        if T.likely(elem_idx < placeholder_10[i0_outer_i1_outer_fused + 1] - placeholder_10[i0_outer_i1_outer_fused]):
+                            placeholder_11 = T.buffer_decl((78656,), data=placeholder_6.data)
+                            placeholder_12 = T.buffer_decl((32768,), data=placeholder_5.data)
+                            placeholder_13 = T.buffer_decl((4916,), "int32", data=placeholder_7.data)
+                            cse_var_1: T.int32 = i_outer_inner * 512 + i_inner * 16 + j
+                            compute_3[cse_var_1] = compute_3[cse_var_1] + placeholder_11[placeholder_10[i0_outer_i1_outer_fused] * 16 + elem_idx * 16 + j] * T.max(placeholder_12[i_outer_inner * 8192 + i_inner * 256 + placeholder_13[placeholder_10[i0_outer_i1_outer_fused] + elem_idx]], T.float32(0))
+                for i0_inner in range(128):
+                    cse_var_2: T.int32 = i0_inner * 512 + i0_outer_i1_outer_fused * 16
+                    compute_4 = T.buffer_decl((65536,), data=compute_1.data)
+                    placeholder_10 = T.buffer_decl((65536,), data=placeholder_9.data)
+                    compute_4[cse_var_2:cse_var_2 + 16] = T.max(compute_3[i0_inner * 16:i0_inner * 16 + 16] + placeholder_10[cse_var_2:cse_var_2 + 16], T.Broadcast(T.float32(0), 16))
 
 
 
@@ -468,7 +460,7 @@ We build the binary and check its correctness and performance.
 
  .. code-block:: none
 
-    Execution time of this operator: 2.954 ms
+    Execution time of this operator: 1.642 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 6110533f88..49b5a71b1d 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
@@ -5,12 +5,12 @@
 
 Computation times
 =================
-**00:56.213** total execution time for **how_to_tune_with_autotvm** files:
+**00:31.885** 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``)           | 00:56.180 | 0.0 MB |
+| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_conv2d_cuda.py` (``tune_conv2d_cuda.py``)           | 00:31.852 | 0.0 MB |
 +--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_tune_with_autotvm_tune_relay_x86.py` (``tune_relay_x86.py``)               | 00:00.019 | 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_cuda.py` (``tune_relay_cuda.py``)             | 00:00.005 | 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 3eb0ac8dc0..8c69a262da 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
@@ -314,7 +314,7 @@ for this template
       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:454
+            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
@@ -370,7 +370,7 @@ for this template
       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:454
+            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
@@ -390,7 +390,7 @@ 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, 8, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 16, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5858902
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 16, 4, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 32]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,10212643
     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)
@@ -437,7 +437,7 @@ for this template
       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:454
+            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
@@ -493,7 +493,7 @@ for this template
       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:454
+            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
@@ -513,9 +513,9 @@ 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, 4, 1, 16]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5705206
-    No: 3   GFLOPS: 2.80/2.80       result: MeasureResult(costs=(0.082569641,), error_no=MeasureErrorNo.NO_ERROR, all_cost=7.368779420852661, timestamp=1673997103.732286)  [('tile_f', [-1, 32, 1, 2]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,5069520
-    No: 4   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 256, 2, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2406378
+    No: 3   GFLOPS: 35.66/35.66     result: MeasureResult(costs=(0.006491297631578947,), error_no=MeasureErrorNo.NO_ERROR, all_cost=5.062892198562622, timestamp=1673997983.7155943)        [('tile_f', [-1, 16, 1, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 1, 8]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3193964
+    No: 4   GFLOPS: 0.00/35.66      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
@@ -561,7 +561,7 @@ for this template
       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:454
+            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
@@ -617,7 +617,7 @@ for this template
       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:454
+            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
@@ -637,8 +637,8 @@ 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, 256, 1, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 128, 2]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,60123
-    No: 5   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 2, 16, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2603126
+    No: 5   GFLOPS: 0.00/35.66      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
@@ -684,7 +684,7 @@ for this template
       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:454
+            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
@@ -740,7 +740,7 @@ for this template
       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:454
+            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
@@ -760,8 +760,8 @@ 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, 2, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 256]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,960972
-    No: 6   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 8, 2, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 4, 64]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3263987
+    No: 6   GFLOPS: 0.00/35.66      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
@@ -807,7 +807,7 @@ for this template
       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:454
+            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
@@ -863,7 +863,7 @@ for this template
       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:454
+            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
@@ -883,8 +883,8 @@ 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, 1, 8, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 8, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9534337
-    No: 7   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 16, 2, 16]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 256]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,5219454
+    No: 7   GFLOPS: 0.00/35.66      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
@@ -930,7 +930,7 @@ for this template
       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:454
+            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
@@ -986,7 +986,7 @@ for this template
       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:454
+            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
@@ -1006,8 +1006,8 @@ 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, 4]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4357867
-    No: 8   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 4, 2, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 4, 128]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4439392
+    No: 8   GFLOPS: 0.00/35.66      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
@@ -1053,7 +1053,7 @@ for this template
       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:454
+            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
@@ -1109,7 +1109,7 @@ for this template
       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:454
+            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
@@ -1129,8 +1129,8 @@ 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, 1, 2, 64]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 1, 256]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9863684
-    No: 9   GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 128, 4, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 256, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9163246
+    No: 9   GFLOPS: 0.00/35.66      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
@@ -1176,7 +1176,7 @@ for this template
       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:454
+            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
@@ -1232,7 +1232,7 @@ for this template
       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:454
+            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
@@ -1252,8 +1252,8 @@ 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, 1, 128, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 2]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1603897
-    No: 10  GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 128, 1, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,8037102
+    No: 10  GFLOPS: 0.00/35.66      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
@@ -1299,7 +1299,7 @@ for this template
       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:454
+            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
@@ -1355,7 +1355,7 @@ for this template
       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:454
+            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
@@ -1375,8 +1375,11 @@ 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, 2, 8]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 64]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,5009985
-    No: 11  GFLOPS: 0.00/2.80       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 2, 64, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 2, 128]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,952206
+    No: 11  GFLOPS: 49.42/49.42     result: MeasureResult(costs=(0.004684357727272727,), error_no=MeasureErrorNo.NO_ERROR, all_cost=4.610525608062744, timestamp=1673997990.6140513)        [('tile_f', [-1, 2, 4, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 128, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,3899740
+    No: 12  GFLOPS: 442.40/442.40   result: MeasureResult(costs=(0.0005232901237623763,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.0484848022460938, timestamp=1673997991.323063)       [('tile_f', [-1, 1, 8, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1232519
+    No: 13  GFLOPS: 48.84/442.40    result: MeasureResult(costs=(0.0047403134242424245,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.498650312423706, timestamp=1673997992.9843655)       [('tile_f', [-1, 4, 1, 4]), ('tile_y', [-1, 1, 7, 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', 0), ('unroll_explicit', 1)],None,5633422
+    No: 14  GFLOPS: 0.00/442.40     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
@@ -1422,7 +1425,7 @@ for this template
       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:454
+            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
@@ -1478,7 +1481,7 @@ for this template
       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:454
+            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
@@ -1498,9 +1501,9 @@ 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, 16, 2, 2]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 32, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,5473888
-    No: 12  GFLOPS: 2.87/2.87       result: MeasureResult(costs=(0.08073779275,), error_no=MeasureErrorNo.NO_ERROR, all_cost=2.3404765129089355, timestamp=1673997108.4687462)      [('tile_f', [-1, 1, 1, 16]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 1, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,810864
-    No: 13  GFLOPS: 0.00/2.87       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 32, 1, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 64, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2993105
+    No: 15  GFLOPS: 47.91/442.40    result: MeasureResult(costs=(0.004832207047619048,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.181091070175171, timestamp=1673997993.7343552)        [('tile_f', [-1, 4, 1, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 16, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,857342
+    No: 16  GFLOPS: 0.00/442.40     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
@@ -1546,7 +1549,7 @@ for this template
       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:454
+            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
@@ -1602,7 +1605,7 @@ for this template
       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:454
+            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
@@ -1622,8 +1625,8 @@ 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, 32, 1, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 512]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9872780
-    No: 14  GFLOPS: 0.00/2.87       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 32, 1, 4]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 128, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6063305
+    No: 17  GFLOPS: 0.00/442.40     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
@@ -1669,7 +1672,7 @@ for this template
       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:454
+            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
@@ -1725,7 +1728,7 @@ for this template
       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:454
+            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
@@ -1745,8 +1748,8 @@ 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, 8, 16]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,8380201
-    No: 15  GFLOPS: 0.00/2.87       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 4, 16, 2]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 1, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1709927
+    No: 18  GFLOPS: 0.00/442.40     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
@@ -1792,7 +1795,7 @@ for this template
       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:454
+            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
@@ -1848,7 +1851,7 @@ for this template
       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:454
+            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
@@ -1868,8 +1871,9 @@ 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, 1, 64, 4]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 256, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6648093
-    No: 16  GFLOPS: 0.00/2.87       result: Traceback (most recent call last):
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 8, 64, 1]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 4, 128]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 1)],None,6182048
+    No: 19  GFLOPS: 176.67/442.40   result: MeasureResult(costs=(0.0013103584155844156,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.035775899887085, timestamp=1673997994.9901366)       [('tile_f', [-1, 1, 2, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,793384
+    No: 20  GFLOPS: 0.00/442.40     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
@@ -1915,7 +1919,7 @@ for this template
       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:454
+            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
@@ -1971,7 +1975,7 @@ for this template
       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:454
+            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
@@ -1991,499 +1995,7 @@ 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, 1, 8]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 64, 8]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2829118
-    No: 17  GFLOPS: 0.00/2.87       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:454
-      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:454
-      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, 128, 1]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 1, 7]), ('tile_rc', [-1, 4, 128]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2699230
-    No: 18  GFLOPS: 0.00/2.87       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:454
-      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:454
-      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, 1, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 8, 64]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,3461539
-    No: 19  GFLOPS: 0.00/2.87       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:454
-      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:454
-      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, 2, 32]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 4, 4]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,850491
-    No: 20  GFLOPS: 0.00/2.87       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:454
-      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:454
-      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, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 4, 32]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 0)],None,2860203
+    tvm.autotvm.task.space.InstantiationError: Skipped because of invalid gpu kernel        [('tile_f', [-1, 2, 8, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 32, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 0)],None,4732280
 
 
 
@@ -2538,9 +2050,9 @@ and measure running time.
     Finish loading 20 records
 
     Best config:
-    [('tile_f', [-1, 1, 1, 16]), ('tile_y', [-1, 7, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 1, 2]), ('tile_ry', [-1, 3, 1]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,810864
+    [('tile_f', [-1, 1, 8, 2]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 2, 4]), ('tile_ry', [-1, 1, 1]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 0), ('unroll_explicit', 0)],None,1232519
     Finish loading 20 records
-    Time cost of this operator: 0.067610
+    Time cost of this operator: 0.000931
 
 
 
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 7b08bc3280..b1897781c5 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  311.6     98.708   (1, 2, 10, 10, 3)  2       1        [311.6]           
-    tvmgen_default_fused_layout_transform_1       tvmgen_default_fused_layout_transform_1       3.105     0.984    (1, 6, 10, 10)     1       1        [3.105]           
-    tvmgen_default_fused_layout_transform         tvmgen_default_fused_layout_transform         0.975     0.309    (1, 1, 10, 10, 3)  1       1        [0.975]           
-    Total_time                                    -                                             315.679   -        -                  -       -        -                 
+    tvmgen_default_fused_nn_contrib_conv2d_NCHWc  tvmgen_default_fused_nn_contrib_conv2d_NCHWc  310.0     98.629   (1, 2, 10, 10, 3)  2       1        [310.0]           
+    tvmgen_default_fused_layout_transform_1       tvmgen_default_fused_layout_transform_1       3.093     0.984    (1, 6, 10, 10)     1       1        [3.093]           
+    tvmgen_default_fused_layout_transform         tvmgen_default_fused_layout_transform         1.217     0.387    (1, 1, 10, 10, 3)  1       1        [1.217]           
+    Total_time                                    -                                             314.31    -        -                  -       -        -                 
 
 
 
@@ -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  102.4     97.405   (1, 6, 10, 10, 1)  2       1        [102.4]           
-    tvmgen_default_fused_layout_transform_1       tvmgen_default_fused_layout_transform_1       1.774     1.688    (1, 6, 10, 10)     1       1        [1.774]           
-    tvmgen_default_fused_layout_transform         tvmgen_default_fused_layout_transform         0.954     0.907    (1, 1, 10, 10, 3)  1       1        [0.954]           
-    Total_time                                    -                                             105.128   -        -                  -       -        -                 
+    tvmgen_default_fused_nn_contrib_conv2d_NCHWc  tvmgen_default_fused_nn_contrib_conv2d_NCHWc  135.0     97.972   (1, 6, 10, 10, 1)  2       1        [135.0]           
+    tvmgen_default_fused_layout_transform_1       tvmgen_default_fused_layout_transform_1       1.832     1.329    (1, 6, 10, 10)     1       1        [1.832]           
+    tvmgen_default_fused_layout_transform         tvmgen_default_fused_layout_transform         0.962     0.698    (1, 1, 10, 10, 3)  1       1        [0.962]           
+    Total_time                                    -                                             137.794   -        -                  -       -        -                 
 
 
 
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 46fa4b1eb6..86b80a3c53 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, 172MB/s]
+
      0%|          | 0.00/3.42M [00:00<?, ?B/s]
    100%|##########| 3.42M/3.42M [00:00<00:00, 105MB/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  8.059 seconds)
+   **Total running time of the script:** ( 1 minutes  7.288 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 c865039436..5675db50f9 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/tmpp7y7ag8l/images/random'
+    '/tmp/tmphugqqvuc/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: [0.0, 1.0], [0.0, 1.0], [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0]
+   :alt: [1.0, 0.0], [0.0, 1.0], [1.0, 0.0], [0.0, 1.0], [1.0, 0.0], [0.0, 1.0], [0.0, 1.0], [0.0, 1.0], [1.0, 0.0], [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/tmpp7y7ag8l/images/target contains 8144 images
-    /tmp/tmpp7y7ag8l/images/random contains 5000 images
+    /tmp/tmphugqqvuc/images/target contains 8144 images
+    /tmp/tmphugqqvuc/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: 0.2165 - accuracy: 0.9255 - val_loss: 0.1829 - val_accuracy: 0.9358 - 47s/epoch - 143ms/step
+    328/328 - 47s - loss: 0.2141 - accuracy: 0.9224 - val_loss: 0.1267 - val_accuracy: 0.9592 - 47s/epoch - 143ms/step
     Epoch 2/3
-    328/328 - 43s - loss: 0.0923 - accuracy: 0.9676 - val_loss: 0.0980 - val_accuracy: 0.9671 - 43s/epoch - 132ms/step
+    328/328 - 43s - loss: 0.0946 - accuracy: 0.9665 - val_loss: 0.1080 - val_accuracy: 0.9637 - 43s/epoch - 131ms/step
     Epoch 3/3
-    328/328 - 43s - loss: 0.0616 - accuracy: 0.9785 - val_loss: 0.1011 - val_accuracy: 0.9645 - 43s/epoch - 131ms/step
+    328/328 - 43s - loss: 0.0587 - accuracy: 0.9772 - val_loss: 0.1381 - val_accuracy: 0.9581 - 43s/epoch - 130ms/step
 
-    <keras.callbacks.History object at 0x7fba407f6dd0>
+    <keras.callbacks.History object at 0x7f4f1bef1d10>
 
 
 
@@ -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  29.208 seconds)
+   **Total running time of the script:** ( 5 minutes  0.332 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 749897fbdd..760bf797ad 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
@@ -5,22 +5,22 @@
 
 Computation times
 =================
-**06:41.212** total execution time for **how_to_work_with_microtvm** files:
+**07:10.918** total execution time for **how_to_work_with_microtvm** files:
 
 +---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_train.py` (``micro_train.py``)               | 04:29.208 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_train.py` (``micro_train.py``)               | 05:00.332 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_pytorch.py` (``micro_pytorch.py``)           | 01:08.059 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_pytorch.py` (``micro_pytorch.py``)           | 01:07.288 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_autotune.py` (``micro_autotune.py``)         | 00:51.341 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_autotune.py` (``micro_autotune.py``)         | 00:50.924 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_aot.py` (``micro_aot.py``)                   | 00:08.797 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_aot.py` (``micro_aot.py``)                   | 00:08.612 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_tflite.py` (``micro_tflite.py``)             | 00:03.807 | 0.0 MB |
-+---------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_microtvm_micro_ethosu.py` (``micro_ethosu.py``)             | 00:00.000 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_tflite.py` (``micro_tflite.py``)             | 00:03.762 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
 | :ref:`sphx_glr_how_to_work_with_microtvm_micro_tvmc.py` (``micro_tvmc.py``)                 | 00:00.000 | 0.0 MB |
 +---------------------------------------------------------------------------------------------+-----------+--------+
+| :ref:`sphx_glr_how_to_work_with_microtvm_micro_ethosu.py` (``micro_ethosu.py``)             | 00:00.000 | 0.0 MB |
++---------------------------------------------------------------------------------------------+-----------+--------+
 | :ref:`sphx_glr_how_to_work_with_microtvm_micro_reference_vm.py` (``micro_reference_vm.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 9d2338b1a1..04383d1c72 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
 =================
-**00:45.353** total execution time for **how_to_work_with_relay** files:
+**00:44.480** 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``) | 00:33.141 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_using_pipeline_executor.py` (``using_pipeline_executor.py``) | 00:32.510 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_using_external_lib.py` (``using_external_lib.py``)           | 00:10.445 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_using_external_lib.py` (``using_external_lib.py``)           | 00:10.353 | 0.0 MB |
 +----------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_relay_build_gcn.py` (``build_gcn.py``)                             | 00:01.761 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_relay_build_gcn.py` (``build_gcn.py``)                             | 00:01.611 | 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 |
 +----------------------------------------------------------------------------------------------------+-----------+--------+
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 137366751a..68aac396a0 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 0x7fb9c243b320>
+    <function my_cuda_math_rule at 0x7f4f1c72b560>
 
 
 
diff --git a/docs/_sources/how_to/work_with_schedules/reduction.rst.txt b/docs/_sources/how_to/work_with_schedules/reduction.rst.txt
index 8caafbcad8..1f162cc8d6 100644
--- a/docs/_sources/how_to/work_with_schedules/reduction.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/reduction.rst.txt
@@ -116,20 +116,24 @@ Before doing anything, let us print out the IR code of default schedule.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i: int32, 0, n) {
-        B_3: Buffer(B_2, float32, [(stride_2*n)], [], type="auto")[(i*stride_2)] = 0f32
-        for (k: int32, 0, m) {
-          B_3[(i*stride_2)] = (B_3[(i*stride_2)] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((i*stride) + (k*stride_1))])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_2,), type="auto")
+            for i in range(n):
+                B_2 = T.buffer_decl((stride_2 * n,), data=B_1.data, type="auto")
+                B_2[i * stride_2] = T.float32(0)
+                for k in range(m):
+                    A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                    B_2[i * stride_2] = B_2[i * stride_2] + A_2[i * stride + k * stride_1]
 
 
 
@@ -159,31 +163,28 @@ axis by different factors. The result is a nested reduction.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.outer: int32, 0, floordiv((n + 31), 32)) {
-        for (i.inner: int32, 0, 32) {
-          if @tir.likely((((i.outer*32) + i.inner) < n), dtype=bool) {
-            B_3: Buffer(B_2, float32, [(stride_2*n)], [], type="auto")[(((i.outer*32) + i.inner)*stride_2)] = 0f32
-          }
-          if @tir.likely((((i.outer*32) + i.inner) < n), dtype=bool) {
-            for (k.outer: int32, 0, floordiv((m + 15), 16)) {
-              for (k.inner: int32, 0, 16) {
-                if @tir.likely((((k.outer*16) + k.inner) < m), dtype=bool) {
-                  let cse_var_1: int32 = ((i.outer*32) + i.inner)
-                  B_3[(cse_var_1*stride_2)] = (B_3[(cse_var_1*stride_2)] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((cse_var_1*stride) + (((k.outer*16) + k.inner)*stride_1))])
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_2,), type="auto")
+            for i_outer, i_inner in T.grid((n + 31) // 32, 32):
+                B_2 = T.buffer_decl((stride_2 * n,), data=B_1.data, type="auto")
+                if T.likely(i_outer * 32 + i_inner < n):
+                    B_2[(i_outer * 32 + i_inner) * stride_2] = T.float32(0)
+                if T.likely(i_outer * 32 + i_inner < n):
+                    for k_outer, k_inner in T.grid((m + 15) // 16, 16):
+                        if T.likely(k_outer * 16 + k_inner < m):
+                            A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                            cse_var_1: T.int32 = i_outer * 32 + i_inner
+                            B_2[cse_var_1 * stride_2] = B_2[cse_var_1 * stride_2] + A_2[cse_var_1 * stride + (k_outer * 16 + k_inner) * stride_1]
 
 
 
@@ -208,29 +209,30 @@ If we are building a GPU kernel, we can bind the rows of B to GPU threads.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 31), 32);
-      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32 {
-        if @tir.likely((((blockIdx.x*32) + threadIdx.x) < n), dtype=bool) {
-          B_3: Buffer(B_2, float32, [(stride_2*n)], [], type="auto")[(((blockIdx.x*32) + threadIdx.x)*stride_2)] = 0f32
-        }
-        for (k.outer: int32, 0, floordiv((m + 15), 16)) {
-          for (k.inner: int32, 0, 16) {
-            if @tir.likely((((blockIdx.x*32) + threadIdx.x) < n), dtype=bool) {
-              if @tir.likely((((k.outer*16) + k.inner) < m), dtype=bool) {
-                B_3[(((blockIdx.x*32) + threadIdx.x)*stride_2)] = (B_3[(((blockIdx.x*32) + threadIdx.x)*stride_2)] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((((blockIdx.x*32) + threadIdx.x)*stride) + (((k.outer*16) + k.inner)*stride_1))])
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_2,), type="auto")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, (n + 31) // 32)
+            threadIdx_x = T.env_thread("threadIdx.x")
+            T.launch_thread(threadIdx_x, 32)
+            B_2 = T.buffer_decl((stride_2 * n,), data=B_1.data, type="auto")
+            if T.likely(blockIdx_x * 32 + threadIdx_x < n):
+                B_2[(blockIdx_x * 32 + threadIdx_x) * stride_2] = T.float32(0)
+            for k_outer, k_inner in T.grid((m + 15) // 16, 16):
+                if T.likely(blockIdx_x * 32 + threadIdx_x < n):
+                    if T.likely(k_outer * 16 + k_inner < m):
+                        A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                        B_2[(blockIdx_x * 32 + threadIdx_x) * stride_2] = B_2[(blockIdx_x * 32 + threadIdx_x) * stride_2] + A_2[(blockIdx_x * 32 + threadIdx_x) * stride + (k_outer * 16 + k_inner) * stride_1]
 
 
 
@@ -266,32 +268,31 @@ result B.rf. The factored dimension becomes the first dimension of B.rf.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      allocate(B.rf: Pointer(global float32), float32, [(n*16)]), storage_scope = global {
-        for (k.inner: int32, 0, 16) {
-          for (i: int32, 0, n) {
-            B.rf_1: Buffer(B.rf, float32, [(16*n)], [])[((k.inner*n) + i)] = 0f32
-            for (k.outer: int32, 0, floordiv((m + 15), 16)) {
-              if @tir.likely((((k.outer*16) + k.inner) < m), dtype=bool) {
-                B.rf_1[((k.inner*n) + i)] = (B.rf_1[((k.inner*n) + i)] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((i*stride) + (((k.outer*16) + k.inner)*stride_1))])
-              }
-            }
-          }
-        }
-        for (ax0: int32, 0, n) {
-          B_3: Buffer(B_2, float32, [(stride_2*n)], [], type="auto")[(ax0*stride_2)] = 0f32
-          for (k.inner.v: int32, 0, 16) {
-            B_3[(ax0*stride_2)] = (B_3[(ax0*stride_2)] + B.rf_1[((k.inner.v*n) + ax0)])
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_2,), type="auto")
+            B_rf = T.allocate([n * 16], "float32", "global")
+            B_rf_1 = T.buffer_decl((16 * n,), data=B_rf)
+            for k_inner, i in T.grid(16, n):
+                B_rf_1[k_inner * n + i] = T.float32(0)
+                for k_outer in range((m + 15) // 16):
+                    if T.likely(k_outer * 16 + k_inner < m):
+                        A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                        B_rf_1[k_inner * n + i] = B_rf_1[k_inner * n + i] + A_2[i * stride + (k_outer * 16 + k_inner) * stride_1]
+            for ax0 in range(n):
+                B_2 = T.buffer_decl((stride_2 * n,), data=B_1.data, type="auto")
+                B_2[ax0 * stride_2] = T.float32(0)
+                for k_inner_v in range(16):
+                    B_2[ax0 * stride_2] = B_2[ax0 * stride_2] + B_rf_1[k_inner_v * n + ax0]
 
 
 
@@ -479,25 +480,24 @@ Here is an example for 2D convolution with filter size = [3, 3] and strides = [1
 
  .. code-block:: none
 
-    @main = primfn(Input_1: handle, Filter_1: handle, Output_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {Input: Buffer(Input_2: Pointer(float32), float32, [n: int32, n], [stride: int32, stride_1: int32], type="auto"),
-                 Filter: Buffer(Filter_2: Pointer(float32), float32, [3, 3], []),
-                 Output: Buffer(Output_2: Pointer(float32), float32, [(n - 2), (n - 2)], [])}
-      buffer_map = {Input_1: Input, Filter_1: Filter, Output_1: Output} {
-      for (i: int32, 0, (n - 2)) {
-        for (j: int32, 0, (n - 2)) {
-          Output_3: Buffer(Output_2, float32, [((n - 2)*(n - 2))], [])[((i*(n - 2)) + j)] = 0f32
-          for (di: int32, 0, 3) {
-            for (dj: int32, 0, 3) {
-              Output_3[((i*(n - 2)) + j)] = (Output_3[((i*(n - 2)) + j)] + (Input_3: Buffer(Input_2, float32, [(stride*n)], [], type="auto")[(((i + di)*stride) + ((j + dj)*stride_1))]*Filter_3: Buffer(Filter_2, float32, [9], [])[((di*3) + dj)]))
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(Input: T.handle, Filter: T.handle, Output: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            Input_1 = T.match_buffer(Input, (n, n), strides=(stride, stride_1), type="auto")
+            Filter_1 = T.match_buffer(Filter, (3, 3))
+            Output_1 = T.match_buffer(Output, (n - 2, n - 2))
+            for i, j in T.grid(n - 2, n - 2):
+                Output_2 = T.buffer_decl(((n - 2) * (n - 2),), data=Output_1.data)
+                Output_2[i * (n - 2) + j] = T.float32(0)
+                for di, dj in T.grid(3, 3):
+                    Input_2 = T.buffer_decl((stride * n,), data=Input_1.data, type="auto")
+                    Filter_2 = T.buffer_decl((9,), data=Filter_1.data)
+                    Output_2[i * (n - 2) + j] = Output_2[i * (n - 2) + j] + Input_2[(i + di) * stride + (j + dj) * stride_1] * Filter_2[di * 3 + dj]
 
 
 
diff --git a/docs/_sources/how_to/work_with_schedules/scan.rst.txt b/docs/_sources/how_to/work_with_schedules/scan.rst.txt
index 784f47d821..a349227ce8 100644
--- a/docs/_sources/how_to/work_with_schedules/scan.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/scan.rst.txt
@@ -122,27 +122,33 @@ To split on the time iteration, user can schedule on scan_op.scan_axis instead.
 
  .. code-block:: none
 
-    @main = primfn(X_1: handle, scan_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {X: Buffer(X_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 scan: Buffer(scan_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
-      buffer_map = {X_1: X, scan_1: scan} {
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 255), 256);
-      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 256;
-      if @tir.likely((((blockIdx.x*256) + threadIdx.x) < n), dtype=bool) {
-        scan_3: Buffer(scan_2, float32, [(stride_2*m)], [], type="auto")[(((blockIdx.x*256) + threadIdx.x)*stride_3)] = X_3: Buffer(X_2, float32, [(stride*m)], [], type="auto")[(((blockIdx.x*256) + threadIdx.x)*stride_1)]
-      }
-      for (scan.idx: int32, 0, (m - 1)) {
-        attr [IterVar(blockIdx.x, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 255), 256);
-        attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 256;
-        if @tir.likely((((blockIdx.x*256) + threadIdx.x) < n), dtype=bool) {
-          let cse_var_1: int32 = (scan.idx + 1)
-          scan_3[((cse_var_1*stride_2) + (((blockIdx.x*256) + threadIdx.x)*stride_3))] = (scan_3[((scan.idx*stride_2) + (((blockIdx.x*256) + threadIdx.x)*stride_3))] + X_3[((cse_var_1*stride) + (((blockIdx.x*256) + threadIdx.x)*stride_1))])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(X: T.handle, scan: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            X_1 = T.match_buffer(X, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            scan_1 = T.match_buffer(scan, (m, n), strides=(stride_2, stride_3), type="auto")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            threadIdx_x = T.env_thread("threadIdx.x")
+            scan_2 = T.buffer_decl((stride_2 * m,), data=scan_1.data, type="auto")
+            X_2 = T.buffer_decl((stride * m,), data=X_1.data, type="auto")
+            with T.launch_thread(blockIdx_x, (n + 255) // 256):
+                T.launch_thread(threadIdx_x, 256)
+                if T.likely(blockIdx_x * 256 + threadIdx_x < n):
+                    scan_2[(blockIdx_x * 256 + threadIdx_x) * stride_3] = X_2[(blockIdx_x * 256 + threadIdx_x) * stride_1]
+            for scan_idx in range(m - 1):
+                T.launch_thread(blockIdx_x, (n + 255) // 256)
+                T.launch_thread(threadIdx_x, 256)
+                if T.likely(blockIdx_x * 256 + threadIdx_x < n):
+                    cse_var_1: T.int32 = scan_idx + 1
+                    scan_2[cse_var_1 * stride_2 + (blockIdx_x * 256 + threadIdx_x) * stride_3] = scan_2[scan_idx * stride_2 + (blockIdx_x * 256 + threadIdx_x) * stride_3] + X_2[cse_var_1 * stride + (blockIdx_x * 256 + threadIdx_x) * stride_1]
 
 
 
@@ -232,35 +238,34 @@ the body of scan to be compute_at locations outside the scan loop.
 
  .. code-block:: none
 
-    @main = primfn(X_1: handle, scan_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {X: Buffer(X_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 scan: Buffer(scan_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
-      buffer_map = {X_1: X, scan_1: scan} {
-      allocate(s1: Pointer(global float32), float32, [32]), storage_scope = global {
-        for (i: int32, 0, n) {
-          scan_3: Buffer(scan_2, float32, [(stride_2*m)], [], type="auto")[(i*stride_3)] = X_3: Buffer(X_2, float32, [(stride*m)], [], type="auto")[(i*stride_1)]
-        }
-        for (scan.idx: int32, 0, (m - 1)) {
-          for (i.outer: int32, 0, floordiv((n + 31), 32)) {
-            for (i_1: int32, 0, 32) {
-              if @tir.likely((((i.outer*32) + i_1) < n), dtype=bool) {
-                s1_1: Buffer(s1, float32, [32], [])[i_1] = (scan_3[((scan.idx*stride_2) + (((i.outer*32) + i_1)*stride_3))]*2f32)
-              }
-            }
-            for (i.inner: int32, 0, 32) {
-              if @tir.likely((((i.outer*32) + i.inner) < n), dtype=bool) {
-                let cse_var_2: int32 = (scan.idx + 1)
-                let cse_var_1: int32 = ((i.outer*32) + i.inner)
-                scan_3[((cse_var_2*stride_2) + (cse_var_1*stride_3))] = (s1_1[i.inner] + X_3[((cse_var_2*stride) + (cse_var_1*stride_1))])
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(X: T.handle, scan: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            X_1 = T.match_buffer(X, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            scan_1 = T.match_buffer(scan, (m, n), strides=(stride_2, stride_3), type="auto")
+            s1 = T.allocate([32], "float32", "global")
+            scan_2 = T.buffer_decl((stride_2 * m,), data=scan_1.data, type="auto")
+            X_2 = T.buffer_decl((stride * m,), data=X_1.data, type="auto")
+            for i in range(n):
+                scan_2[i * stride_3] = X_2[i * stride_1]
+            for scan_idx, i_outer in T.grid(m - 1, (n + 31) // 32):
+                s1_1 = T.buffer_decl((32,), data=s1)
+                for i in range(32):
+                    if T.likely(i_outer * 32 + i < n):
+                        s1_1[i] = scan_2[scan_idx * stride_2 + (i_outer * 32 + i) * stride_3] * T.float32(2)
+                for i_inner in range(32):
+                    if T.likely(i_outer * 32 + i_inner < n):
+                        cse_var_2: T.int32 = scan_idx + 1
+                        cse_var_1: T.int32 = i_outer * 32 + i_inner
+                        scan_2[cse_var_2 * stride_2 + cse_var_1 * stride_3] = s1_1[i_inner] + X_2[cse_var_2 * stride + cse_var_1 * stride_1]
 
 
 
@@ -302,30 +307,36 @@ The following example demonstrates how we can build recurrence with two states.
 
  .. code-block:: none
 
-    @main = primfn(X_1: handle, scan_2: handle, scan_3: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {X: Buffer(X_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 scan: Buffer(scan_4: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
-                 scan_1: Buffer(scan_5: Pointer(float32), float32, [m, l: int32], [stride_4: int32, stride_5: int32], type="auto")}
-      buffer_map = {X_1: X, scan_2: scan, scan_3: scan_1} {
-      for (i: int32, 0, n) {
-        scan_6: Buffer(scan_4, float32, [(stride_2*m)], [], type="auto")[(i*stride_3)] = X_3: Buffer(X_2, float32, [(stride*m)], [], type="auto")[(i*stride_1)]
-      }
-      for (i_1: int32, 0, l) {
-        scan_7: Buffer(scan_5, float32, [(stride_4*m)], [], type="auto")[(i_1*stride_5)] = 0f32
-      }
-      for (scan.idx: int32, 0, (m - 1)) {
-        for (i_2: int32, 0, n) {
-          let cse_var_1: int32 = (scan.idx + 1)
-          scan_6[((cse_var_1*stride_2) + (i_2*stride_3))] = (scan_6[((scan.idx*stride_2) + (i_2*stride_3))] + X_3[((cse_var_1*stride) + (i_2*stride_1))])
-        }
-        for (i_3: int32, 0, l) {
-          scan_7[(((scan.idx + 1)*stride_4) + (i_3*stride_5))] = (scan_7[((scan.idx*stride_4) + (i_3*stride_5))] + scan_6[(scan.idx*stride_2)])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(X: T.handle, scan: T.handle, scan_1: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            X_1 = T.match_buffer(X, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            scan_2 = T.match_buffer(scan, (m, n), strides=(stride_2, stride_3), type="auto")
+            l = T.var("int32")
+            stride_4 = T.var("int32")
+            stride_5 = T.var("int32")
+            scan_3 = T.match_buffer(scan_1, (m, l), strides=(stride_4, stride_5), type="auto")
+            scan_4 = T.buffer_decl((stride_2 * m,), data=scan_2.data, type="auto")
+            X_2 = T.buffer_decl((stride * m,), data=X_1.data, type="auto")
+            for i in range(n):
+                scan_4[i * stride_3] = X_2[i * stride_1]
+            scan_5 = T.buffer_decl((stride_4 * m,), data=scan_3.data, type="auto")
+            for i in range(l):
+                scan_5[i * stride_5] = T.float32(0)
+            for scan_idx in range(m - 1):
+                for i in range(n):
+                    cse_var_1: T.int32 = scan_idx + 1
+                    scan_4[cse_var_1 * stride_2 + i * stride_3] = scan_4[scan_idx * stride_2 + i * stride_3] + X_2[cse_var_1 * stride + i * stride_1]
+                for i in range(l):
+                    scan_5[(scan_idx + 1) * stride_4 + i * stride_5] = scan_5[scan_idx * stride_4 + i * stride_5] + scan_4[scan_idx * stride_2]
 
 
 
diff --git a/docs/_sources/how_to/work_with_schedules/schedule_primitives.rst.txt b/docs/_sources/how_to/work_with_schedules/schedule_primitives.rst.txt
index 0ad18d5e65..1320d7c57d 100644
--- a/docs/_sources/how_to/work_with_schedules/schedule_primitives.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/schedule_primitives.rst.txt
@@ -108,20 +108,27 @@ schedule computes tensor in a serial manner in a row-major order.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m, n], [stride_4: int32, stride_5: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, m) {
-        for (j: int32, 0, n) {
-          C_3: Buffer(C_2, float32, [(stride_4*m)], [], type="auto")[((i*stride_4) + (j*stride_5))] = (A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[((i*stride) + (j*stride_1))]*B_3: Buffer(B_2, float32, [(stride_2*m)], [], type="auto")[((i*stride_2) + (j*stride_3))])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            B_1 = T.match_buffer(B, (m, n), strides=(stride_2, stride_3), type="auto")
+            stride_4 = T.var("int32")
+            stride_5 = T.var("int32")
+            C_1 = T.match_buffer(C, (m, n), strides=(stride_4, stride_5), type="auto")
+            for i, j in T.grid(m, n):
+                C_2 = T.buffer_decl((stride_4 * m,), data=C_1.data, type="auto")
+                A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                B_2 = T.buffer_decl((stride_2 * m,), data=B_1.data, type="auto")
+                C_2[i * stride_4 + j * stride_5] = A_2[i * stride + j * stride_1] * B_2[i * stride_2 + j * stride_3]
 
 
 
@@ -158,22 +165,22 @@ split
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.outer: int32, 0, floordiv((m + 31), 32)) {
-        for (i.inner: int32, 0, 32) {
-          if @tir.likely((((i.outer*32) + i.inner) < m), dtype=bool) {
-            let cse_var_1: int32 = ((i.outer*32) + i.inner)
-            B_3: Buffer(B_2, float32, [(stride_1*m)], [], type="auto")[(cse_var_1*stride_1)] = (A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[(cse_var_1*stride)]*2f32)
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            for i_outer, i_inner in T.grid((m + 31) // 32, 32):
+                if T.likely(i_outer * 32 + i_inner < m):
+                    B_2 = T.buffer_decl((stride_1 * m,), data=B_1.data, type="auto")
+                    A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                    cse_var_1: T.int32 = i_outer * 32 + i_inner
+                    B_2[cse_var_1 * stride_1] = A_2[cse_var_1 * stride] * T.float32(2)
 
 
 
@@ -202,21 +209,21 @@ contrary with :code:`factor`.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.outer: int32, 0, 32) {
-        for (i.inner: int32, 0, floordiv((m + 31), 32)) {
-          if @tir.likely(((i.inner + (i.outer*floordiv((m + 31), 32))) < m), dtype=bool) {
-            B_3: Buffer(B_2, float32, [(stride_1*m)], [], type="auto")[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride_1)] = A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride)]
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            for i_outer, i_inner in T.grid(32, (m + 31) // 32):
+                if T.likely(i_inner + i_outer * ((m + 31) // 32) < m):
+                    B_2 = T.buffer_decl((stride_1 * m,), data=B_1.data, type="auto")
+                    A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                    B_2[(i_inner + i_outer * ((m + 31) // 32)) * stride_1] = A_2[(i_inner + i_outer * ((m + 31) // 32)) * stride]
 
 
 
@@ -247,29 +254,28 @@ axes.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.outer: int32, 0, floordiv((m + 9), 10)) {
-        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
-          for (i.inner: int32, 0, 10) {
-            if @tir.likely((((i.outer*10) + i.inner) < m), dtype=bool) {
-              for (j.inner: int32, 0, 5) {
-                if @tir.likely((((j.outer*5) + j.inner) < n), dtype=bool) {
-                  let cse_var_2: int32 = ((j.outer*5) + j.inner)
-                  let cse_var_1: int32 = ((i.outer*10) + i.inner)
-                  B_3: Buffer(B_2, float32, [(stride_2*m)], [], type="auto")[((cse_var_1*stride_2) + (cse_var_2*stride_3))] = A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[((cse_var_1*stride) + (cse_var_2*stride_1))]
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            B_1 = T.match_buffer(B, (m, n), strides=(stride_2, stride_3), type="auto")
+            for i_outer, j_outer, i_inner in T.grid((m + 9) // 10, (n + 4) // 5, 10):
+                if T.likely(i_outer * 10 + i_inner < m):
+                    for j_inner in range(5):
+                        if T.likely(j_outer * 5 + j_inner < n):
+                            cse_var_2: T.int32 = j_outer * 5 + j_inner
+                            cse_var_1: T.int32 = i_outer * 10 + i_inner
+                            B_2 = T.buffer_decl((stride_2 * m,), data=B_1.data, type="auto")
+                            A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                            B_2[cse_var_1 * stride_2 + cse_var_2 * stride_3] = A_2[cse_var_1 * stride + cse_var_2 * stride_1]
 
 
 
@@ -302,27 +308,27 @@ fuse
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.outer: int32, 0, floordiv((m + 9), 10)) {
-        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
-          for (i.inner.j.inner.fused: int32, 0, 50) {
-            if @tir.likely((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5)) < m), dtype=bool) {
-              if @tir.likely((((j.outer*5) + floormod(i.inner.j.inner.fused, 5)) < n), dtype=bool) {
-                let cse_var_2: int32 = ((j.outer*5) + floormod(i.inner.j.inner.fused, 5))
-                let cse_var_1: int32 = ((i.outer*10) + floordiv(i.inner.j.inner.fused, 5))
-                B_3: Buffer(B_2, float32, [(stride_2*m)], [], type="auto")[((cse_var_1*stride_2) + (cse_var_2*stride_3))] = A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[((cse_var_1*stride) + (cse_var_2*stride_1))]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            B_1 = T.match_buffer(B, (m, n), strides=(stride_2, stride_3), type="auto")
+            for i_outer, j_outer, i_inner_j_inner_fused in T.grid((m + 9) // 10, (n + 4) // 5, 50):
+                if T.likely(i_outer * 10 + i_inner_j_inner_fused // 5 < m):
+                    if T.likely(j_outer * 5 + i_inner_j_inner_fused % 5 < n):
+                        cse_var_2: T.int32 = j_outer * 5 + i_inner_j_inner_fused % 5
+                        cse_var_1: T.int32 = i_outer * 10 + i_inner_j_inner_fused // 5
+                        B_2 = T.buffer_decl((stride_2 * m,), data=B_1.data, type="auto")
+                        A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                        B_2[cse_var_1 * stride_2 + cse_var_2 * stride_3] = A_2[cse_var_1 * stride + cse_var_2 * stride_1]
 
 
 
@@ -355,29 +361,28 @@ reorder
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      for (i.inner: int32, 0, 10) {
-        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
-          for (i.outer: int32, 0, floordiv((m + 9), 10)) {
-            if @tir.likely((((i.outer*10) + i.inner) < m), dtype=bool) {
-              for (j.inner: int32, 0, 5) {
-                if @tir.likely((((j.outer*5) + j.inner) < n), dtype=bool) {
-                  let cse_var_2: int32 = ((j.outer*5) + j.inner)
-                  let cse_var_1: int32 = ((i.outer*10) + i.inner)
-                  B_3: Buffer(B_2, float32, [(stride_2*m)], [], type="auto")[((cse_var_1*stride_2) + (cse_var_2*stride_3))] = A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[((cse_var_1*stride) + (cse_var_2*stride_1))]
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            B_1 = T.match_buffer(B, (m, n), strides=(stride_2, stride_3), type="auto")
+            for i_inner, j_outer, i_outer in T.grid(10, (n + 4) // 5, (m + 9) // 10):
+                if T.likely(i_outer * 10 + i_inner < m):
+                    for j_inner in range(5):
+                        if T.likely(j_outer * 5 + j_inner < n):
+                            cse_var_2: T.int32 = j_outer * 5 + j_inner
+                            cse_var_1: T.int32 = i_outer * 10 + i_inner
+                            B_2 = T.buffer_decl((stride_2 * m,), data=B_1.data, type="auto")
+                            A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                            B_2[cse_var_1 * stride_2 + cse_var_2 * stride_3] = A_2[cse_var_1 * stride + cse_var_2 * stride_1]
 
 
 
@@ -410,19 +415,24 @@ in gpu programming.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_1: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B} {
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 63), 64);
-      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 64;
-      if @tir.likely((((blockIdx.x*64) + threadIdx.x) < n), dtype=bool) {
-        B_3: Buffer(B_2, float32, [(stride_1*n)], [], type="auto")[(((blockIdx.x*64) + threadIdx.x)*stride_1)] = (A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[(((blockIdx.x*64) + threadIdx.x)*stride)]*2f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (n,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_1,), type="auto")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, (n + 63) // 64)
+            threadIdx_x = T.env_thread("threadIdx.x")
+            T.launch_thread(threadIdx_x, 64)
+            if T.likely(blockIdx_x * 64 + threadIdx_x < n):
+                B_2 = T.buffer_decl((stride_1 * n,), data=B_1.data, type="auto")
+                A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                B_2[(blockIdx_x * 64 + threadIdx_x) * stride_1] = A_2[(blockIdx_x * 64 + threadIdx_x) * stride] * T.float32(2)
 
 
 
@@ -453,21 +463,25 @@ tensors at the root separately by default.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, m) {
-        B_3: Buffer(B_2, float32, [(stride_1*m)], [], type="auto")[(i*stride_1)] = (A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[(i*stride)] + 1f32)
-      }
-      for (i_1: int32, 0, m) {
-        C_3: Buffer(C_2, float32, [(stride_2*m)], [], type="auto")[(i_1*stride_2)] = (B_3[(i_1*stride_1)]*2f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (m,), strides=(stride_2,), type="auto")
+            B_2 = T.buffer_decl((stride_1 * m,), data=B_1.data, type="auto")
+            for i in range(m):
+                A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                B_2[i * stride_1] = A_2[i * stride] + T.float32(1)
+            for i in range(m):
+                C_2 = T.buffer_decl((stride_2 * m,), data=C_1.data, type="auto")
+                C_2[i * stride_2] = B_2[i * stride_1] * T.float32(2)
 
 
 
@@ -497,19 +511,24 @@ of computation of `C`.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, m) {
-        B_3: Buffer(B_2, float32, [(stride_1*m)], [], type="auto")[(i*stride_1)] = (A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[(i*stride)] + 1f32)
-        C_3: Buffer(C_2, float32, [(stride_2*m)], [], type="auto")[(i*stride_2)] = (B_3[(i*stride_1)]*2f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (m,), strides=(stride_2,), type="auto")
+            for i in range(m):
+                B_2 = T.buffer_decl((stride_1 * m,), data=B_1.data, type="auto")
+                A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                B_2[i * stride_1] = A_2[i * stride] + T.float32(1)
+                C_2 = T.buffer_decl((stride_2 * m,), data=C_1.data, type="auto")
+                C_2[i * stride_2] = B_2[i * stride_1] * T.float32(2)
 
 
 
@@ -542,18 +561,22 @@ tensor is required.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, m) {
-        C_3: Buffer(C_2, float32, [(stride_2*m)], [], type="auto")[(i*stride_2)] = ((A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[(i*stride)] + 1f32)*2f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (m,), strides=(stride_2,), type="auto")
+            for i in range(m):
+                C_2 = T.buffer_decl((stride_2 * m,), data=C_1.data, type="auto")
+                A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                C_2[i * stride_2] = (A_2[i * stride] + T.float32(1)) * T.float32(2)
 
 
 
@@ -585,21 +608,25 @@ compute_root
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [m: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [m], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, m) {
-        B_3: Buffer(B_2, float32, [(stride_1*m)], [], type="auto")[(i*stride_1)] = (A_3: Buffer(A_2, float32, [(stride*m)], [], type="auto")[(i*stride)] + 1f32)
-      }
-      for (i_1: int32, 0, m) {
-        C_3: Buffer(C_2, float32, [(stride_2*m)], [], type="auto")[(i_1*stride_2)] = (B_3[(i_1*stride_1)]*2f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (m,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (m,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (m,), strides=(stride_2,), type="auto")
+            B_2 = T.buffer_decl((stride_1 * m,), data=B_1.data, type="auto")
+            for i in range(m):
+                A_2 = T.buffer_decl((stride * m,), data=A_1.data, type="auto")
+                B_2[i * stride_1] = A_2[i * stride] + T.float32(1)
+            for i in range(m):
+                C_2 = T.buffer_decl((stride_2 * m,), data=C_1.data, type="auto")
+                C_2[i * stride_2] = B_2[i * stride_1] * T.float32(2)
 
 
 
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 63ef55149b..e104c0dd81 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
 =================
-**00:07.216** total execution time for **how_to_work_with_schedules** files:
+**00:06.147** total execution time for **how_to_work_with_schedules** files:
 
 +------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_intrin_math.py` (``intrin_math.py``)                 | 00:04.707 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_intrin_math.py` (``intrin_math.py``)                 | 00:03.642 | 0.0 MB |
 +------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_tensorize.py` (``tensorize.py``)                     | 00:01.148 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_tensorize.py` (``tensorize.py``)                     | 00:01.152 | 0.0 MB |
 +------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_reduction.py` (``reduction.py``)                     | 00:00.577 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_reduction.py` (``reduction.py``)                     | 00:00.572 | 0.0 MB |
 +------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_scan.py` (``scan.py``)                               | 00:00.556 | 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``)                     | 00:00.125 | 0.0 MB |
+| :ref:`sphx_glr_how_to_work_with_schedules_extern_op.py` (``extern_op.py``)                     | 00:00.114 | 0.0 MB |
 +------------------------------------------------------------------------------------------------+-----------+--------+
 | :ref:`sphx_glr_how_to_work_with_schedules_schedule_primitives.py` (``schedule_primitives.py``) | 00:00.049 | 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_tedd.py` (``tedd.py``)                               | 00:00.032 | 0.0 MB |
 +------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_how_to_work_with_schedules_tuple_inputs.py` (``tuple_inputs.py``)               | 00:00.023 | 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 0f90d71454..c9303a792d 100644
--- a/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/tensorize.rst.txt
@@ -87,24 +87,22 @@ The following lines describe the computation :code:`A * B^T` in TVM.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 64], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [512, 64], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 512], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, 1024) {
-        for (j: int32, 0, 512) {
-          C_3: Buffer(C_2, float32, [524288], [])[((i*512) + j)] = 0f32
-          for (k: int32, 0, 64) {
-            let cse_var_1: int32 = ((i*512) + j)
-            C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [65536], [])[((i*64) + k)]*B_3: Buffer(B_2, float32, [32768], [])[((j*64) + k)]))
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 64))
+            B_1 = T.match_buffer(B, (512, 64))
+            C_1 = T.match_buffer(C, (1024, 512))
+            for i, j in T.grid(1024, 512):
+                C_2 = T.buffer_decl((524288,), data=C_1.data)
+                C_2[i * 512 + j] = T.float32(0)
+                for k in range(64):
+                    cse_var_1: T.int32 = i * 512 + j
+                    A_2 = T.buffer_decl((65536,), data=A_1.data)
+                    B_2 = T.buffer_decl((32768,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[i * 64 + k] * B_2[j * 64 + k]
 
 
 
@@ -139,26 +137,22 @@ Thus we break down the matmul loops to make the innermost loops a (16x64) GEMV.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 64], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [512, 64], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 512], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, 1024) {
-        for (j.outer: int32, 0, 32) {
-          for (j.inner: int32, 0, 16) {
-            C_3: Buffer(C_2, float32, [524288], [])[(((i*512) + (j.outer*16)) + j.inner)] = 0f32
-            for (k: int32, 0, 64) {
-              let cse_var_1: int32 = (((i*512) + (j.outer*16)) + j.inner)
-              C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [65536], [])[((i*64) + k)]*B_3: Buffer(B_2, float32, [32768], [])[(((j.outer*1024) + (j.inner*64)) + k)]))
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 64))
+            B_1 = T.match_buffer(B, (512, 64))
+            C_1 = T.match_buffer(C, (1024, 512))
+            for i, j_outer, j_inner in T.grid(1024, 32, 16):
+                C_2 = T.buffer_decl((524288,), data=C_1.data)
+                C_2[i * 512 + j_outer * 16 + j_inner] = T.float32(0)
+                for k in range(64):
+                    cse_var_1: T.int32 = i * 512 + j_outer * 16 + j_inner
+                    A_2 = T.buffer_decl((65536,), data=A_1.data)
+                    B_2 = T.buffer_decl((32768,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[i * 64 + k] * B_2[j_outer * 1024 + j_inner * 64 + k]
 
 
 
@@ -263,20 +257,16 @@ such placeholder can be put to let TVM automatically bind the inferred value for
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 64], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [512, 64], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 512], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, 1024) {
-        for (j.outer: int32, 0, 32) {
-          @tir.call_extern("gemv_update", @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), C_2, ((i*512) + (j.outer*16)), 16, 2, dtype=handle), @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), A_2, (i*64), 64, 1, dtype=handle), @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), B_2, (j.outer*1024), 1024, 1, dtype=handle), 16, 64, 64, dtype=int32)
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 64))
+            B_1 = T.match_buffer(B, (512, 64))
+            C_1 = T.match_buffer(C, (1024, 512))
+            for i, j_outer in T.grid(1024, 32):
+                T.call_extern("int32", "gemv_update", T.tvm_access_ptr(T.type_annotation("float32"), C_1.data, i * 512 + j_outer * 16, 16, 2), T.tvm_access_ptr(T.type_annotation("float32"), A_1.data, i * 64, 64, 1), T.tvm_access_ptr(T.type_annotation("float32"), B_1.data, j_outer * 1024, 1024, 1), 16, 64, 64)
 
 
 
@@ -341,21 +331,18 @@ The importing needs to happen before the tensorized GEMV being executed.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 64], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [512, 64], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 512], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      attr [IterVar(i: int32, (nullptr), "DataPar", "")] "pragma_import_llvm" = "; ModuleID = '/tmp/tmp1h0clzsh/input0.cc'\nsource_filename = \"/tmp/tmp1h0clzsh/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 floa [...]
-      for (i, 0, 1024) {
-        for (j.outer: int32, 0, 32) {
-          @tir.call_extern("gemv_update", @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), C_2, ((i*512) + (j.outer*16)), 16, 2, dtype=handle), @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), A_2, (i*64), 64, 1, dtype=handle), @tir.tvm_access_ptr(@tir.type_annotation(, dtype=float32), B_2, (j.outer*1024), 1024, 1, dtype=handle), 16, 64, 64, dtype=int32)
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 64))
+            B_1 = T.match_buffer(B, (512, 64))
+            C_1 = T.match_buffer(C, (1024, 512))
+            i = T.var("int32")
+            T.attr(T.iter_var(i, None, "DataPar", ""), "pragma_import_llvm", "; ModuleID = '/tmp/tmp7xql9g8a/input0.cc'\nsource_filename = \"/tmp/tmp7xql9g8a/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_1.data, i * 512 + j_outer * 16, 16, 2), T.tvm_access_ptr(T.type_annotation("float32"), A_1.data, i * 64, 64, 1), T.tvm_access_ptr(T.type_annotation("float32"), B_1.data, j_outer * 1024, 1024, 1), 16, 64, 64)
 
 
 
diff --git a/docs/_sources/how_to/work_with_schedules/tuple_inputs.rst.txt b/docs/_sources/how_to/work_with_schedules/tuple_inputs.rst.txt
index b1a5987e41..74fe801aee 100644
--- a/docs/_sources/how_to/work_with_schedules/tuple_inputs.rst.txt
+++ b/docs/_sources/how_to/work_with_schedules/tuple_inputs.rst.txt
@@ -81,22 +81,32 @@ together in the next schedule procedure.
 
  .. code-block:: none
 
-    @main = primfn(A0_1: handle, A1_1: handle, B_2: handle, B_3: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A0: Buffer(A0_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 A1: Buffer(A1_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
-                 B: Buffer(B_4: Pointer(float32), float32, [m, n], [stride_4: int32, stride_5: int32], type="auto"),
-                 B_1: Buffer(B_5: Pointer(float32), float32, [m, n], [stride_6: int32, stride_7: int32], type="auto")}
-      buffer_map = {A0_1: A0, A1_1: A1, B_2: B, B_3: B_1} {
-      for (i: int32, 0, m) {
-        for (j: int32, 0, n) {
-          B_6: Buffer(B_4, float32, [(stride_4*m)], [], type="auto")[((i*stride_4) + (j*stride_5))] = (A0_3: Buffer(A0_2, float32, [(stride*m)], [], type="auto")[((i*stride) + (j*stride_1))] + 2f32)
-          B_7: Buffer(B_5, float32, [(stride_6*m)], [], type="auto")[((i*stride_6) + (j*stride_7))] = (A1_3: Buffer(A1_2, float32, [(stride_2*m)], [], type="auto")[((i*stride_2) + (j*stride_3))]*3f32)
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A0: T.handle, A1: T.handle, B: T.handle, B_1: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A0_1 = T.match_buffer(A0, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            A1_1 = T.match_buffer(A1, (m, n), strides=(stride_2, stride_3), type="auto")
+            stride_4 = T.var("int32")
+            stride_5 = T.var("int32")
+            B_2 = T.match_buffer(B, (m, n), strides=(stride_4, stride_5), type="auto")
+            stride_6 = T.var("int32")
+            stride_7 = T.var("int32")
+            B_3 = T.match_buffer(B_1, (m, n), strides=(stride_6, stride_7), type="auto")
+            for i, j in T.grid(m, n):
+                B_4 = T.buffer_decl((stride_4 * m,), data=B_2.data, type="auto")
+                A0_2 = T.buffer_decl((stride * m,), data=A0_1.data, type="auto")
+                B_4[i * stride_4 + j * stride_5] = A0_2[i * stride + j * stride_1] + T.float32(2)
+                B_5 = T.buffer_decl((stride_6 * m,), data=B_3.data, type="auto")
+                A1_2 = T.buffer_decl((stride_2 * m,), data=A1_1.data, type="auto")
+                B_5[i * stride_6 + j * stride_7] = A1_2[i * stride_2 + j * stride_3] * T.float32(3)
 
 
 
@@ -154,24 +164,33 @@ with :py:func:`te.comm_reducer` as below:
 
  .. code-block:: none
 
-    @main = primfn(idx_1: handle, val_1: handle, T_2: handle, T_3: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {idx: Buffer(idx_2: Pointer(int32), int32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 val: Buffer(val_2: Pointer(int32), int32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
-                 T: Buffer(T_4: Pointer(int32), int32, [m], [stride_4: int32], type="auto"),
-                 T_1: Buffer(T_5: Pointer(int32), int32, [m], [stride_5: int32], type="auto")}
-      buffer_map = {idx_1: idx, val_1: val, T_2: T, T_3: T_1} {
-      for (i: int32, 0, m) {
-        T_6: Buffer(T_4, int32, [(stride_4*m)], [], type="auto")[(i*stride_4)] = -1
-        T_7: Buffer(T_5, int32, [(stride_5*m)], [], type="auto")[(i*stride_5)] = -2147483648
-        for (k: int32, 0, n) {
-          T_6[(i*stride_4)] = @tir.if_then_else((val_3: Buffer(val_2, int32, [(stride_2*m)], [], type="auto")[((i*stride_2) + (k*stride_3))] <= T_7[(i*stride_5)]), T_6[(i*stride_4)], idx_3: Buffer(idx_2, int32, [(stride*m)], [], type="auto")[((i*stride) + (k*stride_1))], dtype=int32)
-          T_7[(i*stride_5)] = @tir.if_then_else((val_3[((i*stride_2) + (k*stride_3))] <= T_7[(i*stride_5)]), T_7[(i*stride_5)], val_3[((i*stride_2) + (k*stride_3))], dtype=int32)
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(idx: T.handle, val: T.handle, T: T.handle, T_1: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            idx_1 = T.match_buffer(idx, (m, n), "int32", strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            val_1 = T.match_buffer(val, (m, n), "int32", strides=(stride_2, stride_3), type="auto")
+            stride_4 = T.var("int32")
+            T_2 = T.match_buffer(T, (m,), "int32", strides=(stride_4,), type="auto")
+            stride_5 = T.var("int32")
+            T_3 = T.match_buffer(T_1, (m,), "int32", strides=(stride_5,), type="auto")
+            for i in range(m):
+                T_4 = T.buffer_decl((stride_4 * m,), "int32", data=T_2.data, type="auto")
+                T_4[i * stride_4] = -1
+                T_5 = T.buffer_decl((stride_5 * m,), "int32", data=T_3.data, type="auto")
+                T_5[i * stride_5] = -2147483648
+                for k in range(n):
+                    val_2 = T.buffer_decl((stride_2 * m,), "int32", data=val_1.data, type="auto")
+                    idx_2 = T.buffer_decl((stride * m,), "int32", data=idx_1.data, type="auto")
+                    T_4[i * stride_4] = T.if_then_else(val_2[i * stride_2 + k * stride_3] <= T_5[i * stride_5], T_4[i * stride_4], idx_2[i * stride + k * stride_1])
+                    T_5[i * stride_5] = T.if_then_else(val_2[i * stride_2 + k * stride_3] <= T_5[i * stride_5], T_5[i * stride_5], val_2[i * stride_2 + k * stride_3])
 
 
 
@@ -216,26 +235,35 @@ in terms of operation.
 
  .. code-block:: none
 
-    @main = primfn(A0_1: handle, A1_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A0: Buffer(A0_2: Pointer(float32), float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
-                 A1: Buffer(A1_2: Pointer(float32), float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [m, n], [stride_4: int32, stride_5: int32], type="auto")}
-      buffer_map = {A0_1: A0, A1_1: A1, C_1: C} {
-      allocate(B.v0: Pointer(global float32), float32, [n]), storage_scope = global;
-      allocate(B.v1: Pointer(global float32), float32, [n]), storage_scope = global;
-      for (i: int32, 0, m) {
-        for (j: int32, 0, n) {
-          B.v0_1: Buffer(B.v0, float32, [n], [])[j] = (A0_3: Buffer(A0_2, float32, [(stride*m)], [], type="auto")[((i*stride) + (j*stride_1))] + 2f32)
-          B.v1_1: Buffer(B.v1, float32, [n], [])[j] = (A0_3[((i*stride) + (j*stride_1))]*3f32)
-        }
-        for (j_1: int32, 0, n) {
-          C_3: Buffer(C_2, float32, [(stride_4*m)], [], type="auto")[((i*stride_4) + (j_1*stride_5))] = (A1_3: Buffer(A1_2, float32, [(stride_2*m)], [], type="auto")[((i*stride_2) + (j_1*stride_3))] + B.v0_1[j_1])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A0: T.handle, A1: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            m = T.var("int32")
+            n = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A0_1 = T.match_buffer(A0, (m, n), strides=(stride, stride_1), type="auto")
+            stride_2 = T.var("int32")
+            stride_3 = T.var("int32")
+            A1_1 = T.match_buffer(A1, (m, n), strides=(stride_2, stride_3), type="auto")
+            stride_4 = T.var("int32")
+            stride_5 = T.var("int32")
+            C_1 = T.match_buffer(C, (m, n), strides=(stride_4, stride_5), type="auto")
+            B_v0 = T.allocate([n], "float32", "global")
+            B_v1 = T.allocate([n], "float32", "global")
+            for i in range(m):
+                B_v0_1 = T.buffer_decl((n,), data=B_v0)
+                for j in range(n):
+                    A0_2 = T.buffer_decl((stride * m,), data=A0_1.data, type="auto")
+                    B_v0_1[j] = A0_2[i * stride + j * stride_1] + T.float32(2)
+                    B_v1_1 = T.buffer_decl((n,), data=B_v1)
+                    B_v1_1[j] = A0_2[i * stride + j * stride_1] * T.float32(3)
+                for j in range(n):
+                    C_2 = T.buffer_decl((stride_4 * m,), data=C_1.data, type="auto")
+                    A1_2 = T.buffer_decl((stride_2 * m,), data=A1_1.data, type="auto")
+                    C_2[i * stride_4 + j * stride_5] = A1_2[i * stride_2 + j * stride_3] + B_v0_1[j]
 
 
 
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 3007fd5017..f1a8e5e8a9 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
 =================
-**00:30.583** total execution time for **topic_vta_tutorials_autotvm** files:
+**00:29.042** total execution time for **topic_vta_tutorials_autotvm** files:
 
 +---------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``) | 00:30.577 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``) | 00:29.036 | 0.0 MB |
 +---------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_alu_vta.py` (``tune_alu_vta.py``)     | 00:00.007 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_autotvm_tune_alu_vta.py` (``tune_alu_vta.py``)     | 00:00.006 | 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 5c359bf157..7745e2a1c9 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 32.74s!
+    resnet18_v1 inference graph built in 30.82s!
 
 
 
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 581673a74d..d3141302c3 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 22.50s!
+    yolov3-tiny inference graph built in 21.20s!
 
 
 
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 6426a8e813..3fd895dcb3 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
 =================
-**01:39.500** total execution time for **topic_vta_tutorials_frontend** files:
+**01:35.524** total execution time for **topic_vta_tutorials_frontend** files:
 
 +------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``)           | 00:49.840 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``)           | 00:47.956 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``) | 00:49.659 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``) | 00:47.568 | 0.0 MB |
 +------------------------------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/topic/vta/tutorials/matrix_multiply.rst.txt b/docs/_sources/topic/vta/tutorials/matrix_multiply.rst.txt
index 7f11482d1f..a21a495d46 100644
--- a/docs/_sources/topic/vta/tutorials/matrix_multiply.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/matrix_multiply.rst.txt
@@ -346,52 +346,37 @@ After we construct the schedule, by default the schedule computes
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(int8), int8, [1, 16, 1, 16], []),
-                 B: Buffer(B_2: Pointer(int8), int8, [16, 16, 16, 16], []),
-                 C: Buffer(C_2: Pointer(int8), int8, [1, 16, 1, 16], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(A_buf: Pointer(global int8), int8, [256]), storage_scope = global;
-      allocate(B_buf: Pointer(global int8), int8, [65536]), storage_scope = global;
-      allocate(C_buf: Pointer(global int32), int32, [256]), storage_scope = global {
-        for (i1: int32, 0, 16) {
-          for (i3: int32, 0, 16) {
-            let cse_var_1: int32 = ((i1*16) + i3)
-            A_buf_1: Buffer(A_buf, int8, [256], [])[cse_var_1] = A_3: Buffer(A_2, int8, [256], [])[cse_var_1]
-          }
-        }
-        for (i0: int32, 0, 16) {
-          for (i1_1: int32, 0, 16) {
-            for (i2: int32, 0, 16) {
-              for (i3_1: int32, 0, 16) {
-                let cse_var_2: int32 = ((((i0*4096) + (i1_1*256)) + (i2*16)) + i3_1)
-                B_buf_1: Buffer(B_buf, int8, [65536], [])[cse_var_2] = B_3: Buffer(B_2, int8, [65536], [])[cse_var_2]
-              }
-            }
-          }
-        }
-        for (co: int32, 0, 16) {
-          for (ci: int32, 0, 16) {
-            C_buf_1: Buffer(C_buf, int32, [256], [])[((co*16) + ci)] = 0
-            for (ko: int32, 0, 16) {
-              for (ki: int32, 0, 16) {
-                let cse_var_3: int32 = ((co*16) + ci)
-                C_buf_1[cse_var_3] = (C_buf_1[cse_var_3] + (cast(int32, A_buf_1[((ko*16) + ki)])*cast(int32, B_buf_1[((((co*4096) + (ko*256)) + (ci*16)) + ki)])))
-              }
-            }
-          }
-        }
-        for (i1_2: int32, 0, 16) {
-          for (i3_2: int32, 0, 16) {
-            let cse_var_4: int32 = ((i1_2*16) + i3_2)
-            C_3: Buffer(C_2, int8, [256], [])[cse_var_4] = cast(int8, C_buf_1[cse_var_4])
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1, 16, 1, 16), "int8")
+            B_1 = T.match_buffer(B, (16, 16, 16, 16), "int8")
+            C_1 = T.match_buffer(C, (1, 16, 1, 16), "int8")
+            A_buf = T.allocate([256], "int8", "global")
+            B_buf = T.allocate([65536], "int8", "global")
+            C_buf = T.allocate([256], "int32", "global")
+            A_buf_1 = T.buffer_decl((256,), "int8", data=A_buf)
+            for i1, i3 in T.grid(16, 16):
+                cse_var_1: T.int32 = i1 * 16 + i3
+                A_2 = T.buffer_decl((256,), "int8", data=A_1.data)
+                A_buf_1[cse_var_1] = A_2[cse_var_1]
+            B_buf_1 = T.buffer_decl((65536,), "int8", data=B_buf)
+            for i0, i1, i2, i3 in T.grid(16, 16, 16, 16):
+                cse_var_2: T.int32 = i0 * 4096 + i1 * 256 + i2 * 16 + i3
+                B_2 = T.buffer_decl((65536,), "int8", data=B_1.data)
+                B_buf_1[cse_var_2] = B_2[cse_var_2]
+            C_buf_1 = T.buffer_decl((256,), "int32", data=C_buf)
+            for co, ci in T.grid(16, 16):
+                C_buf_1[co * 16 + ci] = 0
+                for ko, ki in T.grid(16, 16):
+                    cse_var_3: T.int32 = co * 16 + ci
+                    C_buf_1[cse_var_3] = C_buf_1[cse_var_3] + T.Cast("int32", A_buf_1[ko * 16 + ki]) * T.Cast("int32", B_buf_1[co * 4096 + ko * 256 + ci * 16 + ki])
+            for i1, i3 in T.grid(16, 16):
+                cse_var_4: T.int32 = i1 * 16 + i3
+                C_2 = T.buffer_decl((256,), "int8", data=C_1.data)
+                C_2[cse_var_4] = T.Cast("int8", C_buf_1[cse_var_4])
 
 
 
@@ -509,45 +494,42 @@ moving the copy operations into the matrix multiplication loop.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(int8), int8, [1, 16, 1, 16], []),
-                 B: Buffer(B_2: Pointer(int8), int8, [16, 16, 16, 16], []),
-                 C: Buffer(C_2: Pointer(int8), int8, [1, 16, 1, 16], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(C_buf: Pointer(local.acc_buffer int32), int32, [256]), storage_scope = local.acc_buffer;
-      allocate(A_buf: Pointer(local.inp_buffer int8), int8, [16]), storage_scope = local.inp_buffer;
-      allocate(B_buf: Pointer(local.wgt_buffer int8), int8, [16]), storage_scope = local.wgt_buffer {
-        for (co: int32, 0, 16) {
-          for (ci: int32, 0, 16) {
-            C_buf_1: Buffer(C_buf, int32, [256], [], scope="local.acc_buffer", align=16)[((co*16) + ci)] = 0
-            for (ko: int32, 0, 16) {
-              attr [IterVar(i0: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
-              for (i3: int32, 0, 16) {
-                A_buf_1: Buffer(A_buf, int8, [16], [], scope="local.inp_buffer", align=16)[i3] = A_3: Buffer(A_2, int8, [256], [])[((ko*16) + i3)]
-              }
-              attr [IterVar(i0_1: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
-              for (i3_1: int32, 0, 16) {
-                B_buf_1: Buffer(B_buf, int8, [16], [], scope="local.wgt_buffer", align=256)[i3_1] = B_3: Buffer(B_2, int8, [65536], [])[((((co*4096) + (ko*256)) + (ci*16)) + i3_1)]
-              }
-              for (ki: int32, 0, 16) {
-                let cse_var_1: int32 = ((co*16) + ci)
-                C_buf_1[cse_var_1] = (C_buf_1[cse_var_1] + (cast(int32, A_buf_1[ki])*cast(int32, B_buf_1[ki])))
-              }
-            }
-          }
-        }
-        attr [IterVar(i0_2: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
-        for (i1: int32, 0, 16) {
-          for (i3_2: int32, 0, 16) {
-            let cse_var_2: int32 = ((i1*16) + i3_2)
-            C_3: Buffer(C_2, int8, [256], [])[cse_var_2] = cast(int8, C_buf_1[cse_var_2])
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1, 16, 1, 16), "int8")
+            B_1 = T.match_buffer(B, (16, 16, 16, 16), "int8")
+            C_1 = T.match_buffer(C, (1, 16, 1, 16), "int8")
+            C_buf = T.allocate([256], "int32", "local.acc_buffer")
+            A_buf = T.allocate([16], "int8", "local.inp_buffer")
+            B_buf = T.allocate([16], "int8", "local.wgt_buffer")
+            C_buf_1 = T.buffer_decl((256,), "int32", data=C_buf, scope="local.acc_buffer", align=16)
+            for co, ci in T.grid(16, 16):
+                C_buf_1[co * 16 + ci] = 0
+                for ko in range(16):
+                    i0 = T.var("int32")
+                    A_buf_1 = T.buffer_decl((16,), "int8", data=A_buf, scope="local.inp_buffer", align=16)
+                    with T.attr(T.iter_var(i0, None, "DataPar", ""), "pragma_dma_copy", 1):
+                        for i3 in range(16):
+                            A_2 = T.buffer_decl((256,), "int8", data=A_1.data)
+                            A_buf_1[i3] = A_2[ko * 16 + i3]
+                    i0_1 = T.var("int32")
+                    B_buf_1 = T.buffer_decl((16,), "int8", data=B_buf, scope="local.wgt_buffer", align=256)
+                    with T.attr(T.iter_var(i0_1, None, "DataPar", ""), "pragma_dma_copy", 1):
+                        for i3 in range(16):
+                            B_2 = T.buffer_decl((65536,), "int8", data=B_1.data)
+                            B_buf_1[i3] = B_2[co * 4096 + ko * 256 + ci * 16 + i3]
+                    for ki in range(16):
+                        cse_var_1: T.int32 = co * 16 + ci
+                        C_buf_1[cse_var_1] = C_buf_1[cse_var_1] + T.Cast("int32", A_buf_1[ki]) * T.Cast("int32", B_buf_1[ki])
+            i0 = T.var("int32")
+            T.attr(T.iter_var(i0, None, "DataPar", ""), "pragma_dma_copy", 1)
+            for i1, i3 in T.grid(16, 16):
+                cse_var_2: T.int32 = i1 * 16 + i3
+                C_2 = T.buffer_decl((256,), "int8", data=C_1.data)
+                C_2[cse_var_2] = T.Cast("int8", C_buf_1[cse_var_2])
 
 
 
@@ -595,47 +577,40 @@ by the VTA runtime JIT compiler.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(int8), int8, [1, 16, 1, 16], []),
-                 B: Buffer(B_2: Pointer(int8), int8, [16, 16, 16, 16], []),
-                 C: Buffer(C_2: Pointer(int8), int8, [1, 16, 1, 16], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-          @tir.call_extern("VTAUopLoopBegin", 16, 1, 0, 0, dtype=int32)
-          @tir.vta.uop_push(0, 1, 0, 0, 0, 0, 0, 0, dtype=int32)
-          @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-        }
-        @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-      }
-      for (ko: int32, 0, 16) {
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
-          @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-          @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), A_2, ko, 1, 1, 1, 0, 0, 0, 0, 0, 2, dtype=int32)
-          @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), B_2, ko, 1, 16, 16, 0, 0, 0, 0, 0, 1, dtype=int32)
-          @tir.vta.coproc_dep_push(1, 2, dtype=int32)
-        }
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-          @tir.vta.coproc_dep_pop(1, 2, dtype=int32)
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-            @tir.call_extern("VTAUopLoopBegin", 16, 1, 0, 1, dtype=int32)
-            @tir.vta.uop_push(0, 0, 0, 0, 0, 0, 0, 0, dtype=int32)
-            @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-          }
-          @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-        }
-      }
-      @tir.vta.coproc_dep_push(2, 3, dtype=int32)
-      @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-      attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
-        @tir.vta.coproc_dep_pop(2, 3, dtype=int32)
-        @tir.call_extern("VTAStoreBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), 0, 4, C_2, 0, 16, 1, 16, dtype=int32)
-      }
-      @tir.vta.coproc_sync(, dtype=int32)
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1, 16, 1, 16), "int8")
+            B_1 = T.match_buffer(B, (16, 16, 16, 16), "int8")
+            C_1 = T.match_buffer(C, (1, 16, 1, 16), "int8")
+            vta = T.var("int32")
+            with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2):
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                    T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 0, 0)
+                    T.tir.vta.uop_push(0, 1, 0, 0, 0, 0, 0, 0)
+                    T.call_extern("int32", "VTAUopLoopEnd")
+                T.tir.vta.coproc_dep_push(2, 1)
+            for ko in range(16):
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 1):
+                    T.tir.vta.coproc_dep_pop(2, 1)
+                    T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), A_1.data, ko, 1, 1, 1, 0, 0, 0, 0, 0, 2)
+                    T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), B_1.data, ko, 1, 16, 16, 0, 0, 0, 0, 0, 1)
+                    T.tir.vta.coproc_dep_push(1, 2)
+                T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2)
+                T.tir.vta.coproc_dep_pop(1, 2)
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                    T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 0, 1)
+                    T.tir.vta.uop_push(0, 0, 0, 0, 0, 0, 0, 0)
+                    T.call_extern("int32", "VTAUopLoopEnd")
+                T.tir.vta.coproc_dep_push(2, 1)
+            T.tir.vta.coproc_dep_push(2, 3)
+            T.tir.vta.coproc_dep_pop(2, 1)
+            with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 3):
+                T.tir.vta.coproc_dep_pop(2, 3)
+                T.call_extern("int32", "VTAStoreBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), 0, 4, C_1.data, 0, 16, 1, 16)
+            T.tir.vta.coproc_sync()
 
 
 
diff --git a/docs/_sources/topic/vta/tutorials/optimize/convolution_opt.rst.txt b/docs/_sources/topic/vta/tutorials/optimize/convolution_opt.rst.txt
index 4f8c45b4ee..f8ab9652a4 100644
--- a/docs/_sources/topic/vta/tutorials/optimize/convolution_opt.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/optimize/convolution_opt.rst.txt
@@ -274,103 +274,50 @@ Those include:
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], []),
-                 kernel: Buffer(kernel_2: Pointer(int8), int8, [16, 16, 3, 3, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], [])}
-      buffer_map = {data_1: data, kernel_1: kernel, res_1: res} {
-      allocate(data_buf: Pointer(global int8), int8, [65536]), storage_scope = global;
-      allocate(kernel_buf: Pointer(global int8), int8, [589824]), storage_scope = global;
-      allocate(res_conv: Pointer(global int32), int32, [50176]), storage_scope = global {
-        for (i1: int32, 0, 16) {
-          for (i2: int32, 0, 16) {
-            for (i3: int32, 0, 16) {
-              for (i5: int32, 0, 16) {
-                let cse_var_1: int32 = (i3*16)
-                data_buf_1: Buffer(data_buf, int8, [65536], [])[((((i1*4096) + (i2*256)) + cse_var_1) + i5)] = @tir.if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), data_3: Buffer(data_2, int8, [50176], [])[(((((i1*3136) + (i2*224)) + cse_var_1) + i5) - 240)], 0i8, dtype=int8)
-              }
-            }
-          }
-        }
-        for (i0: int32, 0, 16) {
-          for (i1_1: int32, 0, 16) {
-            for (i2_1: int32, 0, 3) {
-              for (i3_1: int32, 0, 3) {
-                for (i4: int32, 0, 16) {
-                  for (i5_1: int32, 0, 16) {
-                    let cse_var_2: int32 = ((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)
-                    kernel_buf_1: Buffer(kernel_buf, int8, [589824], [])[cse_var_2] = kernel_3: Buffer(kernel_2, int8, [589824], [])[cse_var_2]
-                  }
-                }
-              }
-            }
-          }
-        }
-        for (co: int32, 0, 16) {
-          for (i: int32, 0, 14) {
-            for (j: int32, 0, 14) {
-              for (ci: int32, 0, 16) {
-                res_conv_1: Buffer(res_conv, int32, [50176], [])[((((co*3136) + (i*224)) + (j*16)) + ci)] = 0
-                for (ic: int32, 0, 16) {
-                  for (dy: int32, 0, 3) {
-                    for (dx: int32, 0, 3) {
-                      for (ic_tns: int32, 0, 16) {
-                        let cse_var_4: int32 = (j*16)
-                        let cse_var_3: int32 = ((((co*3136) + (i*224)) + cse_var_4) + ci)
-                        res_conv_1[cse_var_3] = (res_conv_1[cse_var_3] + (cast(int32, data_buf_1[((((((ic*4096) + (i*256)) + (dy*256)) + cse_var_4) + (dx*16)) + ic_tns)])*cast(int32, kernel_buf_1[((((((co*36864) + (ic*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)])))
-                      }
-                    }
-                  }
-                }
-              }
-            }
-          }
-        }
-        for (i1_2: int32, 0, 16) {
-          for (i2_2: int32, 0, 14) {
-            for (i3_2: int32, 0, 14) {
-              for (i5_2: int32, 0, 16) {
-                let cse_var_5: int32 = ((((i1_2*3136) + (i2_2*224)) + (i3_2*16)) + i5_2)
-                res_conv_2: Buffer(res_conv, int32, [50176], [])[cse_var_5] = @tir.shift_right(res_conv_1[cse_var_5], 8, dtype=int32)
-              }
-            }
-          }
-        }
-        for (i1_3: int32, 0, 16) {
-          for (i2_3: int32, 0, 14) {
-            for (i3_3: int32, 0, 14) {
-              for (i5_3: int32, 0, 16) {
-                let cse_var_6: int32 = ((((i1_3*3136) + (i2_3*224)) + (i3_3*16)) + i5_3)
-                res_conv_3: Buffer(res_conv, int32, [50176], [])[cse_var_6] = max(res_conv_2[cse_var_6], 0)
-              }
-            }
-          }
-        }
-        for (i1_4: int32, 0, 16) {
-          for (i2_4: int32, 0, 14) {
-            for (i3_4: int32, 0, 14) {
-              for (i5_4: int32, 0, 16) {
-                let cse_var_7: int32 = ((((i1_4*3136) + (i2_4*224)) + (i3_4*16)) + i5_4)
-                res_conv_4: Buffer(res_conv, int32, [50176], [])[cse_var_7] = min(res_conv_3[cse_var_7], 127)
-              }
-            }
-          }
-        }
-        for (i1_5: int32, 0, 16) {
-          for (i2_5: int32, 0, 14) {
-            for (i3_5: int32, 0, 14) {
-              for (i5_5: int32, 0, 16) {
-                let cse_var_8: int32 = ((((i1_5*3136) + (i2_5*224)) + (i3_5*16)) + i5_5)
-                res_3: Buffer(res_2, int8, [50176], [])[cse_var_8] = cast(int8, res_conv_4[cse_var_8])
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, kernel: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 16, 14, 14, 1, 16), "int8")
+            kernel_1 = T.match_buffer(kernel, (16, 16, 3, 3, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 16, 14, 14, 1, 16), "int8")
+            data_buf = T.allocate([65536], "int8", "global")
+            kernel_buf = T.allocate([589824], "int8", "global")
+            res_conv = T.allocate([50176], "int32", "global")
+            data_buf_1 = T.buffer_decl((65536,), "int8", data=data_buf)
+            for i1, i2, i3, i5 in T.grid(16, 16, 16, 16):
+                cse_var_1: T.int32 = i3 * 16
+                data_2 = T.buffer_decl((50176,), "int8", data=data_1.data)
+                data_buf_1[i1 * 4096 + i2 * 256 + cse_var_1 + i5] = T.if_then_else(1 <= i2 and i2 < 15 and 1 <= i3 and i3 < 15, data_2[i1 * 3136 + i2 * 224 + cse_var_1 + i5 - 240], T.int8(0))
+            kernel_buf_1 = T.buffer_decl((589824,), "int8", data=kernel_buf)
+            for i0, i1, i2, i3, i4, i5 in T.grid(16, 16, 3, 3, 16, 16):
+                cse_var_2: T.int32 = i0 * 36864 + i1 * 2304 + i2 * 768 + i3 * 256 + i4 * 16 + i5
+                kernel_2 = T.buffer_decl((589824,), "int8", data=kernel_1.data)
+                kernel_buf_1[cse_var_2] = kernel_2[cse_var_2]
+            res_conv_1 = T.buffer_decl((50176,), "int32", data=res_conv)
+            for co, i, j, ci in T.grid(16, 14, 14, 16):
+                res_conv_1[co * 3136 + i * 224 + j * 16 + ci] = 0
+                for ic, dy, dx, ic_tns in T.grid(16, 3, 3, 16):
+                    cse_var_4: T.int32 = j * 16
+                    cse_var_3: T.int32 = co * 3136 + i * 224 + cse_var_4 + ci
+                    res_conv_1[cse_var_3] = res_conv_1[cse_var_3] + T.Cast("int32", data_buf_1[ic * 4096 + i * 256 + dy * 256 + cse_var_4 + dx * 16 + ic_tns]) * T.Cast("int32", kernel_buf_1[co * 36864 + ic * 2304 + dy * 768 + dx * 256 + ci * 16 + ic_tns])
+            res_conv_2 = T.buffer_decl((50176,), "int32", data=res_conv)
+            for i1, i2, i3, i5 in T.grid(16, 14, 14, 16):
+                cse_var_5: T.int32 = i1 * 3136 + i2 * 224 + i3 * 16 + i5
+                res_conv_2[cse_var_5] = T.shift_right(res_conv_1[cse_var_5], 8)
+            res_conv_3 = T.buffer_decl((50176,), "int32", data=res_conv)
+            for i1, i2, i3, i5 in T.grid(16, 14, 14, 16):
+                cse_var_6: T.int32 = i1 * 3136 + i2 * 224 + i3 * 16 + i5
+                res_conv_3[cse_var_6] = T.max(res_conv_2[cse_var_6], 0)
+            res_conv_4 = T.buffer_decl((50176,), "int32", data=res_conv)
+            for i1, i2, i3, i5 in T.grid(16, 14, 14, 16):
+                cse_var_7: T.int32 = i1 * 3136 + i2 * 224 + i3 * 16 + i5
+                res_conv_4[cse_var_7] = T.min(res_conv_3[cse_var_7], 127)
+            for i1, i2, i3, i5 in T.grid(16, 14, 14, 16):
+                cse_var_8: T.int32 = i1 * 3136 + i2 * 224 + i3 * 16 + i5
+                res_2 = T.buffer_decl((50176,), "int8", data=res_1.data)
+                res_2[cse_var_8] = T.Cast("int8", res_conv_4[cse_var_8])
 
 
 
@@ -481,141 +428,67 @@ below.
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], []),
-                 kernel: Buffer(kernel_2: Pointer(int8), int8, [16, 16, 3, 3, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], [])}
-      buffer_map = {data_1: data, kernel_1: kernel, res_1: res} {
-      allocate(data_buf: Pointer(global int8), int8, [65536]), storage_scope = global;
-      allocate(kernel_buf: Pointer(global int8), int8, [589824]), storage_scope = global;
-      allocate(res_conv: Pointer(global int32), int32, [25088]), storage_scope = global {
-        for (i1: int32, 0, 16) {
-          for (i2: int32, 0, 16) {
-            for (i3: int32, 0, 16) {
-              for (i5: int32, 0, 16) {
-                let cse_var_1: int32 = (i3*16)
-                data_buf_1: Buffer(data_buf, int8, [65536], [])[((((i1*4096) + (i2*256)) + cse_var_1) + i5)] = @tir.if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), data_3: Buffer(data_2, int8, [50176], [])[(((((i1*3136) + (i2*224)) + cse_var_1) + i5) - 240)], 0i8, dtype=int8)
-              }
-            }
-          }
-        }
-        for (i0: int32, 0, 16) {
-          for (i1_1: int32, 0, 16) {
-            for (i2_1: int32, 0, 3) {
-              for (i3_1: int32, 0, 3) {
-                for (i4: int32, 0, 16) {
-                  for (i5_1: int32, 0, 16) {
-                    let cse_var_2: int32 = ((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)
-                    kernel_buf_1: Buffer(kernel_buf, int8, [589824], [])[cse_var_2] = kernel_3: Buffer(kernel_2, int8, [589824], [])[cse_var_2]
-                  }
-                }
-              }
-            }
-          }
-        }
-        for (i2.outer: int32, 0, 2) {
-          for (co.init: int32, 0, 8) {
-            for (i.init: int32, 0, 7) {
-              for (j.init: int32, 0, 14) {
-                for (ci.init: int32, 0, 16) {
-                  let cse_var_3: int32 = ((((co.init*1568) + (i.init*224)) + (j.init*16)) + ci.init)
-                   {
-                    res_conv_1: Buffer(res_conv, int32, [157351936], [])[cse_var_3] = 0
-                    res_conv_1[(cse_var_3 + 12544)] = 0
-                  }
-                }
-              }
-            }
-          }
-          for (ic.outer: int32, 0, 16) {
-            for (co: int32, 0, 8) {
-              for (i: int32, 0, 7) {
-                for (dy: int32, 0, 3) {
-                  for (dx: int32, 0, 3) {
-                    for (j: int32, 0, 14) {
-                      for (ci: int32, 0, 16) {
-                        for (ic_tns: int32, 0, 16) {
-                          let cse_var_8: int32 = (j*16)
-                          let cse_var_7: int32 = ((((co*1568) + (i*224)) + cse_var_8) + ci)
-                          let cse_var_6: int32 = (cse_var_7 + 12544)
-                          let cse_var_5: int32 = ((((((co*36864) + (ic.outer*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)
-                          let cse_var_4: int32 = (((((((ic.outer*4096) + (i2.outer*1792)) + (i*256)) + (dy*256)) + cse_var_8) + (dx*16)) + ic_tns)
-                           {
-                            res_conv_1[cse_var_7] = (res_conv_1[cse_var_7] + (cast(int32, data_buf_1[cse_var_4])*cast(int32, kernel_buf_1[cse_var_5])))
-                            res_conv_1[cse_var_6] = (res_conv_1[cse_var_6] + (cast(int32, data_buf_1[cse_var_4])*cast(int32, kernel_buf_1[(cse_var_5 + 294912)])))
-                          }
-                        }
-                      }
-                    }
-                  }
-                }
-              }
-            }
-          }
-          for (i1_2: int32, 0, 8) {
-            for (i2_2: int32, 0, 7) {
-              for (i3_2: int32, 0, 14) {
-                for (i5_2: int32, 0, 16) {
-                  let cse_var_10: int32 = ((((i1_2*1568) + (i2_2*224)) + (i3_2*16)) + i5_2)
-                  let cse_var_9: int32 = (cse_var_10 + 12544)
-                   {
-                    res_conv_2: Buffer(res_conv, int32, [157351936], [])[cse_var_10] = @tir.shift_right(res_conv_1[cse_var_10], 8, dtype=int32)
-                    res_conv_2[cse_var_9] = @tir.shift_right(res_conv_1[cse_var_9], 8, dtype=int32)
-                  }
-                }
-              }
-            }
-          }
-          for (i1_3: int32, 0, 8) {
-            for (i2_3: int32, 0, 7) {
-              for (i3_3: int32, 0, 14) {
-                for (i5_3: int32, 0, 16) {
-                  let cse_var_12: int32 = ((((i1_3*1568) + (i2_3*224)) + (i3_3*16)) + i5_3)
-                  let cse_var_11: int32 = (cse_var_12 + 12544)
-                   {
-                    res_conv_3: Buffer(res_conv, int32, [157351936], [])[cse_var_12] = max(res_conv_2[cse_var_12], 0)
-                    res_conv_3[cse_var_11] = max(res_conv_2[cse_var_11], 0)
-                  }
-                }
-              }
-            }
-          }
-          for (i1_4: int32, 0, 8) {
-            for (i2_4: int32, 0, 7) {
-              for (i3_4: int32, 0, 14) {
-                for (i5_4: int32, 0, 16) {
-                  let cse_var_14: int32 = ((((i1_4*1568) + (i2_4*224)) + (i3_4*16)) + i5_4)
-                  let cse_var_13: int32 = (cse_var_14 + 12544)
-                   {
-                    res_conv_4: Buffer(res_conv, int32, [157351936], [])[cse_var_14] = min(res_conv_3[cse_var_14], 127)
-                    res_conv_4[cse_var_13] = min(res_conv_3[cse_var_13], 127)
-                  }
-                }
-              }
-            }
-          }
-          for (i1.inner: int32, 0, 8) {
-            for (i2.inner: int32, 0, 7) {
-              for (i3.inner: int32, 0, 14) {
-                for (i5_5: int32, 0, 16) {
-                  let cse_var_18: int32 = (i2.inner*224)
-                  let cse_var_17: int32 = (i3.inner*16)
-                  let cse_var_16: int32 = ((((i1.inner*1568) + cse_var_18) + cse_var_17) + i5_5)
-                  let cse_var_15: int32 = (((((i1.inner*3136) + (i2.outer*1568)) + cse_var_18) + cse_var_17) + i5_5)
-                   {
-                    res_3: Buffer(res_2, int8, [50176], [])[cse_var_15] = cast(int8, res_conv_4[cse_var_16])
-                    res_3[(cse_var_15 + 25088)] = cast(int8, res_conv_4[(cse_var_16 + 12544)])
-                  }
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, kernel: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 16, 14, 14, 1, 16), "int8")
+            kernel_1 = T.match_buffer(kernel, (16, 16, 3, 3, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 16, 14, 14, 1, 16), "int8")
+            data_buf = T.allocate([65536], "int8", "global")
+            kernel_buf = T.allocate([589824], "int8", "global")
+            res_conv = T.allocate([25088], "int32", "global")
+            data_buf_1 = T.buffer_decl((65536,), "int8", data=data_buf)
+            for i1, i2, i3, i5 in T.grid(16, 16, 16, 16):
+                cse_var_1: T.int32 = i3 * 16
+                data_2 = T.buffer_decl((50176,), "int8", data=data_1.data)
+                data_buf_1[i1 * 4096 + i2 * 256 + cse_var_1 + i5] = T.if_then_else(1 <= i2 and i2 < 15 and 1 <= i3 and i3 < 15, data_2[i1 * 3136 + i2 * 224 + cse_var_1 + i5 - 240], T.int8(0))
+            kernel_buf_1 = T.buffer_decl((589824,), "int8", data=kernel_buf)
+            for i0, i1, i2, i3, i4, i5 in T.grid(16, 16, 3, 3, 16, 16):
+                cse_var_2: T.int32 = i0 * 36864 + i1 * 2304 + i2 * 768 + i3 * 256 + i4 * 16 + i5
+                kernel_2 = T.buffer_decl((589824,), "int8", data=kernel_1.data)
+                kernel_buf_1[cse_var_2] = kernel_2[cse_var_2]
+            for i2_outer in range(2):
+                res_conv_1 = T.buffer_decl((157351936,), "int32", data=res_conv)
+                for co_init, i_init, j_init, ci_init in T.grid(8, 7, 14, 16):
+                    cse_var_3: T.int32 = co_init * 1568 + i_init * 224 + j_init * 16 + ci_init
+                    res_conv_1[cse_var_3] = 0
+                    res_conv_1[cse_var_3 + 12544] = 0
+                for ic_outer, co, i, dy, dx, j, ci, ic_tns in T.grid(16, 8, 7, 3, 3, 14, 16, 16):
+                    cse_var_8: T.int32 = j * 16
+                    cse_var_7: T.int32 = co * 1568 + i * 224 + cse_var_8 + ci
+                    cse_var_6: T.int32 = cse_var_7 + 12544
+                    cse_var_5: T.int32 = co * 36864 + ic_outer * 2304 + dy * 768 + dx * 256 + ci * 16 + ic_tns
+                    cse_var_4: T.int32 = ic_outer * 4096 + i2_outer * 1792 + i * 256 + dy * 256 + cse_var_8 + dx * 16 + ic_tns
+                    res_conv_1[cse_var_7] = res_conv_1[cse_var_7] + T.Cast("int32", data_buf_1[cse_var_4]) * T.Cast("int32", kernel_buf_1[cse_var_5])
+                    res_conv_1[cse_var_6] = res_conv_1[cse_var_6] + T.Cast("int32", data_buf_1[cse_var_4]) * T.Cast("int32", kernel_buf_1[cse_var_5 + 294912])
+                res_conv_2 = T.buffer_decl((157351936,), "int32", data=res_conv)
+                for i1, i2, i3, i5 in T.grid(8, 7, 14, 16):
+                    cse_var_10: T.int32 = i1 * 1568 + i2 * 224 + i3 * 16 + i5
+                    cse_var_9: T.int32 = cse_var_10 + 12544
+                    res_conv_2[cse_var_10] = T.shift_right(res_conv_1[cse_var_10], 8)
+                    res_conv_2[cse_var_9] = T.shift_right(res_conv_1[cse_var_9], 8)
+                res_conv_3 = T.buffer_decl((157351936,), "int32", data=res_conv)
+                for i1, i2, i3, i5 in T.grid(8, 7, 14, 16):
+                    cse_var_12: T.int32 = i1 * 1568 + i2 * 224 + i3 * 16 + i5
+                    cse_var_11: T.int32 = cse_var_12 + 12544
+                    res_conv_3[cse_var_12] = T.max(res_conv_2[cse_var_12], 0)
+                    res_conv_3[cse_var_11] = T.max(res_conv_2[cse_var_11], 0)
+                res_conv_4 = T.buffer_decl((157351936,), "int32", data=res_conv)
+                for i1, i2, i3, i5 in T.grid(8, 7, 14, 16):
+                    cse_var_14: T.int32 = i1 * 1568 + i2 * 224 + i3 * 16 + i5
+                    cse_var_13: T.int32 = cse_var_14 + 12544
+                    res_conv_4[cse_var_14] = T.min(res_conv_3[cse_var_14], 127)
+                    res_conv_4[cse_var_13] = T.min(res_conv_3[cse_var_13], 127)
+                for i1_inner, i2_inner, i3_inner, i5 in T.grid(8, 7, 14, 16):
+                    cse_var_18: T.int32 = i2_inner * 224
+                    cse_var_17: T.int32 = i3_inner * 16
+                    cse_var_16: T.int32 = i1_inner * 1568 + cse_var_18 + cse_var_17 + i5
+                    cse_var_15: T.int32 = i1_inner * 3136 + i2_outer * 1568 + cse_var_18 + cse_var_17 + i5
+                    res_2 = T.buffer_decl((50176,), "int8", data=res_1.data)
+                    res_2[cse_var_15] = T.Cast("int8", res_conv_4[cse_var_16])
+                    res_2[cse_var_15 + 25088] = T.Cast("int8", res_conv_4[cse_var_16 + 12544])
 
 
 
@@ -697,115 +570,85 @@ and mapping the shift, and clipping computation to the vector ALU.
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], []),
-                 kernel: Buffer(kernel_2: Pointer(int8), int8, [16, 16, 3, 3, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 16, 14, 14, 1, 16], [])}
-      buffer_map = {data_1: data, kernel_1: kernel, res_1: res} {
-      @tir.vta.coproc_dep_push(3, 2, dtype=int32)
-      @tir.vta.coproc_dep_push(3, 2, dtype=int32)
-      for (i2.outer: int32, 0, 2) {
-        for (cthread.s: int32, 0, 2) {
-          attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-            @tir.vta.coproc_dep_pop(3, 2, dtype=int32)
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-              @tir.call_extern("VTAUopLoopBegin", 8, 98, 0, 0, dtype=int32)
-              @tir.call_extern("VTAUopLoopBegin", 7, 14, 0, 0, dtype=int32)
-              for (j.init: int32, 0, 14) {
-                @tir.vta.uop_push(0, 1, ((cthread.s*784) + j.init), 0, 0, 0, 0, 0, dtype=int32)
-              }
-              @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-              @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-            }
-            @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-          }
-        }
-        for (ic.outer: int32, 0, 16) {
-          let cse_var_6: int32 = (i2.outer*7)
-          let cse_var_5: int32 = (ic.outer*9)
-          let cse_var_4: int32 = max((1 - cse_var_6), 0)
-          let cse_var_3: int32 = max((cse_var_6 - 6), 0)
-          let cse_var_2: int32 = ((9 - cse_var_4) - cse_var_3)
-          let cse_var_1: int32 = ((((ic.outer*196) + (i2.outer*98)) + (cse_var_4*14)) - 14)
-           {
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
-              @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), data_2, cse_var_1, 14, cse_var_2, 14, 1, cse_var_4, 1, cse_var_3, 0, 2, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), kernel_2, cse_var_5, 9, 8, 144, 0, 0, 0, 0, 0, 1, dtype=int32)
-              @tir.vta.coproc_dep_push(1, 2, dtype=int32)
-            }
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
-              @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), data_2, cse_var_1, 14, cse_var_2, 14, 1, cse_var_4, 1, cse_var_3, 144, 2, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), kernel_2, (cse_var_5 + 1152), 9, 8, 144, 0, 0, 0, 0, 72, 1, dtype=int32)
-              @tir.vta.coproc_dep_push(1, 2, dtype=int32)
-            }
-            for (cthread.s_1: int32, 0, 2) {
-              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-                @tir.vta.coproc_dep_pop(1, 2, dtype=int32)
-                attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-                  @tir.call_extern("VTAUopLoopBegin", 8, 98, 0, 9, dtype=int32)
-                  @tir.call_extern("VTAUopLoopBegin", 7, 14, 16, 0, dtype=int32)
-                  for (dy: int32, 0, 3) {
-                    for (dx: int32, 0, 3) {
-                      for (j: int32, 0, 14) {
-                        @tir.vta.uop_push(0, 0, ((cthread.s_1*784) + j), ((((cthread.s_1*144) + (dy*16)) + j) + dx), (((cthread.s_1*72) + (dy*3)) + dx), 0, 0, 0, dtype=int32)
-                      }
-                    }
-                  }
-                  @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-                  @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-                }
-                @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-              }
-            }
-          }
-        }
-        @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-        @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-        for (cthread.s_2: int32, 0, 2) {
-          let cse_var_7: int32 = (cthread.s_2*784)
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-              @tir.call_extern("VTAUopLoopBegin", 784, 1, 1, 0, dtype=int32)
-              @tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 3, 1, 8, dtype=int32)
-              @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-            }
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-              @tir.call_extern("VTAUopLoopBegin", 784, 1, 1, 0, dtype=int32)
-              @tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 1, 1, 0, dtype=int32)
-              @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-            }
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-              @tir.call_extern("VTAUopLoopBegin", 784, 1, 1, 0, dtype=int32)
-              @tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 0, 1, 127, dtype=int32)
-              @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-            }
-            @tir.vta.coproc_dep_push(2, 3, dtype=int32)
-          }
-        }
-        for (cthread.s_3: int32, 0, 2) {
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
-            @tir.vta.coproc_dep_pop(2, 3, dtype=int32)
-            for (i1.inner: int32, 0, 8) {
-              for (i2.inner: int32, 0, 7) {
-                for (i3.inner: int32, 0, 14) {
-                  let cse_var_8: int32 = (i2.inner*14)
-                  @tir.call_extern("VTAStoreBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), ((((cthread.s_3*784) + (i1.inner*98)) + cse_var_8) + i3.inner), 4, res_2, (((((cthread.s_3*1568) + (i1.inner*196)) + (i2.outer*98)) + cse_var_8) + i3.inner), 1, 1, 1, dtype=int32)
-                }
-              }
-            }
-            @tir.vta.coproc_dep_push(3, 2, dtype=int32)
-          }
-        }
-      }
-      @tir.vta.coproc_dep_pop(3, 2, dtype=int32)
-      @tir.vta.coproc_dep_pop(3, 2, dtype=int32)
-      @tir.vta.coproc_sync(, dtype=int32)
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, kernel: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 16, 14, 14, 1, 16), "int8")
+            kernel_1 = T.match_buffer(kernel, (16, 16, 3, 3, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 16, 14, 14, 1, 16), "int8")
+            T.tir.vta.coproc_dep_push(3, 2)
+            T.tir.vta.coproc_dep_push(3, 2)
+            for i2_outer in range(2):
+                vta = T.var("int32")
+                for cthread_s in range(2):
+                    T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2)
+                    T.tir.vta.coproc_dep_pop(3, 2)
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 8, 98, 0, 0)
+                        T.call_extern("int32", "VTAUopLoopBegin", 7, 14, 0, 0)
+                        for j_init in range(14):
+                            T.tir.vta.uop_push(0, 1, cthread_s * 784 + j_init, 0, 0, 0, 0, 0)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    T.tir.vta.coproc_dep_push(2, 1)
+                for ic_outer in range(16):
+                    cse_var_6: T.int32 = i2_outer * 7
+                    cse_var_5: T.int32 = ic_outer * 9
+                    cse_var_4: T.int32 = T.max(1 - cse_var_6, 0)
+                    cse_var_3: T.int32 = T.max(cse_var_6 - 6, 0)
+                    cse_var_2: T.int32 = 9 - cse_var_4 - cse_var_3
+                    cse_var_1: T.int32 = ic_outer * 196 + i2_outer * 98 + cse_var_4 * 14 - 14
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 1):
+                        T.tir.vta.coproc_dep_pop(2, 1)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), data_1.data, cse_var_1, 14, cse_var_2, 14, 1, cse_var_4, 1, cse_var_3, 0, 2)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), kernel_1.data, cse_var_5, 9, 8, 144, 0, 0, 0, 0, 0, 1)
+                        T.tir.vta.coproc_dep_push(1, 2)
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 1):
+                        T.tir.vta.coproc_dep_pop(2, 1)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), data_1.data, cse_var_1, 14, cse_var_2, 14, 1, cse_var_4, 1, cse_var_3, 144, 2)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), kernel_1.data, cse_var_5 + 1152, 9, 8, 144, 0, 0, 0, 0, 72, 1)
+                        T.tir.vta.coproc_dep_push(1, 2)
+                    for cthread_s in range(2):
+                        T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2)
+                        T.tir.vta.coproc_dep_pop(1, 2)
+                        with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                            T.call_extern("int32", "VTAUopLoopBegin", 8, 98, 0, 9)
+                            T.call_extern("int32", "VTAUopLoopBegin", 7, 14, 16, 0)
+                            for dy, dx, j in T.grid(3, 3, 14):
+                                T.tir.vta.uop_push(0, 0, cthread_s * 784 + j, cthread_s * 144 + dy * 16 + j + dx, cthread_s * 72 + dy * 3 + dx, 0, 0, 0)
+                            T.call_extern("int32", "VTAUopLoopEnd")
+                            T.call_extern("int32", "VTAUopLoopEnd")
+                        T.tir.vta.coproc_dep_push(2, 1)
+                T.tir.vta.coproc_dep_pop(2, 1)
+                T.tir.vta.coproc_dep_pop(2, 1)
+                for cthread_s in range(2):
+                    cse_var_7: T.int32 = cthread_s * 784
+                    T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2)
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 784, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 3, 1, 8)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 784, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 1, 1, 0)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 784, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, cse_var_7, cse_var_7, 0, 0, 1, 127)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    T.tir.vta.coproc_dep_push(2, 3)
+                for cthread_s in range(2):
+                    T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 3)
+                    T.tir.vta.coproc_dep_pop(2, 3)
+                    for i1_inner, i2_inner, i3_inner in T.grid(8, 7, 14):
+                        cse_var_8: T.int32 = i2_inner * 14
+                        T.call_extern("int32", "VTAStoreBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), cthread_s * 784 + i1_inner * 98 + cse_var_8 + i3_inner, 4, res_1.data, cthread_s * 1568 + i1_inner * 196 + i2_outer * 98 + cse_var_8 + i3_inner, 1, 1, 1)
+                    T.tir.vta.coproc_dep_push(3, 2)
+            T.tir.vta.coproc_dep_pop(3, 2)
+            T.tir.vta.coproc_dep_pop(3, 2)
+            T.tir.vta.coproc_sync()
 
 
 
diff --git a/docs/_sources/topic/vta/tutorials/optimize/matrix_multiply_opt.rst.txt b/docs/_sources/topic/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
index 29770c6c38..d7eec75c7e 100644
--- a/docs/_sources/topic/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
@@ -211,70 +211,49 @@ Those include:
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 64, 1, 16], []),
-                 weight: Buffer(weight_2: Pointer(int8), int8, [64, 64, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 64, 1, 16], [])}
-      buffer_map = {data_1: data, weight_1: weight, res_1: res} {
-      allocate(data_buf: Pointer(global int8), int8, [1024]), storage_scope = global;
-      allocate(weight_buf: Pointer(global int8), int8, [1048576]), storage_scope = global;
-      allocate(res_gem: Pointer(global int32), int32, [1024]), storage_scope = global {
-        for (i1: int32, 0, 64) {
-          for (i3: int32, 0, 16) {
-            let cse_var_1: int32 = ((i1*16) + i3)
-            data_buf_1: Buffer(data_buf, int8, [1024], [])[cse_var_1] = data_3: Buffer(data_2, int8, [1024], [])[cse_var_1]
-          }
-        }
-        for (i0: int32, 0, 64) {
-          for (i1_1: int32, 0, 64) {
-            for (i2: int32, 0, 16) {
-              for (i3_1: int32, 0, 16) {
-                let cse_var_2: int32 = ((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)
-                weight_buf_1: Buffer(weight_buf, int8, [1048576], [])[cse_var_2] = weight_3: Buffer(weight_2, int8, [1048576], [])[cse_var_2]
-              }
-            }
-          }
-        }
-        for (co: int32, 0, 64) {
-          for (ci: int32, 0, 16) {
-            res_gem_1: Buffer(res_gem, int32, [1024], [])[((co*16) + ci)] = 0
-            for (ic: int32, 0, 64) {
-              for (ic_tns: int32, 0, 16) {
-                let cse_var_3: int32 = ((co*16) + ci)
-                res_gem_1[cse_var_3] = (res_gem_1[cse_var_3] + (cast(int32, data_buf_1[((ic*16) + ic_tns)])*cast(int32, weight_buf_1[((((co*16384) + (ic*256)) + (ci*16)) + ic_tns)])))
-              }
-            }
-          }
-        }
-        for (i1_2: int32, 0, 64) {
-          for (i3_2: int32, 0, 16) {
-            let cse_var_4: int32 = ((i1_2*16) + i3_2)
-            res_gem_2: Buffer(res_gem, int32, [1024], [])[cse_var_4] = @tir.shift_right(res_gem_1[cse_var_4], 8, dtype=int32)
-          }
-        }
-        for (i1_3: int32, 0, 64) {
-          for (i3_3: int32, 0, 16) {
-            let cse_var_5: int32 = ((i1_3*16) + i3_3)
-            res_gem_3: Buffer(res_gem, int32, [1024], [])[cse_var_5] = max(res_gem_2[cse_var_5], 0)
-          }
-        }
-        for (i1_4: int32, 0, 64) {
-          for (i3_4: int32, 0, 16) {
-            let cse_var_6: int32 = ((i1_4*16) + i3_4)
-            res_gem_4: Buffer(res_gem, int32, [1024], [])[cse_var_6] = min(res_gem_3[cse_var_6], 127)
-          }
-        }
-        for (i1_5: int32, 0, 64) {
-          for (i3_5: int32, 0, 16) {
-            let cse_var_7: int32 = ((i1_5*16) + i3_5)
-            res_3: Buffer(res_2, int8, [1024], [])[cse_var_7] = cast(int8, res_gem_4[cse_var_7])
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, weight: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 64, 1, 16), "int8")
+            weight_1 = T.match_buffer(weight, (64, 64, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 64, 1, 16), "int8")
+            data_buf = T.allocate([1024], "int8", "global")
+            weight_buf = T.allocate([1048576], "int8", "global")
+            res_gem = T.allocate([1024], "int32", "global")
+            data_buf_1 = T.buffer_decl((1024,), "int8", data=data_buf)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_1: T.int32 = i1 * 16 + i3
+                data_2 = T.buffer_decl((1024,), "int8", data=data_1.data)
+                data_buf_1[cse_var_1] = data_2[cse_var_1]
+            weight_buf_1 = T.buffer_decl((1048576,), "int8", data=weight_buf)
+            for i0, i1, i2, i3 in T.grid(64, 64, 16, 16):
+                cse_var_2: T.int32 = i0 * 16384 + i1 * 256 + i2 * 16 + i3
+                weight_2 = T.buffer_decl((1048576,), "int8", data=weight_1.data)
+                weight_buf_1[cse_var_2] = weight_2[cse_var_2]
+            res_gem_1 = T.buffer_decl((1024,), "int32", data=res_gem)
+            for co, ci in T.grid(64, 16):
+                res_gem_1[co * 16 + ci] = 0
+                for ic, ic_tns in T.grid(64, 16):
+                    cse_var_3: T.int32 = co * 16 + ci
+                    res_gem_1[cse_var_3] = res_gem_1[cse_var_3] + T.Cast("int32", data_buf_1[ic * 16 + ic_tns]) * T.Cast("int32", weight_buf_1[co * 16384 + ic * 256 + ci * 16 + ic_tns])
+            res_gem_2 = T.buffer_decl((1024,), "int32", data=res_gem)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_4: T.int32 = i1 * 16 + i3
+                res_gem_2[cse_var_4] = T.shift_right(res_gem_1[cse_var_4], 8)
+            res_gem_3 = T.buffer_decl((1024,), "int32", data=res_gem)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_5: T.int32 = i1 * 16 + i3
+                res_gem_3[cse_var_5] = T.max(res_gem_2[cse_var_5], 0)
+            res_gem_4 = T.buffer_decl((1024,), "int32", data=res_gem)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_6: T.int32 = i1 * 16 + i3
+                res_gem_4[cse_var_6] = T.min(res_gem_3[cse_var_6], 127)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_7: T.int32 = i1 * 16 + i3
+                res_2 = T.buffer_decl((1024,), "int8", data=res_1.data)
+                res_2[cse_var_7] = T.Cast("int8", res_gem_4[cse_var_7])
 
 
 
@@ -379,78 +358,50 @@ below:
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 64, 1, 16], []),
-                 weight: Buffer(weight_2: Pointer(int8), int8, [64, 64, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 64, 1, 16], [])}
-      buffer_map = {data_1: data, weight_1: weight, res_1: res} {
-      allocate(data_buf: Pointer(global int8), int8, [1024]), storage_scope = global;
-      allocate(weight_buf: Pointer(global int8), int8, [1048576]), storage_scope = global;
-      allocate(res_gem: Pointer(global int32), int32, [256]), storage_scope = global {
-        for (i1: int32, 0, 64) {
-          for (i3: int32, 0, 16) {
-            let cse_var_1: int32 = ((i1*16) + i3)
-            data_buf_1: Buffer(data_buf, int8, [1024], [])[cse_var_1] = data_3: Buffer(data_2, int8, [1024], [])[cse_var_1]
-          }
-        }
-        for (i0: int32, 0, 64) {
-          for (i1_1: int32, 0, 64) {
-            for (i2: int32, 0, 16) {
-              for (i3_1: int32, 0, 16) {
-                let cse_var_2: int32 = ((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)
-                weight_buf_1: Buffer(weight_buf, int8, [1048576], [])[cse_var_2] = weight_3: Buffer(weight_2, int8, [1048576], [])[cse_var_2]
-              }
-            }
-          }
-        }
-        for (i1.outer: int32, 0, 4) {
-          for (co.init: int32, 0, 16) {
-            for (ci.init: int32, 0, 16) {
-              res_gem_1: Buffer(res_gem, int32, [256], [])[((co.init*16) + ci.init)] = 0
-            }
-          }
-          for (ic.outer: int32, 0, 4) {
-            for (co: int32, 0, 16) {
-              for (ic.inner: int32, 0, 16) {
-                for (ci: int32, 0, 16) {
-                  for (ic_tns: int32, 0, 16) {
-                    let cse_var_3: int32 = ((co*16) + ci)
-                    res_gem_1[cse_var_3] = (res_gem_1[cse_var_3] + (cast(int32, data_buf_1[(((ic.outer*256) + (ic.inner*16)) + ic_tns)])*cast(int32, weight_buf_1[((((((i1.outer*262144) + (co*16384)) + (ic.outer*4096)) + (ic.inner*256)) + (ci*16)) + ic_tns)])))
-                  }
-                }
-              }
-            }
-          }
-          for (i1_2: int32, 0, 16) {
-            for (i3_2: int32, 0, 16) {
-              let cse_var_4: int32 = ((i1_2*16) + i3_2)
-              res_gem_2: Buffer(res_gem, int32, [256], [])[cse_var_4] = @tir.shift_right(res_gem_1[cse_var_4], 8, dtype=int32)
-            }
-          }
-          for (i1_3: int32, 0, 16) {
-            for (i3_3: int32, 0, 16) {
-              let cse_var_5: int32 = ((i1_3*16) + i3_3)
-              res_gem_3: Buffer(res_gem, int32, [256], [])[cse_var_5] = max(res_gem_2[cse_var_5], 0)
-            }
-          }
-          for (i1_4: int32, 0, 16) {
-            for (i3_4: int32, 0, 16) {
-              let cse_var_6: int32 = ((i1_4*16) + i3_4)
-              res_gem_4: Buffer(res_gem, int32, [256], [])[cse_var_6] = min(res_gem_3[cse_var_6], 127)
-            }
-          }
-          for (i1.inner: int32, 0, 16) {
-            for (i3_5: int32, 0, 16) {
-              let cse_var_7: int32 = (i1.inner*16)
-              res_3: Buffer(res_2, int8, [1024], [])[(((i1.outer*256) + cse_var_7) + i3_5)] = cast(int8, res_gem_4[(cse_var_7 + i3_5)])
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, weight: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 64, 1, 16), "int8")
+            weight_1 = T.match_buffer(weight, (64, 64, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 64, 1, 16), "int8")
+            data_buf = T.allocate([1024], "int8", "global")
+            weight_buf = T.allocate([1048576], "int8", "global")
+            res_gem = T.allocate([256], "int32", "global")
+            data_buf_1 = T.buffer_decl((1024,), "int8", data=data_buf)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_1: T.int32 = i1 * 16 + i3
+                data_2 = T.buffer_decl((1024,), "int8", data=data_1.data)
+                data_buf_1[cse_var_1] = data_2[cse_var_1]
+            weight_buf_1 = T.buffer_decl((1048576,), "int8", data=weight_buf)
+            for i0, i1, i2, i3 in T.grid(64, 64, 16, 16):
+                cse_var_2: T.int32 = i0 * 16384 + i1 * 256 + i2 * 16 + i3
+                weight_2 = T.buffer_decl((1048576,), "int8", data=weight_1.data)
+                weight_buf_1[cse_var_2] = weight_2[cse_var_2]
+            for i1_outer in range(4):
+                res_gem_1 = T.buffer_decl((256,), "int32", data=res_gem)
+                for co_init, ci_init in T.grid(16, 16):
+                    res_gem_1[co_init * 16 + ci_init] = 0
+                for ic_outer, co, ic_inner, ci, ic_tns in T.grid(4, 16, 16, 16, 16):
+                    cse_var_3: T.int32 = co * 16 + ci
+                    res_gem_1[cse_var_3] = res_gem_1[cse_var_3] + T.Cast("int32", data_buf_1[ic_outer * 256 + ic_inner * 16 + ic_tns]) * T.Cast("int32", weight_buf_1[i1_outer * 262144 + co * 16384 + ic_outer * 4096 + ic_inner * 256 + ci * 16 + ic_tns])
+                res_gem_2 = T.buffer_decl((256,), "int32", data=res_gem)
+                for i1, i3 in T.grid(16, 16):
+                    cse_var_4: T.int32 = i1 * 16 + i3
+                    res_gem_2[cse_var_4] = T.shift_right(res_gem_1[cse_var_4], 8)
+                res_gem_3 = T.buffer_decl((256,), "int32", data=res_gem)
+                for i1, i3 in T.grid(16, 16):
+                    cse_var_5: T.int32 = i1 * 16 + i3
+                    res_gem_3[cse_var_5] = T.max(res_gem_2[cse_var_5], 0)
+                res_gem_4 = T.buffer_decl((256,), "int32", data=res_gem)
+                for i1, i3 in T.grid(16, 16):
+                    cse_var_6: T.int32 = i1 * 16 + i3
+                    res_gem_4[cse_var_6] = T.min(res_gem_3[cse_var_6], 127)
+                for i1_inner, i3 in T.grid(16, 16):
+                    cse_var_7: T.int32 = i1_inner * 16
+                    res_2 = T.buffer_decl((1024,), "int8", data=res_1.data)
+                    res_2[i1_outer * 256 + cse_var_7 + i3] = T.Cast("int8", res_gem_4[cse_var_7 + i3])
 
 
 
@@ -532,77 +483,62 @@ and mapping the shift, and clipping computation to the vector ALU.
 
  .. code-block:: none
 
-    @main = primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {data: Buffer(data_2: Pointer(int8), int8, [1, 64, 1, 16], []),
-                 weight: Buffer(weight_2: Pointer(int8), int8, [64, 64, 16, 16], []),
-                 res: Buffer(res_2: Pointer(int8), int8, [1, 64, 1, 16], [])}
-      buffer_map = {data_1: data, weight_1: weight, res_1: res} {
-      @tir.vta.coproc_dep_push(3, 2, dtype=int32)
-      for (i1.outer: int32, 0, 4) {
-        attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-          @tir.vta.coproc_dep_pop(3, 2, dtype=int32)
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-            @tir.call_extern("VTAUopLoopBegin", 16, 1, 0, 0, dtype=int32)
-            @tir.vta.uop_push(0, 1, 0, 0, 0, 0, 0, 0, dtype=int32)
-            @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-          }
-          @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-        }
-        for (ic.outer: int32, 0, 4) {
-          let cse_var_1: int32 = (ic.outer*16)
-           {
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
-              @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), data_2, cse_var_1, 16, 1, 16, 0, 0, 0, 0, 0, 2, dtype=int32)
-              @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), weight_2, ((i1.outer*1024) + cse_var_1), 16, 16, 64, 0, 0, 0, 0, 0, 1, dtype=int32)
-              @tir.vta.coproc_dep_push(1, 2, dtype=int32)
-            }
-            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-              @tir.vta.coproc_dep_pop(1, 2, dtype=int32)
-              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
-                @tir.call_extern("VTAUopLoopBegin", 16, 1, 0, 16, dtype=int32)
-                @tir.call_extern("VTAUopLoopBegin", 16, 0, 1, 1, dtype=int32)
-                @tir.vta.uop_push(0, 0, 0, 0, 0, 0, 0, 0, dtype=int32)
-                @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-                @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-              }
-              @tir.vta.coproc_dep_push(2, 1, dtype=int32)
-            }
-          }
-        }
-        @tir.vta.coproc_dep_pop(2, 1, dtype=int32)
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-            @tir.call_extern("VTAUopLoopBegin", 16, 1, 1, 0, dtype=int32)
-            @tir.vta.uop_push(1, 0, 0, 0, 0, 3, 1, 8, dtype=int32)
-            @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-          }
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-            @tir.call_extern("VTAUopLoopBegin", 16, 1, 1, 0, dtype=int32)
-            @tir.vta.uop_push(1, 0, 0, 0, 0, 1, 1, 0, dtype=int32)
-            @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-          }
-          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-            @tir.call_extern("VTAUopLoopBegin", 16, 1, 1, 0, dtype=int32)
-            @tir.vta.uop_push(1, 0, 0, 0, 0, 0, 1, 127, dtype=int32)
-            @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-          }
-          @tir.vta.coproc_dep_push(2, 3, dtype=int32)
-        }
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
-          @tir.vta.coproc_dep_pop(2, 3, dtype=int32)
-          for (i1.inner: int32, 0, 16) {
-            @tir.call_extern("VTAStoreBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), i1.inner, 4, res_2, ((i1.outer*16) + i1.inner), 1, 1, 1, dtype=int32)
-          }
-          @tir.vta.coproc_dep_push(3, 2, dtype=int32)
-        }
-      }
-      @tir.vta.coproc_sync(, dtype=int32)
-      @tir.vta.coproc_dep_pop(3, 2, dtype=int32)
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(data: T.handle, weight: T.handle, res: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            data_1 = T.match_buffer(data, (1, 64, 1, 16), "int8")
+            weight_1 = T.match_buffer(weight, (64, 64, 16, 16), "int8")
+            res_1 = T.match_buffer(res, (1, 64, 1, 16), "int8")
+            T.tir.vta.coproc_dep_push(3, 2)
+            for i1_outer in range(4):
+                vta = T.var("int32")
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2):
+                    T.tir.vta.coproc_dep_pop(3, 2)
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 0, 0)
+                        T.tir.vta.uop_push(0, 1, 0, 0, 0, 0, 0, 0)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    T.tir.vta.coproc_dep_push(2, 1)
+                for ic_outer in range(4):
+                    cse_var_1: T.int32 = ic_outer * 16
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 1):
+                        T.tir.vta.coproc_dep_pop(2, 1)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), data_1.data, cse_var_1, 16, 1, 16, 0, 0, 0, 0, 0, 2)
+                        T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), weight_1.data, i1_outer * 1024 + cse_var_1, 16, 16, 64, 0, 0, 0, 0, 0, 1)
+                        T.tir.vta.coproc_dep_push(1, 2)
+                    T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2)
+                    T.tir.vta.coproc_dep_pop(1, 2)
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushGEMMOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 0, 16)
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 0, 1, 1)
+                        T.tir.vta.uop_push(0, 0, 0, 0, 0, 0, 0, 0)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    T.tir.vta.coproc_dep_push(2, 1)
+                T.tir.vta.coproc_dep_pop(2, 1)
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2):
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, 0, 0, 0, 3, 1, 8)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, 0, 0, 0, 1, 1, 0)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                        T.call_extern("int32", "VTAUopLoopBegin", 16, 1, 1, 0)
+                        T.tir.vta.uop_push(1, 0, 0, 0, 0, 0, 1, 127)
+                        T.call_extern("int32", "VTAUopLoopEnd")
+                    T.tir.vta.coproc_dep_push(2, 3)
+                T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 3)
+                T.tir.vta.coproc_dep_pop(2, 3)
+                for i1_inner in range(16):
+                    T.call_extern("int32", "VTAStoreBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), i1_inner, 4, res_1.data, i1_outer * 16 + i1_inner, 1, 1, 1)
+                T.tir.vta.coproc_dep_push(3, 2)
+            T.tir.vta.coproc_sync()
+            T.tir.vta.coproc_dep_pop(3, 2)
 
 
 
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 25f0fd1cb2..f17ef743dd 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
 =================
-**00:03.285** total execution time for **topic_vta_tutorials_optimize** files:
+**00:03.164** total execution time for **topic_vta_tutorials_optimize** files:
 
 +--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``)         | 00:02.814 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``)         | 00:02.703 | 0.0 MB |
 +--------------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``) | 00:00.471 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``) | 00:00.461 | 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 2fe833c2ea..af48bc737d 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
 =================
-**00:00.838** total execution time for **topic_vta_tutorials** files:
+**00:00.816** total execution time for **topic_vta_tutorials** files:
 
 +---------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``) | 00:00.446 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``) | 00:00.435 | 0.0 MB |
 +---------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_topic_vta_tutorials_vta_get_started.py` (``vta_get_started.py``) | 00:00.392 | 0.0 MB |
+| :ref:`sphx_glr_topic_vta_tutorials_vta_get_started.py` (``vta_get_started.py``) | 00:00.381 | 0.0 MB |
 +---------------------------------------------------------------------------------+-----------+--------+
diff --git a/docs/_sources/topic/vta/tutorials/vta_get_started.rst.txt b/docs/_sources/topic/vta/tutorials/vta_get_started.rst.txt
index fb2e99b084..ae51f76072 100644
--- a/docs/_sources/topic/vta/tutorials/vta_get_started.rst.txt
+++ b/docs/_sources/topic/vta/tutorials/vta_get_started.rst.txt
@@ -352,42 +352,34 @@ After we construct the schedule, by default the schedule computes
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(int32), int32, [1, 64, 1, 16], []),
-                 B: Buffer(B_2: Pointer(int32), int32, [1, 64, 1, 16], []),
-                 C: Buffer(C_2: Pointer(int8), int8, [1, 64, 1, 16], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(A_buf: Pointer(global int32), int32, [1024]), storage_scope = global;
-      allocate(B_buf: Pointer(global int32), int32, [1024]), storage_scope = global {
-        for (i1: int32, 0, 64) {
-          for (i3: int32, 0, 16) {
-            let cse_var_1: int32 = ((i1*16) + i3)
-            A_buf_1: Buffer(A_buf, int32, [1024], [])[cse_var_1] = A_3: Buffer(A_2, int32, [1024], [])[cse_var_1]
-          }
-        }
-        for (i1_1: int32, 0, 64) {
-          for (i3_1: int32, 0, 16) {
-            let cse_var_2: int32 = ((i1_1*16) + i3_1)
-            B_buf_1: Buffer(B_buf, int32, [1024], [])[cse_var_2] = B_3: Buffer(B_2, int32, [1024], [])[cse_var_2]
-          }
-        }
-        for (i1_2: int32, 0, 64) {
-          for (i3_2: int32, 0, 16) {
-            let cse_var_3: int32 = ((i1_2*16) + i3_2)
-            A_buf_2: Buffer(A_buf, int32, [1024], [])[cse_var_3] = (A_buf_1[cse_var_3] + B_buf_1[cse_var_3])
-          }
-        }
-        for (i1_3: int32, 0, 64) {
-          for (i3_3: int32, 0, 16) {
-            let cse_var_4: int32 = ((i1_3*16) + i3_3)
-            C_3: Buffer(C_2, int8, [1024], [])[cse_var_4] = cast(int8, A_buf_2[cse_var_4])
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1, 64, 1, 16), "int32")
+            B_1 = T.match_buffer(B, (1, 64, 1, 16), "int32")
+            C_1 = T.match_buffer(C, (1, 64, 1, 16), "int8")
+            A_buf = T.allocate([1024], "int32", "global")
+            B_buf = T.allocate([1024], "int32", "global")
+            A_buf_1 = T.buffer_decl((1024,), "int32", data=A_buf)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_1: T.int32 = i1 * 16 + i3
+                A_2 = T.buffer_decl((1024,), "int32", data=A_1.data)
+                A_buf_1[cse_var_1] = A_2[cse_var_1]
+            B_buf_1 = T.buffer_decl((1024,), "int32", data=B_buf)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_2: T.int32 = i1 * 16 + i3
+                B_2 = T.buffer_decl((1024,), "int32", data=B_1.data)
+                B_buf_1[cse_var_2] = B_2[cse_var_2]
+            A_buf_2 = T.buffer_decl((1024,), "int32", data=A_buf)
+            for i1, i3 in T.grid(64, 16):
+                cse_var_3: T.int32 = i1 * 16 + i3
+                A_buf_2[cse_var_3] = A_buf_1[cse_var_3] + B_buf_1[cse_var_3]
+            for i1, i3 in T.grid(64, 16):
+                cse_var_4: T.int32 = i1 * 16 + i3
+                C_2 = T.buffer_decl((1024,), "int8", data=C_1.data)
+                C_2[cse_var_4] = T.Cast("int8", A_buf_2[cse_var_4])
 
 
 
@@ -495,30 +487,27 @@ with an :code:`env.alu` pragma.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(int32), int32, [1, 64, 1, 16], []),
-                 B: Buffer(B_2: Pointer(int32), int32, [1, 64, 1, 16], []),
-                 C: Buffer(C_2: Pointer(int8), int8, [1, 64, 1, 16], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
-        @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), A_2, 0, 64, 1, 64, 0, 0, 0, 0, 0, 3, dtype=int32)
-        @tir.call_extern("VTALoadBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), B_2, 0, 64, 1, 64, 0, 0, 0, 0, 64, 3, dtype=int32)
-        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
-          @tir.call_extern("VTAUopLoopBegin", 64, 1, 1, 0, dtype=int32)
-          @tir.vta.uop_push(1, 0, 0, 64, 0, 2, 0, 0, dtype=int32)
-          @tir.call_extern("VTAUopLoopEnd", dtype=int32)
-        }
-        @tir.vta.coproc_dep_push(2, 3, dtype=int32)
-      }
-      attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
-        @tir.vta.coproc_dep_pop(2, 3, dtype=int32)
-        @tir.call_extern("VTAStoreBuffer2D", @tir.tvm_thread_context(@tir.vta.command_handle(, dtype=handle), dtype=handle), 0, 4, C_2, 0, 64, 1, 64, dtype=int32)
-      }
-      @tir.vta.coproc_sync(, dtype=int32)
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1, 64, 1, 16), "int32")
+            B_1 = T.match_buffer(B, (1, 64, 1, 16), "int32")
+            C_1 = T.match_buffer(C, (1, 64, 1, 16), "int8")
+            vta = T.var("int32")
+            with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 2):
+                T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), A_1.data, 0, 64, 1, 64, 0, 0, 0, 0, 0, 3)
+                T.call_extern("int32", "VTALoadBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), B_1.data, 0, 64, 1, 64, 0, 0, 0, 0, 64, 3)
+                with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_uop_scope", "VTAPushALUOp"):
+                    T.call_extern("int32", "VTAUopLoopBegin", 64, 1, 1, 0)
+                    T.tir.vta.uop_push(1, 0, 0, 64, 0, 2, 0, 0)
+                    T.call_extern("int32", "VTAUopLoopEnd")
+                T.tir.vta.coproc_dep_push(2, 3)
+            with T.attr(T.iter_var(vta, None, "ThreadIndex", "vta"), "coproc_scope", 3):
+                T.tir.vta.coproc_dep_pop(2, 3)
+                T.call_extern("int32", "VTAStoreBuffer2D", T.tvm_thread_context(T.tir.vta.command_handle()), 0, 4, C_1.data, 0, 64, 1, 64)
+            T.tir.vta.coproc_sync()
 
 
 
diff --git a/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt b/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
index 3fe53b60a9..f362515456 100644
--- a/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
+++ b/docs/_sources/tutorial/auto_scheduler_matmul_x86.rst.txt
@@ -207,13 +207,6 @@ trials, we can load the best schedule from the log file and apply it.
 
 
 
-.. rst-class:: sphx-glr-script-out
-
- .. code-block:: none
-
-    *E
-
-
 
 
 
@@ -243,52 +236,42 @@ operator fusion.
  .. code-block:: none
 
     Lowered TIR:
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle, out_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], []),
-                 out: Buffer(out_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C, out_1: out} {
-      allocate(auto_scheduler_layout_transform: Pointer(global float32), float32, [1048576]), storage_scope = global {
-        for (ax0.ax1.fused.ax2.fused: int32, 0, 128) "parallel" {
-          for (ax4: int32, 0, 256) {
-            for (ax6: int32, 0, 4) {
-              for (ax7: int32, 0, 8) {
-                auto_scheduler_layout_transform_1: Buffer(auto_scheduler_layout_transform, float32, [1048576], [])[((((ax0.ax1.fused.ax2.fused*8192) + (ax4*32)) + (ax6*8)) + ax7)] = B_3: Buffer(B_2, float32, [1048576], [])[((((ax4*4096) + (ax6*1024)) + (ax0.ax1.fused.ax2.fused*8)) + ax7)]
-              }
-            }
-          }
-        }
-        for (i.outer.outer.j.outer.outer.fused: int32, 0, 16384) "parallel" {
-          allocate(matmul: Pointer(global float32x8), float32x8, [4]), storage_scope = global;
-          for (i.outer.inner: int32, 0, 2) {
-            matmul_1: Buffer(matmul, float32x8, [4], [])[0] = broadcast(0f32, 8)
-            matmul_1[1] = broadcast(0f32, 8)
-            matmul_1[2] = broadcast(0f32, 8)
-            matmul_1[3] = broadcast(0f32, 8)
-            for (k.outer: int32, 0, 256) {
-              for (k.inner: int32, 0, 4) {
-                let cse_var_2: int32 = (((floormod(i.outer.outer.j.outer.outer.fused, 128)*8192) + (k.outer*32)) + (k.inner*8))
-                let cse_var_1: int32 = ((((floordiv(i.outer.outer.j.outer.outer.fused, 128)*8192) + (i.outer.inner*4096)) + (k.outer*4)) + k.inner)
-                 {
-                  matmul_1[0] = (matmul_1[0] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[cse_var_1], 8)*auto_scheduler_layout_transform_1[ramp(cse_var_2, 1, 8)]))
-                  matmul_1[1] = (matmul_1[1] + (broadcast(A_3[(cse_var_1 + 1024)], 8)*auto_scheduler_layout_transform_1[ramp(cse_var_2, 1, 8)]))
-                  matmul_1[2] = (matmul_1[2] + (broadcast(A_3[(cse_var_1 + 2048)], 8)*auto_scheduler_layout_transform_1[ramp(cse_var_2, 1, 8)]))
-                  matmul_1[3] = (matmul_1[3] + (broadcast(A_3[(cse_var_1 + 3072)], 8)*auto_scheduler_layout_transform_1[ramp(cse_var_2, 1, 8)]))
-                }
-              }
-            }
-            for (i.inner: int32, 0, 4) {
-              let cse_var_3: int32 = ((((floordiv(i.outer.outer.j.outer.outer.fused, 128)*8192) + (i.outer.inner*4096)) + (i.inner*1024)) + (floormod(i.outer.outer.j.outer.outer.fused, 128)*8))
-              out_3: Buffer(out_2, float32, [1048576], [])[ramp(cse_var_3, 1, 8)] = (matmul_1[i.inner] + C_3: Buffer(C_2, float32, [1048576], [])[ramp(cse_var_3, 1, 8)])
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle, out: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            out_1 = T.match_buffer(out, (1024, 1024))
+            auto_scheduler_layout_transform = T.allocate([1048576], "float32", "global")
+            auto_scheduler_layout_transform_1 = T.buffer_decl((1048576,), data=auto_scheduler_layout_transform)
+            for ax0_ax1_fused_ax2_fused in T.parallel(128):
+                for ax4, ax6, ax7 in T.grid(256, 4, 8):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    auto_scheduler_layout_transform_1[ax0_ax1_fused_ax2_fused * 8192 + ax4 * 32 + ax6 * 8 + ax7] = B_2[ax4 * 4096 + ax6 * 1024 + ax0_ax1_fused_ax2_fused * 8 + ax7]
+            for i_outer_outer_j_outer_outer_fused in T.parallel(16384):
+                matmul = T.allocate([4], "float32x8", "global")
+                for i_outer_inner in range(2):
+                    matmul_1 = T.buffer_decl((4,), "float32x8", data=matmul)
+                    matmul_1[0] = T.Broadcast(T.float32(0), 8)
+                    matmul_1[1] = T.Broadcast(T.float32(0), 8)
+                    matmul_1[2] = T.Broadcast(T.float32(0), 8)
+                    matmul_1[3] = T.Broadcast(T.float32(0), 8)
+                    for k_outer, k_inner in T.grid(256, 4):
+                        cse_var_2: T.int32 = i_outer_outer_j_outer_outer_fused % 128 * 8192 + k_outer * 32 + k_inner * 8
+                        cse_var_1: T.int32 = i_outer_outer_j_outer_outer_fused // 128 * 8192 + i_outer_inner * 4096 + k_outer * 4 + k_inner
+                        A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                        matmul_1[0] = matmul_1[0] + T.Broadcast(A_2[cse_var_1], 8) * auto_scheduler_layout_transform_1[cse_var_2:cse_var_2 + 8]
+                        matmul_1[1] = matmul_1[1] + T.Broadcast(A_2[cse_var_1 + 1024], 8) * auto_scheduler_layout_transform_1[cse_var_2:cse_var_2 + 8]
+                        matmul_1[2] = matmul_1[2] + T.Broadcast(A_2[cse_var_1 + 2048], 8) * auto_scheduler_layout_transform_1[cse_var_2:cse_var_2 + 8]
+                        matmul_1[3] = matmul_1[3] + T.Broadcast(A_2[cse_var_1 + 3072], 8) * auto_scheduler_layout_transform_1[cse_var_2:cse_var_2 + 8]
+                    for i_inner in range(4):
+                        cse_var_3: T.int32 = i_outer_outer_j_outer_outer_fused // 128 * 8192 + i_outer_inner * 4096 + i_inner * 1024 + i_outer_outer_j_outer_outer_fused % 128 * 8
+                        out_2 = T.buffer_decl((1048576,), data=out_1.data)
+                        C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                        out_2[cse_var_3:cse_var_3 + 8] = matmul_1[i_inner] + C_2[cse_var_3:cse_var_3 + 8]
 
 
 
@@ -336,7 +319,7 @@ We build the binary and check its correctness and performance.
 
  .. code-block:: none
 
-    Execution time of this operator: 98.473 ms
+    Execution time of this operator: 96.150 ms
 
 
 
@@ -454,7 +437,7 @@ operations.
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 1 minutes  28.083 seconds)
+   **Total running time of the script:** ( 1 minutes  13.404 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 27d450d793..9e85c96d02 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: 2.53/2.53       result: MeasureResult(costs=(0.1061960336,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.962169885635376, timestamp=1673995581.1326325)        [('tile_y', [-1, 8]), ('tile_x', [-1, 4])],None,23
-    No: 2   GFLOPS: 13.40/13.40     result: MeasureResult(costs=(0.020038429400000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.570389986038208, timestamp=1673995581.7051456)        [('tile_y', [-1, 256]), ('tile_x', [-1, 64])],None,68
-    No: 3   GFLOPS: 0.50/13.40      result: MeasureResult(costs=(0.5352776608,), error_no=MeasureErrorNo.NO_ERROR, all_cost=8.805688619613647, timestamp=1673995591.2741747)        [('tile_y', [-1, 64]), ('tile_x', [-1, 1])],None,6
-    No: 4   GFLOPS: 11.94/13.40     result: MeasureResult(costs=(0.0224818592,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6071557998657227, timestamp=1673995592.6464636)       [('tile_y', [-1, 32]), ('tile_x', [-1, 256])],None,85
-    No: 5   GFLOPS: 12.65/13.40     result: MeasureResult(costs=(0.021226347200000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6028244495391846, timestamp=1673995593.3651145)       [('tile_y', [-1, 64]), ('tile_x', [-1, 256])],None,86
-    No: 6   GFLOPS: 12.77/13.40     result: MeasureResult(costs=(0.021023800000000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.5934467315673828, timestamp=1673995594.719418)        [('tile_y', [-1, 4]), ('tile_x', [-1, 256])],None,82
-    No: 7   GFLOPS: 2.07/13.40      result: MeasureResult(costs=(0.1293966956,), error_no=MeasureErrorNo.NO_ERROR, all_cost=2.307110071182251, timestamp=1673995597.0376809)        [('tile_y', [-1, 256]), ('tile_x', [-1, 4])],None,28
-    No: 8   GFLOPS: 12.33/13.40     result: MeasureResult(costs=(0.0217772238,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6025233268737793, timestamp=1673995597.6416461)       [('tile_y', [-1, 8]), ('tile_x', [-1, 256])],None,83
-    No: 9   GFLOPS: 10.56/13.40     result: MeasureResult(costs=(0.025431841599999998,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6511471271514893, timestamp=1673995598.4557238)       [('tile_y', [-1, 4]), ('tile_x', [-1, 128])],None,72
-    No: 10  GFLOPS: 12.76/13.40     result: MeasureResult(costs=(0.021030583999999998,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6265027523040771, timestamp=1673995599.0469368)       [('tile_y', [-1, 256]), ('tile_x', [-1, 128])],None,78
+    No: 1   GFLOPS: 12.07/12.07     result: MeasureResult(costs=(0.022236667800000002,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.6587295532226562, timestamp=1673996505.5195735)       [('tile_y', [-1, 64]), ('tile_x', [-1, 32])],None,56
+    No: 2   GFLOPS: 1.55/12.07      result: MeasureResult(costs=(0.1730933026,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.0049822330474854, timestamp=1673996508.5317776)       [('tile_y', [-1, 64]), ('tile_x', [-1, 4])],None,26
+    No: 3   GFLOPS: 3.76/12.07      result: MeasureResult(costs=(0.0713306806,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3891253471374512, timestamp=1673996510.6750538)       [('tile_y', [-1, 256]), ('tile_x', [-1, 16])],None,48
+    No: 4   GFLOPS: 10.55/12.07     result: MeasureResult(costs=(0.025447205,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.7185344696044922, timestamp=1673996511.3389113)        [('tile_y', [-1, 1]), ('tile_x', [-1, 128])],None,70
+    No: 5   GFLOPS: 0.90/12.07      result: MeasureResult(costs=(0.2986694228,), error_no=MeasureErrorNo.NO_ERROR, all_cost=5.002816915512085, timestamp=1673996516.4702854)        [('tile_y', [-1, 64]), ('tile_x', [-1, 2])],None,16
+    No: 6   GFLOPS: 4.47/12.07      result: MeasureResult(costs=(0.060065444999999995,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.2160968780517578, timestamp=1673996518.4282773)       [('tile_y', [-1, 8]), ('tile_x', [-1, 16])],None,43
+    No: 7   GFLOPS: 1.30/12.07      result: MeasureResult(costs=(0.206241246,), error_no=MeasureErrorNo.NO_ERROR, all_cost=3.5250444412231445, timestamp=1673996522.729668) [('tile_y', [-1, 1]), ('tile_x', [-1, 2])],None,10
+    No: 8   GFLOPS: 3.28/12.07      result: MeasureResult(costs=(0.081944288,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.5534205436706543, timestamp=1673996524.2894173)        [('tile_y', [-1, 32]), ('tile_x', [-1, 8])],None,35
+    No: 9   GFLOPS: 3.67/12.07      result: MeasureResult(costs=(0.0730662638,), error_no=MeasureErrorNo.NO_ERROR, all_cost=1.3784840106964111, timestamp=1673996525.7949874)       [('tile_y', [-1, 8]), ('tile_x', [-1, 8])],None,33
+    No: 10  GFLOPS: 12.63/12.63     result: MeasureResult(costs=(0.0212478146,), error_no=MeasureErrorNo.NO_ERROR, all_cost=0.5707542896270752, timestamp=1673996526.390976)        [('tile_y', [-1, 128]), ('tile_x', [-1, 256])],None,87
 
 
 
diff --git a/docs/_sources/tutorial/autotvm_relay_x86.rst.txt b/docs/_sources/tutorial/autotvm_relay_x86.rst.txt
index 609be1607c..53fa541011 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': 521.6796194400035, 'median': 522.0597105500076, 'std': 2.9126359836131974}
+    {'mean': 507.22305444999967, 'median': 506.30012315000386, 'std': 5.391398088033922}
 
 
 
@@ -545,31 +545,30 @@ 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:    8.80/  23.55 GFLOPS | Progress: (4/20) | 8.95 s
    [Task  1/25]  Current/Best:   22.01/  23.55 GFLOPS | Progress: (8/20) | 12.53 s
    [Task  1/25]  Current/Best:   22.11/  23.55 GFLOPS | Progress: (12/20) | 14.80 s
    [Task  1/25]  Current/Best:   22.77/  23.55 GFLOPS | Progress: (16/20) | 16.90 s
    [Task  1/25]  Current/Best:   10.97/  23.55 GFLOPS | Progress: (20/20) | 20.71 s Done.
-
    [Task  2/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  2/25]  Current/Best:    2.07/  18.63 GFLOPS | Progress: (4/20) | 3.44 s
    [Task  2/25]  Current/Best:   11.14/  19.02 GFLOPS | Progress: (8/20) | 5.31 s
    [Task  2/25]  Current/Best:   16.18/  19.02 GFLOPS | Progress: (12/20) | 7.01 s
    [Task  2/25]  Current/Best:    8.57/  19.02 GFLOPS | Progress: (16/20) | 8.61 s
    [Task  2/25]  Current/Best:   10.25/  19.02 GFLOPS | Progress: (20/20) | 11.53 s Done.
-
    [Task  3/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  3/25]  Current/Best:   12.56/  12.56 GFLOPS | Progress: (4/20) | 4.47 s
    [Task  3/25]  Current/Best:    8.34/  17.10 GFLOPS | Progress: (8/20) | 7.25 s
    [Task  3/25]  Current/Best:    6.98/  17.68 GFLOPS | Progress: (12/20) | 10.13 s
    [Task  3/25]  Current/Best:    6.37/  17.68 GFLOPS | Progress: (16/20) | 12.70 s
    [Task  3/25]  Current/Best:    9.28/  18.43 GFLOPS | Progress: (20/20) | 16.33 s Done.
-
    [Task  4/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  4/25]  Current/Best:   15.45/  17.63 GFLOPS | Progress: (4/20) | 3.87 s
    [Task  4/25]  Current/Best:    9.68/  17.63 GFLOPS | Progress: (8/20) | 6.75 s
    [Task  4/25]  Current/Best:    6.87/  18.62 GFLOPS | Progress: (12/20) | 12.11 s
    [Task  4/25]  Current/Best:   12.65/  18.62 GFLOPS | Progress: (16/20) | 14.01 s
    [Task  4/25]  Current/Best:   18.84/  18.84 GFLOPS | Progress: (20/20) | 15.89 s Done.
-
    [Task  5/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  5/25]  Current/Best:    5.75/  20.46 GFLOPS | Progress: (4/20) | 4.32 s
    [Task  5/25]  Current/Best:   11.38/  20.46 GFLOPS | Progress: (8/20) | 6.29 s
    [Task  5/25]  Current/Best:   14.84/  20.46 GFLOPS | Progress: (12/20) | 7.89 s
    [Task  5/25]  Current/Best:   15.90/  20.46 GFLOPS | Progress: (16/20) | 10.47 s
    [Task  5/25]  Current/Best:   16.48/  20.46 GFLOPS | Progress: (20/20) | 12.84 s Done.
-
    [Task  6/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  6/25]  Current/Best:   14.25/  14.51 GFLOPS | Progress: (4/20) | 4.17 s
    [Task  6/25]  Current/Best:    9.57/  14.51 GFLOPS | Progress: (8/20) | 7.21 s
    [Task  6/25]  Current/Best:   14.69/  14.69 GFLOPS | Progress: (12/20) | 10.80 s
    [Task  6/25]  Current/Best:   12.05/  14.82 GFLOPS | Progress: (16/20) | 13.20 s
    [Task  6/25]  Current/Best:    9.84/  15.85 GFLOPS | Progress: (20/20) | 20.15 s Done.
-
    [Task  7/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  7/25]  Current/Best:   15.11/  16.80 GFLOPS | Progress: (4/20) | 4.04 s
    [Task  7/25]  Current/Best:   12.04/  21.99 GFLOPS | Progress: (8/20) | 6.63 s
    [Task  7/25]  Current/Best:   15.86/  21.99 GFLOPS | Progress: (12/20) | 8.72 s
    [Task  7/25]  Current/Best:   11.15/  21.99 GFLOPS | Progress: (16/20) | 11.46 s
    [Task  7/25]  Current/Best:    1.58/  21.99 GFLOPS | Progress: (20/20) | 16.05 s Done.
-
    [Task  8/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  8/25]  Current/Best:   15.45/  15.45 GFLOPS | Progress: (4/20) | 13.32 s
    [Task  8/25]  Current/Best:   14.02/  15.45 GFLOPS | Progress: (8/20) | 16.81 s
    [Task  8/25]  Current/Best:    9.87/  15.45 GFLOPS | Progress: (12/20) | 22.57 s
    [Task  8/25]  Current/Best:   10.83/  15.45 GFLOPS | Progress: (16/20) | 29.65 s
    [Task  8/25]  Current/Best:   11.52/  15.45 GFLOPS | Progress: (20/20) | 34.11 s
    [Task  9/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  9/25]  Current/Best:   19.89/  19.89 GFLOPS | Progress: (4/20) | 12.95 s
    [Task  9/25]  Current/Best:   16.36/  19.89 GFLOPS | Progress: (8/20) | 15.52 s
    [Task  9/25]  Current/Best:   12.37/  19.89 GFLOPS | Progress: (12/20) | 17.14 s
    [Task  9/25]  Current/Best:   11.57/  19.89 GFLOPS | Progress: (16/20) | 22.41 s
    [Task  9/25]  Current/Best:    8.27/  19.89 GFLOPS | Progress: (2
 0/20) | 24.02 s
    [Task 10/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 10/25]  Current/Best:    5.02/  12.62 GFLOPS | Progress: (4/20) | 3.95 s
    [Task 10/25]  Current/Best:    5.90/  18.93 GFLOPS | Progress: (8/20) | 6.02 s
    [Task 10/25]  Current/Best:    4.76/  18.93 GFLOPS | Progress: (12/20) | 8.41 s
    [Task 10/25]  Current/Best:   15.91/  18.93 GFLOPS | Progress: (16/20) | 10.43 s
    [Task 10/25]  Current/Best:    1.60/  18.93 GFLOPS | Progress: (20/20) | 15.08 s Done.
-
    [Task 11/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 11/25]  Current/Best:   20.87/  22.60 GFLOPS | Progress: (4/20) | 3.67 s
    [Task 11/25]  Current/Best:   12.56/  22.60 GFLOPS | Progress: (8/20) | 6.33 s
    [Task 11/25]  Current/Best:    9.76/  22.60 GFLOPS | Progress: (12/20) | 8.74 s
    [Task 11/25]  Current/Best:    6.86/  22.60 GFLOPS | Progress: (16/20) | 12.08 s
    [Task 11/25]  Current/Best:   10.07/  22.60 GFLOPS | Progress: (20/20) | 15.02 s Done.
-
    [Task 12/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 12/25]  Current/Best:   12.50/  12.88 GFLOPS | Progress: (4/20) | 4.02 s
    [Task 12/25]  Current/Best:   14.99/  18.11 GFLOPS | Progress: (8/20) | 6.13 s
    [Task 12/25]  Current/Best:   18.74/  18.74 GFLOPS | Progress: (12/20) | 9.75 s
    [Task 12/25]  Current/Best:   14.84/  21.93 GFLOPS | Progress: (16/20) | 12.12 s
    [Task 12/25]  Current/Best:    9.13/  21.93 GFLOPS | Progress: (20/20) | 20.45 s Done.
-
    [Task 13/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 13/25]  Current/Best:   13.67/  17.52 GFLOPS | Progress: (4/20) | 4.39 s
    [Task 13/25]  Current/Best:    9.71/  17.52 GFLOPS | Progress: (8/20) | 7.69 s
    [Task 13/25]  Current/Best:   15.82/  19.71 GFLOPS | Progress: (12/20) | 10.30 s
    [Task 13/25]  Current/Best:   10.81/  19.71 GFLOPS | Progress: (16/20) | 13.12 s
    [Task 13/25]  Current/Best:    7.98/  19.71 GFLOPS | Progress: (20/20) | 15.78 s Done.
-
    [Task 14/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 14/25]  Current/Best:    9.52/  22.01 GFLOPS | Progress: (4/20) | 4.07 s
    [Task 14/25]  Current/Best:    6.24/  22.01 GFLOPS | Progress: (8/20) | 10.01 s
    [Task 14/25]  Current/Best:    4.15/  22.01 GFLOPS | Progress: (12/20) | 13.27 s
    [Task 14/25]  Current/Best:    4.97/  22.01 GFLOPS | Progress: (16/20) | 16.22 s Done.
-     Done.
-
    [Task 14/25]  Current/Best:   17.55/  22.01 GFLOPS | Progress: (20/20) | 19.14 s Done.
-
    [Task 15/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 15/25]  Current/Best:   19.33/  19.33 GFLOPS | Progress: (4/20) | 4.38 s
    [Task 15/25]  Current/Best:   21.52/  21.52 GFLOPS | Progress: (8/20) | 7.42 s
    [Task 15/25]  Current/Best:    8.18/  21.52 GFLOPS | Progress: (12/20) | 9.53 s
    [Task 15/25]  Current/Best:   13.40/  22.11 GFLOPS | Progress: (16/20) | 12.33 s
    [Task 15/25]  Current/Best:   23.16/  23.16 GFLOPS | Progress: (20/20) | 14.08 s
    [Task 16/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 16/25]  Current/Best:   21.15/  21.15 GFLOPS | Progress: (4/20) | 4.27 s
    [Task 16/25]  Current/Best:   18.38/  21.15 GFLOPS | Progress: (8/20) | 6.02 s
    [Task 16/25]  Current/Best:    4.97/  21.15 GFLOPS | Progress: (12/20) | 8.41 s
    [Task 16/25]  Current/Best:   18.04/  21.15 GFLOPS | Progress: (16/20) | 9.94 s
    [Task 16/25]  Current/Best:   16.37/  21.15 GFLOPS | Progress: (20/20) |
  11.51 s Done.
-
    [Task 17/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 17/25]  Current/Best:   17.01/  17.97 GFLOPS | Progress: (4/20) | 4.39 s
    [Task 17/25]  Current/Best:    6.14/  17.99 GFLOPS | Progress: (8/20) | 7.36 s
    [Task 17/25]  Current/Best:   17.35/  17.99 GFLOPS | Progress: (12/20) | 10.02 s
    [Task 17/25]  Current/Best:   10.80/  22.81 GFLOPS | Progress: (16/20) | 12.57 s
    [Task 17/25]  Current/Best:    6.13/  22.81 GFLOPS | Progress: (20/20) | 16.19 s Done.
-
    [Task 18/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 18/25]  Current/Best:   19.12/  19.12 GFLOPS | Progress: (4/20) | 5.51 s
    [Task 18/25]  Current/Best:   10.65/  19.12 GFLOPS | Progress: (8/20) | 7.47 s
    [Task 18/25]  Current/Best:   19.62/  19.62 GFLOPS | Progress: (12/20) | 10.70 s
    [Task 18/25]  Current/Best:   12.63/  19.62 GFLOPS | Progress: (16/20) | 13.51 s
    [Task 18/25]  Current/Best:    8.18/  19.62 GFLOPS | Progress: (20/20) | 16.91 s Done.
-
    [Task 19/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 19/25]  Current/Best:    9.17/  14.57 GFLOPS | Progress: (4/20) | 4.34 s
    [Task 19/25]  Current/Best:    8.34/  19.15 GFLOPS | Progress: (8/20) | 8.38 s
    [Task 19/25]  Current/Best:   16.18/  19.15 GFLOPS | Progress: (12/20) | 11.62 s
    [Task 19/25]  Current/Best:   11.66/  22.97 GFLOPS | Progress: (16/20) | 18.02 s
    [Task 19/25]  Current/Best:   17.82/  22.97 GFLOPS | Progress: (20/20) | 20.42 s Done.
-
    [Task 20/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 20/25]  Current/Best:   17.73/  17.73 GFLOPS | Progress: (4/20) | 6.67 s
    [Task 20/25]  Current/Best:   15.44/  17.73 GFLOPS | Progress: (8/20) | 8.11 s
    [Task 20/25]  Current/Best:    9.97/  17.73 GFLOPS | Progress: (12/20) | 12.71 s
    [Task 20/25]  Current/Best:    6.17/  17.73 GFLOPS | Progress: (16/20) | 15.84 s
    [Task 20/25]  Current/Best:   11.36/  17.73 GFLOPS | Progress: (20/20) | 17.59 s
    [Task 21/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
-     Done.
-
    [Task 21/25]  Current/Best:   13.53/  13.53 GFLOPS | Progress: (4/20) | 4.42 s
    [Task 21/25]  Current/Best:   19.58/  20.29 GFLOPS | Progress: (8/20) | 6.58 s
    [Task 21/25]  Current/Best:   12.95/  20.29 GFLOPS | Progress: (12/20) | 8.47 s
    [Task 21/25]  Current/Best:   23.55/  23.55 GFLOPS | Progress: (16/20) | 11.79 s
    [Task 21/25]  Current/Best:   17.04/  23.55 GFLOPS | Progress: (20/20) | 14.16 s
    [Task 22/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 22/25]  Current/Best:    6.19/  20.95 GFLOPS | Progress: (4/20) | 3.39 s
    [Task 22/25]  Current/Best:   15.79/  20.95 GFLOPS | Progress: (8/20) | 6.35 s
    [Task 22/25]  Current/Best:   10.26/  20.95 GFLOPS | Progress: (12/20) | 9.12 s
    [Task 22/25]  Current/Best:   12.86/  20.95 GFLOPS | Progress: (16/20) | 10.95 s
    [Task 22/25]  Current/Best:   17.46/  21.32 GFLOPS | Progress: (20/20) | 12.67 s Done.
-
    [Task 23/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 23/25]  Current/Best:    2.69/  11.00 GFLOPS | Progress: (4/20) | 5.74 s
    [Task 23/25]  Current/Best:   10.51/  11.76 GFLOPS | Progress: (8/20) | 10.41 s
    [Task 23/25]  Current/Best:   13.28/  21.52 GFLOPS | Progress: (12/20) | 12.76 s
    [Task 23/25]  Current/Best:   11.12/  21.52 GFLOPS | Progress: (16/20) | 16.02 s
    [Task 23/25]  Current/Best:    9.07/  21.52 GFLOPS | Progress: (20/20) | 18.98 s Done.
-
    [Task 24/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 24/25]  Current/Best:    6.87/   6.87 GFLOPS | Progress: (4/20) | 8.09 s
    [Task 24/25]  Current/Best:    4.04/   6.87 GFLOPS | Progress: (8/20) | 19.04 s
    [Task 24/25]  Current/Best:    9.05/   9.05 GFLOPS | Progress: (12/20) | 22.43 s
    [Task 24/25]  Current/Best:    0.90/   9.05 GFLOPS | Progress: (16/20) | 29.07 s
    [Task 24/25]  Current/Best:    4.55/   9.05 GFLOPS | Progress: (20/20) | 35.49 s
    [Task 25/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
-     Done.
-
    [Task 25/25]  Current/Best:    9.86/   9.86 GFLOPS | Progress: (4/20) | 13.51 s
    [Task 25/25]  Current/Best:    2.90/   9.86 GFLOPS | Progress: (8/20) | 25.23 s
    [Task 25/25]  Current/Best:    2.98/   9.86 GFLOPS | Progress: (12/20) | 29.69 s
    [Task 25/25]  Current/Best:    7.77/   9.86 GFLOPS | Progress: (16/20) | 31.20 s
    [Task 25/25]  Current/Best:    4.33/   9.86 GFLOPS | Progress: (20/20) | 33.20 s
+
    [Task  1/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  1/25]  Current/Best:    8.77/  22.61 GFLOPS | Progress: (4/20) | 8.62 s
    [Task  1/25]  Current/Best:    9.95/  23.43 GFLOPS | Progress: (8/20) | 12.74 s
    [Task  1/25]  Current/Best:    6.38/  23.43 GFLOPS | Progress: (12/20) | 15.81 s
    [Task  1/25]  Current/Best:   15.11/  23.43 GFLOPS | Progress: (16/20) | 18.06 s
    [Task  1/25]  Current/Best:    7.38/  23.62 GFLOPS | Progress: (20/20) | 20.31 s Done.
+
    [Task  2/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  2/25]  Current/Best:   19.14/  19.14 GFLOPS | Progress: (4/20) | 3.78 s
    [Task  2/25]  Current/Best:    8.80/  19.14 GFLOPS | Progress: (8/20) | 5.58 s
    [Task  2/25]  Current/Best:   16.41/  19.14 GFLOPS | Progress: (12/20) | 7.03 s
    [Task  2/25]  Current/Best:    5.95/  19.14 GFLOPS | Progress: (16/20) | 8.55 s
    [Task  2/25]  Current/Best:    8.74/  19.14 GFLOPS | Progress: (20/20) | 9.92 s Done.
+
    [Task  3/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  3/25]  Current/Best:   14.96/  23.88 GFLOPS | Progress: (4/20) | 3.93 s
    [Task  3/25]  Current/Best:    6.73/  23.88 GFLOPS | Progress: (8/20) | 6.94 s
    [Task  3/25]  Current/Best:   18.57/  23.88 GFLOPS | Progress: (12/20) | 9.44 s
    [Task  3/25]  Current/Best:   21.50/  23.88 GFLOPS | Progress: (16/20) | 11.53 s
    [Task  3/25]  Current/Best:   12.85/  23.88 GFLOPS | Progress: (20/20) | 14.28 s Done.
+
    [Task  4/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  4/25]  Current/Best:   15.79/  15.79 GFLOPS | Progress: (4/20) | 8.60 s
    [Task  4/25]  Current/Best:   11.96/  15.79 GFLOPS | Progress: (8/20) | 12.53 s
    [Task  4/25]  Current/Best:   16.66/  20.93 GFLOPS | Progress: (12/20) | 14.39 s
    [Task  4/25]  Current/Best:   15.52/  20.93 GFLOPS | Progress: (16/20) | 16.34 s
    [Task  4/25]  Current/Best:    7.01/  21.73 GFLOPS | Progress: (20/20) | 18.62 s Done.
+
    [Task  5/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  5/25]  Current/Best:    9.31/  14.86 GFLOPS | Progress: (4/20) | 3.80 s
    [Task  5/25]  Current/Best:    9.11/  16.94 GFLOPS | Progress: (8/20) | 5.91 s
    [Task  5/25]  Current/Best:   13.10/  19.93 GFLOPS | Progress: (12/20) | 7.59 s
    [Task  5/25]  Current/Best:   12.19/  19.93 GFLOPS | Progress: (16/20) | 10.89 s
    [Task  5/25]  Current/Best:   13.89/  19.93 GFLOPS | Progress: (20/20) | 13.68 s Done.
+
    [Task  6/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  6/25]  Current/Best:   15.56/  17.86 GFLOPS | Progress: (4/20) | 4.69 s
    [Task  6/25]  Current/Best:    6.89/  17.86 GFLOPS | Progress: (8/20) | 7.13 s
    [Task  6/25]  Current/Best:    6.09/  17.86 GFLOPS | Progress: (12/20) | 9.72 s
    [Task  6/25]  Current/Best:    8.58/  17.86 GFLOPS | Progress: (16/20) | 13.11 s
    [Task  6/25]  Current/Best:   13.47/  17.86 GFLOPS | Progress: (20/20) | 15.78 s Done.
+
    [Task  7/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  7/25]  Current/Best:    7.82/  19.40 GFLOPS | Progress: (4/20) | 3.99 s
    [Task  7/25]  Current/Best:   12.28/  19.40 GFLOPS | Progress: (8/20) | 6.15 s
    [Task  7/25]  Current/Best:   12.96/  19.40 GFLOPS | Progress: (12/20) | 9.13 s
    [Task  7/25]  Current/Best:   18.29/  19.40 GFLOPS | Progress: (16/20) | 11.42 s
    [Task  7/25]  Current/Best:   18.54/  19.40 GFLOPS | Progress: (20/20) | 13.77 s Done.
+
    [Task  8/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  8/25]  Current/Best:   10.76/  21.13 GFLOPS | Progress: (4/20) | 6.82 s
    [Task  8/25]  Current/Best:    8.83/  21.13 GFLOPS | Progress: (8/20) | 9.24 s
    [Task  8/25]  Current/Best:    4.07/  21.13 GFLOPS | Progress: (12/20) | 21.17 s
    [Task  8/25]  Current/Best:   11.52/  22.76 GFLOPS | Progress: (16/20) | 23.53 s
    [Task  8/25]  Current/Best:    5.19/  22.76 GFLOPS | Progress: (20/20) | 26.63 s Done.
+
    [Task  9/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task  9/25]  Current/Best:    3.19/  15.36 GFLOPS | Progress: (4/20) | 7.37 s
    [Task  9/25]  Current/Best:   19.20/  19.20 GFLOPS | Progress: (8/20) | 9.98 s
    [Task  9/25]  Current/Best:   12.07/  19.60 GFLOPS | Progress: (12/20) | 12.28 s
    [Task  9/25]  Current/Best:   15.91/  19.60 GFLOPS | Progress: (16/20) | 14.93 s
    [Task  9/25]  Current/Best:   21.88/  21.88 GFLOPS | Progress: (20/20) | 19.83 s Done.
+
    [Task 10/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 10/25]  Current/Best:   11.10/  15.69 GFLOPS | Progress: (4/20) | 3.66 s
    [Task 10/25]  Current/Best:   15.10/  15.69 GFLOPS | Progress: (8/20) | 5.55 s
    [Task 10/25]  Current/Best:    5.41/  15.69 GFLOPS | Progress: (12/20) | 8.83 s
    [Task 10/25]  Current/Best:   17.15/  20.97 GFLOPS | Progress: (16/20) | 10.26 s
    [Task 10/25]  Current/Best:    5.25/  20.97 GFLOPS | Progress: (20/20) | 12.14 s Done.
+
    [Task 11/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 11/25]  Current/Best:   16.25/  16.25 GFLOPS | Progress: (4/20) | 3.96 s
    [Task 11/25]  Current/Best:   21.11/  21.11 GFLOPS | Progress: (8/20) | 6.32 s
    [Task 11/25]  Current/Best:   14.40/  21.11 GFLOPS | Progress: (12/20) | 8.89 s
    [Task 11/25]  Current/Best:    6.75/  21.11 GFLOPS | Progress: (16/20) | 11.47 s
    [Task 11/25]  Current/Best:    9.66/  21.11 GFLOPS | Progress: (20/20) | 14.13 s Done.
+
    [Task 12/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 12/25]  Current/Best:    4.77/  14.21 GFLOPS | Progress: (4/20) | 4.71 s
    [Task 12/25]  Current/Best:    6.15/  14.49 GFLOPS | Progress: (8/20) | 7.11 s
    [Task 12/25]  Current/Best:   10.34/  14.49 GFLOPS | Progress: (12/20) | 10.15 s
    [Task 12/25]  Current/Best:    9.72/  22.39 GFLOPS | Progress: (16/20) | 15.79 s
    [Task 12/25]  Current/Best:   13.54/  22.39 GFLOPS | Progress: (20/20) | 18.25 s Done.
+
    [Task 13/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 13/25]  Current/Best:    3.10/  16.32 GFLOPS | Progress: (4/20) | 5.94 s
    [Task 13/25]  Current/Best:   12.16/  21.63 GFLOPS | Progress: (8/20) | 9.15 s
    [Task 13/25]  Current/Best:    6.72/  21.63 GFLOPS | Progress: (12/20) | 12.20 s
    [Task 13/25]  Current/Best:   17.29/  21.63 GFLOPS | Progress: (16/20) | 15.07 s
    [Task 13/25]  Current/Best:   11.61/  21.63 GFLOPS | Progress: (20/20) | 19.25 s Done.
+
    [Task 14/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 14/25]  Current/Best:   15.16/  15.16 GFLOPS | Progress: (4/20) | 5.35 s
    [Task 14/25]  Current/Best:   12.75/  15.16 GFLOPS | Progress: (8/20) | 7.84 s
    [Task 14/25]  Current/Best:    9.87/  21.34 GFLOPS | Progress: (12/20) | 13.48 s
    [Task 14/25]  Current/Best:    3.16/  21.34 GFLOPS | Progress: (16/20) | 18.32 s
    [Task 14/25]  Current/Best:   11.49/  21.34 GFLOPS | Progress: (20/20) | 21.35 s
    [Task 15/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 15/25]  Current/Best:    7.11/  21.93 GFLOPS | Progress: (4/20) | 5.43 s
    [Task 15/25]  Current/Best:   11.49/  21.93 GFLOPS | Progress: (8/20) | 8.27 s
    [Task 15/25]  Current/Best:   11.09/  21.93 GFLOPS | Progress: (12/20) | 11.56 s
    [Task 15/25]  Current/Best:   16.79/  21.93 GFLOPS | Progress: (16/20) | 13.08 s Done.
+
    [Task 15/25]  Current/Best:   13.50/  21.93 GFLOPS | Progress: (20/20) | 15.55 s
    [Task 16/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 16/25]  Current/Best:   18.60/  18.60 GFLOPS | Progress: (4/20) | 4.25 s
    [Task 16/25]  Current/Best:    5.74/  18.60 GFLOPS | Progress: (8/20) | 6.73 s
    [Task 16/25]  Current/Best:    2.81/  18.60 GFLOPS | Progress: (12/20) | 9.01 s
    [Task 16/25]  Current/Best:   14.42/  19.68 GFLOPS | Progress: (16/20) | 10.84 s
    [Task 16/25]  Current/Best:   13.92/  19.97 GFLOPS | Progress: (20/20) | 13.02 s Done.
+
    [Task 17/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 17/25]  Current/Best:    9.66/  12.66 GFLOPS | Progress: (4/20) | 5.01 s
    [Task 17/25]  Current/Best:   13.89/  13.89 GFLOPS | Progress: (8/20) | 7.50 s
    [Task 17/25]  Current/Best:   12.30/  17.99 GFLOPS | Progress: (12/20) | 10.26 s
    [Task 17/25]  Current/Best:    6.16/  17.99 GFLOPS | Progress: (16/20) | 12.77 s
    [Task 17/25]  Current/Best:    9.85/  19.00 GFLOPS | Progress: (20/20) | 15.51 s Done.
+
    [Task 18/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 18/25]  Current/Best:    5.59/  18.48 GFLOPS | Progress: (4/20) | 5.22 s
    [Task 18/25]  Current/Best:   15.15/  18.48 GFLOPS | Progress: (8/20) | 8.09 s
    [Task 18/25]  Current/Best:   18.23/  21.10 GFLOPS | Progress: (12/20) | 10.72 s
    [Task 18/25]  Current/Best:    6.25/  21.10 GFLOPS | Progress: (16/20) | 15.77 s
    [Task 18/25]  Current/Best:   11.86/  21.10 GFLOPS | Progress: (20/20) | 18.26 s Done.
+
    [Task 19/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 19/25]  Current/Best:   11.96/  21.67 GFLOPS | Progress: (4/20) | 4.69 s
    [Task 19/25]  Current/Best:   21.22/  21.67 GFLOPS | Progress: (8/20) | 7.22 s
    [Task 19/25]  Current/Best:    6.16/  21.67 GFLOPS | Progress: (12/20) | 10.77 s
    [Task 19/25]  Current/Best:   10.23/  21.67 GFLOPS | Progress: (16/20) | 15.23 s
    [Task 19/25]  Current/Best:    8.34/  21.67 GFLOPS | Progress: (20/20) | 18.79 s Done.
+
    [Task 20/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 20/25]  Current/Best:   17.78/  17.78 GFLOPS | Progress: (4/20) | 5.05 s
    [Task 20/25]  Current/Best:   11.08/  17.78 GFLOPS | Progress: (8/20) | 6.84 s
    [Task 20/25]  Current/Best:   11.35/  18.42 GFLOPS | Progress: (12/20) | 12.44 s
    [Task 20/25]  Current/Best:    5.21/  18.42 GFLOPS | Progress: (16/20) | 15.46 s
    [Task 20/25]  Current/Best:    5.97/  18.42 GFLOPS | Progress: (20/20) | 17.76 s
    [Task 21/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 21/25]  Current/Best:   16.77/  16.77 GFLOPS | Progress: (4/20) | 3.39 s
    [Task 21/25]  Current/Best:    1.61/  19.11 GFLOPS | Progress: (8/20) | 6.03 s
    [Task 21/25]  Current/Best:    8.07/  19.11 GFLOPS | Progress: (12/20) | 8.45 s
    [Task 21/25]  Current/Best:    8.78/  19.11 GFLOPS | Progress: (16/20) | 10.78 s Done.
+
    [Task 21/25]  Current/Best:   13.26/  19.11 GFLOPS | Progress: (20/20) | 13.18 s Done.
+
    [Task 22/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 22/25]  Current/Best:   11.85/  18.05 GFLOPS | Progress: (4/20) | 4.39 s
    [Task 22/25]  Current/Best:    4.68/  18.05 GFLOPS | Progress: (8/20) | 6.80 s
    [Task 22/25]  Current/Best:   15.08/  18.05 GFLOPS | Progress: (12/20) | 10.02 s
    [Task 22/25]  Current/Best:   12.66/  18.05 GFLOPS | Progress: (16/20) | 15.03 s
    [Task 22/25]  Current/Best:   10.08/  18.05 GFLOPS | Progress: (20/20) | 17.93 s Done.
+
    [Task 23/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 23/25]  Current/Best:    9.51/  20.33 GFLOPS | Progress: (4/20) | 5.23 s
    [Task 23/25]  Current/Best:    9.65/  20.33 GFLOPS | Progress: (8/20) | 10.92 s
    [Task 23/25]  Current/Best:   15.44/  21.47 GFLOPS | Progress: (12/20) | 13.65 s
    [Task 23/25]  Current/Best:    9.64/  21.47 GFLOPS | Progress: (16/20) | 16.44 s
    [Task 23/25]  Current/Best:    1.55/  21.47 GFLOPS | Progress: (20/20) | 20.87 s Done.
+
    [Task 24/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s
    [Task 24/25]  Current/Best:    4.59/   4.59 GFLOPS | Progress: (4/20) | 12.17 s
    [Task 24/25]  Current/Best:    5.44/   7.72 GFLOPS | Progress: (8/20) | 23.98 s
    [Task 24/25]  Current/Best:    5.46/   7.72 GFLOPS | Progress: (12/20) | 35.77 s
    [Task 24/25]  Current/Best:    1.55/   9.79 GFLOPS | Progress: (16/20) | 46.71 s
    [Task 24/25]  Current/Best:    3.77/   9.79 GFLOPS | Progress: (20/20) | 57.67 s
    [Task 25/25]  Current/Best:    0.00/   0.00 GFLOPS | Progress: (0/20) | 0.00 s Done.
+
    [Task 25/25]  Current/Best:    5.84/   8.97 GFLOPS | Progress: (4/20) | 3.92 s
    [Task 25/25]  Current/Best:    1.54/   8.97 GFLOPS | Progress: (8/20) | 12.23 s
    [Task 25/25]  Current/Best:    8.96/   8.97 GFLOPS | Progress: (12/20) | 22.59 s
    [Task 25/25]  Current/Best:    5.05/   8.97 GFLOPS | Progress: (16/20) | 34.66 s
    [Task 25/25]  Current/Best:    5.92/   8.97 GFLOPS | Progress: (20/20) | 45.30 s
 
 
 
@@ -666,7 +665,7 @@ Verify that the optimized model runs and produces the same results:
  .. code-block:: none
 
     class='n02123045 tabby, tabby cat' with probability=0.621104
-    class='n02123159 tiger cat' with probability=0.356378
+    class='n02123159 tiger cat' with probability=0.356377
     class='n02124075 Egyptian cat' with probability=0.019712
     class='n02129604 tiger, Panthera tigris' with probability=0.001215
     class='n04040759 radiator' with probability=0.000262
@@ -723,8 +722,8 @@ improvement in comparing the optimized model to the unoptimized model.
 
  .. code-block:: none
 
-    optimized: {'mean': 415.72972476999894, 'median': 415.55295190000834, 'std': 1.6521591847146238}
-    unoptimized: {'mean': 521.6796194400035, 'median': 522.0597105500076, 'std': 2.9126359836131974}
+    optimized: {'mean': 414.4166316200017, 'median': 413.3563886999923, 'std': 4.5748836455747}
+    unoptimized: {'mean': 507.22305444999967, 'median': 506.30012315000386, 'std': 5.391398088033922}
 
 
 
@@ -747,7 +746,7 @@ profiling/benchmarking.
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** ( 11 minutes  32.342 seconds)
+   **Total running time of the script:** ( 11 minutes  48.544 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 249a986fac..83d057d45e 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
 
-    1.337e-07 secs/op
+    1.27e-07 secs/op
 
 
 
diff --git a/docs/_sources/tutorial/intro_topi.rst.txt b/docs/_sources/tutorial/intro_topi.rst.txt
index baf1ae09a1..026f23c0ef 100644
--- a/docs/_sources/tutorial/intro_topi.rst.txt
+++ b/docs/_sources/tutorial/intro_topi.rst.txt
@@ -96,20 +96,23 @@ and to examine the IR code in human readable format, we can do
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto")}
-      buffer_map = {A_1: A} {
-      allocate(B: Pointer(global float32), float32, [n]), storage_scope = global;
-      for (i: int32, 0, n) {
-        B_1: Buffer(B, float32, [n], [])[i] = 0f32
-        for (k: int32, 0, m) {
-          B_1[i] = (B_1[i] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((i*stride) + (k*stride_1))])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            B = T.allocate([n], "float32", "global")
+            for i in range(n):
+                B_1 = T.buffer_decl((n,), data=B)
+                B_1[i] = T.float32(0)
+                for k in range(m):
+                    A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                    B_1[i] = B_1[i] + A_2[i * stride + k * stride_1]
 
 
 
@@ -137,20 +140,23 @@ Fortunately, we can replace those two lines with simple :code:`topi.sum` much li
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto")}
-      buffer_map = {A_1: A} {
-      allocate(A_red: Pointer(global float32), float32, [n]), storage_scope = global;
-      for (ax0: int32, 0, n) {
-        A_red_1: Buffer(A_red, float32, [n], [])[ax0] = 0f32
-        for (k1: int32, 0, m) {
-          A_red_1[ax0] = (A_red_1[ax0] + A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[((ax0*stride) + (k1*stride_1))])
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            m = T.var("int32")
+            stride = T.var("int32")
+            stride_1 = T.var("int32")
+            A_1 = T.match_buffer(A, (n, m), strides=(stride, stride_1), type="auto")
+            A_red = T.allocate([n], "float32", "global")
+            for ax0 in range(n):
+                A_red_1 = T.buffer_decl((n,), data=A_red)
+                A_red_1[ax0] = T.float32(0)
+                for k1 in range(m):
+                    A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                    A_red_1[ax0] = A_red_1[ax0] + A_2[ax0 * stride + k1 * stride_1]
 
 
 
@@ -216,30 +222,31 @@ we can schedule the following series of operations ending with :code:`topi.sum`
 
     /workspace/python/tvm/target/target.py:397: UserWarning: Try specifying cuda arch by adding 'arch=sm_xx' to your target.
       warnings.warn("Try specifying cuda arch by adding 'arch=sm_xx' to your target.")
-    @main = primfn(a_1: handle, b_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {a: Buffer(a_2: Pointer(float32), float32, [100, 10, 10], []),
-                 b: Buffer(b_2: Pointer(float32), float32, [10, 10], [])}
-      buffer_map = {a_1: a, b_1: b} {
-      allocate(T_divide_red: Pointer(global float32), float32, [1]), storage_scope = global;
-      attr [IterVar(threadIdx.x: int32, [0:1024], "ThreadIndex", "threadIdx.x")] "thread_extent" = 1024;
-      allocate(T_divide_red.rf: Pointer(local float32), float32, [1]), storage_scope = local;
-      allocate(reduce_temp0: Pointer(local float32), float32, [1]), storage_scope = local {
-        T_divide_red.rf_1: Buffer(T_divide_red.rf, float32, [1], [], scope="local", align=4)[0] = 0f32
-        for (k0.k1.fused.k2.fused.outer: int32, 0, 10) {
-          if @tir.likely((((((k0.k1.fused.k2.fused.outer*64) + floordiv(threadIdx.x, 16)) < 625) && (((k0.k1.fused.k2.fused.outer*64) + floordiv(threadIdx.x, 16)) < 625)) && (((k0.k1.fused.k2.fused.outer*64) + floordiv(threadIdx.x, 16)) < 625)), dtype=bool) {
-            T_divide_red.rf_1[0] = (T_divide_red.rf_1[0] + (((a_3: Buffer(a_2, float32, [10000], [])[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)] + b_3: Buffer(b_2, float32, [100], [])[((floordiv(floormod(((k0.k1.fused.k2.fused.outer*12) + floordiv(threadIdx.x, 2)), 50), 5)*10) + floormod(((k0.k1.fused.k2.fused.outer*4) + threadIdx.x), 10))]) + (a_3[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)]*b_3[((floordiv(floormod(((k0.k1.fused.k2.fused.outer*12) + floordiv(threadIdx.x, 2) [...]
-          }
-        }
-        attr [meta[tir.CommReducer][0]] "reduce_scope" = @tir.reinterpret(0u64, dtype=handle);
-        @tir.tvm_thread_allreduce(1u32, T_divide_red.rf_1[0], True, reduce_temp0_1: Buffer(reduce_temp0, float32, [1], [], scope="local")[0], threadIdx.x, dtype=handle)
-        if (threadIdx.x == 0) {
-          T_divide_red_1: Buffer(T_divide_red, float32, [1], [], align=4)[0] = reduce_temp0_1[0]
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(a: T.handle, b: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            a_1 = T.match_buffer(a, (100, 10, 10))
+            b_1 = T.match_buffer(b, (10, 10))
+            T_divide_red = T.allocate([1], "float32", "global")
+            threadIdx_x = T.env_thread("threadIdx.x")
+            T.launch_thread(threadIdx_x, 1024)
+            T_divide_red_rf = T.allocate([1], "float32", "local")
+            reduce_temp0 = T.allocate([1], "float32", "local")
+            T_divide_red_rf_1 = T.buffer_decl((1,), data=T_divide_red_rf, scope="local", align=4)
+            T_divide_red_rf_1[0] = T.float32(0)
+            for k0_k1_fused_k2_fused_outer in range(10):
+                if T.likely(k0_k1_fused_k2_fused_outer * 64 + threadIdx_x // 16 < 625 and k0_k1_fused_k2_fused_outer * 64 + threadIdx_x // 16 < 625 and k0_k1_fused_k2_fused_outer * 64 + threadIdx_x // 16 < 625):
+                    a_2 = T.buffer_decl((10000,), data=a_1.data)
+                    b_2 = T.buffer_decl((100,), data=b_1.data)
+                    T_divide_red_rf_1[0] = T_divide_red_rf_1[0] + (a_2[k0_k1_fused_k2_fused_outer * 1024 + threadIdx_x] + b_2[(k0_k1_fused_k2_fused_outer * 12 + threadIdx_x // 2) % 50 // 5 * 10 + (k0_k1_fused_k2_fused_outer * 4 + threadIdx_x) % 10] + a_2[k0_k1_fused_k2_fused_outer * 1024 + threadIdx_x] * b_2[(k0_k1_fused_k2_fused_outer * 12 + threadIdx_x // 2) % 50 // 5 * 10 + (k0_k1_fused_k2_fused_outer * 4 + threadIdx_x) % 10]) * T.float32(0.5)
+            reduce_temp0_1 = T.buffer_decl((1,), data=reduce_temp0, scope="local")
+            with T.attr(T.comm_reducer(lambda x, y: x + y, [T.float32(0)]), "reduce_scope", T.reinterpret("handle", T.uint64(0))):
+                T.tvm_thread_allreduce(T.uint32(1), T_divide_red_rf_1[0], True, reduce_temp0_1[0], threadIdx_x)
+            if threadIdx_x == 0:
+                T_divide_red_1 = T.buffer_decl((1,), data=T_divide_red, align=4)
+                T_divide_red_1[0] = reduce_temp0_1[0]
 
 
 
@@ -263,7 +270,7 @@ As you can see, scheduled stages of computation have been accumulated and we can
 
  .. code-block:: none
 
-    [stage(a, placeholder(a, 0x209d2fe0)), stage(b, placeholder(b, 0xcc73c60)), 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, 0x192f5ee0)), stage(b, placeholder(b, 0x14d5dbe0)), 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 [...]
 
 
 
@@ -318,44 +325,47 @@ TOPI also provides common neural nets operations such as _softmax_ with optimize
 
  .. code-block:: none
 
-    @main = primfn(tarray_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {tarray: Buffer(tarray_2: Pointer(float32), float32, [512, 512], [])}
-      buffer_map = {tarray_1: tarray} {
-      allocate(T_softmax_norm: Pointer(global float32x4), float32x4, [65536]), storage_scope = global;
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 512;
-      allocate(normal_reduce_temp0: Pointer(local float32), float32, [1]), storage_scope = local;
-      allocate(reduce_temp0: Pointer(local float32), float32, [1]), storage_scope = local;
-      allocate(T_softmax_exp: Pointer(warp float32), float32, [512]), storage_scope = warp;
-      allocate(normal_reduce_temp0_1: Pointer(local float32), float32, [1]), storage_scope = local;
-      allocate(reduce_temp0_1: Pointer(local float32), float32, [1]), storage_scope = local {
-        attr [IterVar(threadIdx.x: int32, [0:32], "ThreadIndex", "threadIdx.x")] "thread_extent" = 32 {
-          normal_reduce_temp0_2: Buffer(normal_reduce_temp0, float32, [1], [], scope="local")[0] = -3.40282e+38f32
-          for (k.inner: int32, 0, 16) {
-            normal_reduce_temp0_2[0] = max(normal_reduce_temp0_2[0], tarray_3: Buffer(tarray_2, float32, [262144], [])[(((blockIdx.x*512) + (threadIdx.x*16)) + k.inner)])
-          }
-          attr [meta[tir.CommReducer][0]] "reduce_scope" = @tir.reinterpret(0u64, dtype=handle);
-          @tir.tvm_thread_allreduce(1u32, normal_reduce_temp0_2[0], True, reduce_temp0_2: Buffer(reduce_temp0, float32, [1], [], scope="local")[0], threadIdx.x, dtype=handle)
-          for (i1.inner.outer: int32, 0, 4) {
-            let cse_var_1: int32 = (i1.inner.outer*4)
-            T_softmax_exp_1: Buffer(T_softmax_exp, float32, [512], [], scope="warp")[ramp(((threadIdx.x*16) + cse_var_1), 1, 4)] = @tir.exp((tarray_3[ramp((((blockIdx.x*512) + (threadIdx.x*16)) + cse_var_1), 1, 4)] - broadcast(reduce_temp0_3: Buffer(reduce_temp0, float32, [1], [], scope="local", align=4)[0], 4)), dtype=float32x4)
-          }
-        }
-        attr [IterVar(threadIdx.x, [0:32], "ThreadIndex", "threadIdx.x")] "thread_extent" = 32 {
-          normal_reduce_temp0_3: Buffer(normal_reduce_temp0_1, float32, [1], [], scope="local")[0] = 0f32
-          for (k.inner_1: int32, 0, 16) {
-            normal_reduce_temp0_3[0] = (normal_reduce_temp0_3[0] + T_softmax_exp_1[((threadIdx.x*16) + k.inner_1)])
-          }
-          attr [meta[tir.CommReducer][1]] "reduce_scope" = @tir.reinterpret(0u64, dtype=handle);
-          @tir.tvm_thread_allreduce(1u32, normal_reduce_temp0_3[0], True, reduce_temp0_4: Buffer(reduce_temp0_1, float32, [1], [], scope="local")[0], threadIdx.x, dtype=handle)
-          for (i1.inner.outer_1: int32, 0, 4) {
-            T_softmax_norm_1: Buffer(T_softmax_norm, float32x4, [65536], [])[(((blockIdx.x*128) + (threadIdx.x*4)) + i1.inner.outer_1)] = (T_softmax_exp_1[ramp(((threadIdx.x*16) + (i1.inner.outer_1*4)), 1, 4)] / broadcast(reduce_temp0_5: Buffer(reduce_temp0_1, float32, [1], [], scope="local", align=4)[0], 4))
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(tarray: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            tarray_1 = T.match_buffer(tarray, (512, 512))
+            T_softmax_norm = T.allocate([65536], "float32x4", "global")
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, 512)
+            normal_reduce_temp0 = T.allocate([1], "float32", "local")
+            reduce_temp0 = T.allocate([1], "float32", "local")
+            T_softmax_exp = T.allocate([512], "float32", "warp")
+            normal_reduce_temp0_1 = T.allocate([1], "float32", "local")
+            reduce_temp0_1 = T.allocate([1], "float32", "local")
+            threadIdx_x = T.env_thread("threadIdx.x")
+            T_softmax_exp_1 = T.buffer_decl((512,), data=T_softmax_exp, scope="warp")
+            with T.launch_thread(threadIdx_x, 32):
+                normal_reduce_temp0_2 = T.buffer_decl((1,), data=normal_reduce_temp0, scope="local")
+                normal_reduce_temp0_2[0] = T.float32(-3.4028234663852886e+38)
+                tarray_2 = T.buffer_decl((262144,), data=tarray_1.data)
+                for k_inner in range(16):
+                    normal_reduce_temp0_2[0] = T.max(normal_reduce_temp0_2[0], tarray_2[blockIdx_x * 512 + threadIdx_x * 16 + k_inner])
+                with T.attr(T.comm_reducer(lambda x, y: T.max(x, y), [T.float32(-3.4028234663852886e+38)]), "reduce_scope", T.reinterpret("handle", T.uint64(0))):
+                    reduce_temp0_2 = T.buffer_decl((1,), data=reduce_temp0, scope="local")
+                    T.tvm_thread_allreduce(T.uint32(1), normal_reduce_temp0_2[0], True, reduce_temp0_2[0], threadIdx_x)
+                for i1_inner_outer in range(4):
+                    cse_var_1: T.int32 = i1_inner_outer * 4
+                    reduce_temp0_2 = T.buffer_decl((1,), data=reduce_temp0, scope="local", align=4)
+                    T_softmax_exp_1[threadIdx_x * 16 + cse_var_1:threadIdx_x * 16 + cse_var_1 + 4] = T.exp(tarray_2[blockIdx_x * 512 + threadIdx_x * 16 + cse_var_1:blockIdx_x * 512 + threadIdx_x * 16 + cse_var_1 + 4] - T.Broadcast(reduce_temp0_2[0], 4))
+            T.launch_thread(threadIdx_x, 32)
+            normal_reduce_temp0_2 = T.buffer_decl((1,), data=normal_reduce_temp0_1, scope="local")
+            normal_reduce_temp0_2[0] = T.float32(0)
+            for k_inner in range(16):
+                normal_reduce_temp0_2[0] = normal_reduce_temp0_2[0] + T_softmax_exp_1[threadIdx_x * 16 + k_inner]
+            with T.attr(T.comm_reducer(lambda x, y: x + y, [T.float32(0)]), "reduce_scope", T.reinterpret("handle", T.uint64(0))):
+                reduce_temp0_2 = T.buffer_decl((1,), data=reduce_temp0_1, scope="local")
+                T.tvm_thread_allreduce(T.uint32(1), normal_reduce_temp0_2[0], True, reduce_temp0_2[0], threadIdx_x)
+            for i1_inner_outer in range(4):
+                T_softmax_norm_1 = T.buffer_decl((65536,), "float32x4", data=T_softmax_norm)
+                reduce_temp0_2 = T.buffer_decl((1,), data=reduce_temp0_1, scope="local", align=4)
+                T_softmax_norm_1[blockIdx_x * 128 + threadIdx_x * 4 + i1_inner_outer] = T_softmax_exp_1[threadIdx_x * 16 + i1_inner_outer * 4:threadIdx_x * 16 + i1_inner_outer * 4 + 4] / T.Broadcast(reduce_temp0_2[0], 4)
 
 
 
@@ -396,212 +406,215 @@ We can fuse :code:`topi.nn.conv2d` and :code:`topi.nn.relu` together.
 
  .. code-block:: none
 
-    @main = primfn(placeholder_2: handle, placeholder_3: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {placeholder: Buffer(placeholder_4: Pointer(float32), float32, [1, 3, 224, 224], []),
-                 placeholder_1: Buffer(placeholder_5: Pointer(float32), float32, [10, 3, 5, 5], [])}
-      buffer_map = {placeholder_2: placeholder, placeholder_3: placeholder_1} {
-      allocate(compute: Pointer(global float32), float32, [501760]), storage_scope = global;
-      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 5;
-      allocate(conv2d_nchw: Pointer(local float32), float32, [14]), storage_scope = local;
-      allocate(pad_temp.shared: Pointer(shared float32), float32, [112]), storage_scope = shared;
-      allocate(placeholder.shared: Pointer(shared float32), float32, [2]), storage_scope = shared;
-      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 224;
-      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
-      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-        conv2d_nchw_1: Buffer(conv2d_nchw, float32, [4], [], scope="local", align=8)[0] = 0f32
-        conv2d_nchw_1[2] = 0f32
-        conv2d_nchw_1[4] = 0f32
-        conv2d_nchw_1[6] = 0f32
-        conv2d_nchw_1[8] = 0f32
-        conv2d_nchw_1[10] = 0f32
-        conv2d_nchw_1[12] = 0f32
-        conv2d_nchw_1[1] = 0f32
-        conv2d_nchw_1[3] = 0f32
-        conv2d_nchw_1[5] = 0f32
-        conv2d_nchw_1[7] = 0f32
-        conv2d_nchw_1[9] = 0f32
-        conv2d_nchw_1[11] = 0f32
-        conv2d_nchw_1[13] = 0f32
-        for (rc.outer: int32, 0, 3) {
-          for (ry.outer: int32, 0, 5) {
-            attr [IterVar(threadIdx.z_1: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_1: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_1: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-              pad_temp.shared_1: Buffer(pad_temp.shared, float32, [112], [], scope="shared")[(threadIdx.x_1*7)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*56) + floordiv((threadIdx.x_1*7), 2)))), placeholder_6: Buffer(placeholder_4, float32, [150528], [])[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 450)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 1)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*56) + floordiv(((threadIdx.x_1*7) + 1), 2)))), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 449)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 2)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 3)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 4)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 5)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 6)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)], 0f32, dtype=float32)
-            }
-            attr [IterVar(threadIdx.z_2: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_2: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_2: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
-            if @tir.likely((threadIdx.x_2 < 2), dtype=bool) {
-              placeholder.shared_1: Buffer(placeholder.shared, float32, [2], [], scope="shared", align=8)[threadIdx.x_2] = placeholder_7: Buffer(placeholder_5, float32, [750], [])[((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5))]
-            }
-            conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[0]))
-            conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[8] = (conv2d_nchw_1[8] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[10] = (conv2d_nchw_1[10] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[12] = (conv2d_nchw_1[12] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[1]))
-            conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[7] = (conv2d_nchw_1[7] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[9] = (conv2d_nchw_1[9] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[11] = (conv2d_nchw_1[11] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[13] = (conv2d_nchw_1[13] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[1]))
-            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-              pad_temp.shared_1[(threadIdx.x_1*7)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*56) + floordiv(((threadIdx.x_1*7) + 1), 2)))), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 449)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 1)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 2)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 3)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 4)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 5)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 6)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)], 0f32, dtype=float32)
-            }
-            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
-            if @tir.likely((threadIdx.x_2 < 2), dtype=bool) {
-              placeholder.shared_1[threadIdx.x_2] = placeholder_7[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 1)]
-            }
-            conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[0]))
-            conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[8] = (conv2d_nchw_1[8] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[10] = (conv2d_nchw_1[10] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[12] = (conv2d_nchw_1[12] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[1]))
-            conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[7] = (conv2d_nchw_1[7] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[9] = (conv2d_nchw_1[9] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[11] = (conv2d_nchw_1[11] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[13] = (conv2d_nchw_1[13] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[1]))
-            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-              pad_temp.shared_1[(threadIdx.x_1*7)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 1)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 2)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 3)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 4)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 5)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 6)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)], 0f32, dtype=float32)
-            }
-            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
-            if @tir.likely((threadIdx.x_2 < 2), dtype=bool) {
-              placeholder.shared_1[threadIdx.x_2] = placeholder_7[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 2)]
-            }
-            conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[0]))
-            conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[8] = (conv2d_nchw_1[8] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[10] = (conv2d_nchw_1[10] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[12] = (conv2d_nchw_1[12] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[1]))
-            conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[7] = (conv2d_nchw_1[7] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[9] = (conv2d_nchw_1[9] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[11] = (conv2d_nchw_1[11] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[13] = (conv2d_nchw_1[13] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[1]))
-            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-              pad_temp.shared_1[(threadIdx.x_1*7)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 1)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 2)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 3)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 4)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 5)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 6)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*56) + floordiv(((threadIdx.x_1*7) + 9), 2)) < 113)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 441)], 0f32, dtype=float32)
-            }
-            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
-            if @tir.likely((threadIdx.x_2 < 2), dtype=bool) {
-              placeholder.shared_1[threadIdx.x_2] = placeholder_7[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 3)]
-            }
-            conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[0]))
-            conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[8] = (conv2d_nchw_1[8] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[10] = (conv2d_nchw_1[10] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[12] = (conv2d_nchw_1[12] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[1]))
-            conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[7] = (conv2d_nchw_1[7] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[9] = (conv2d_nchw_1[9] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[11] = (conv2d_nchw_1[11] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[13] = (conv2d_nchw_1[13] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[1]))
-            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
-              pad_temp.shared_1[(threadIdx.x_1*7)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 1)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 2)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 3)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 4)] = @tir.if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 5)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*56) + floordiv(((threadIdx.x_1*7) + 9), 2)) < 113)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 441)], 0f32, dtype=float32)
-              pad_temp.shared_1[((threadIdx.x_1*7) + 6)] = @tir.if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*56) + floordiv((threadIdx.x_1*7), 2)) < 108)), placeholder_6[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 440)], 0f32, dtype=float32)
-            }
-            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
-            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
-            if @tir.likely((threadIdx.x_2 < 2), dtype=bool) {
-              placeholder.shared_1[threadIdx.x_2] = placeholder_7[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 4)]
-            }
-            conv2d_nchw_1[0] = (conv2d_nchw_1[0] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[0]))
-            conv2d_nchw_1[2] = (conv2d_nchw_1[2] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[4] = (conv2d_nchw_1[4] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[6] = (conv2d_nchw_1[6] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[8] = (conv2d_nchw_1[8] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[10] = (conv2d_nchw_1[10] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[12] = (conv2d_nchw_1[12] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[0]))
-            conv2d_nchw_1[1] = (conv2d_nchw_1[1] + (pad_temp.shared_1[threadIdx.x]*placeholder.shared_1[1]))
-            conv2d_nchw_1[3] = (conv2d_nchw_1[3] + (pad_temp.shared_1[(threadIdx.x + 16)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[5] = (conv2d_nchw_1[5] + (pad_temp.shared_1[(threadIdx.x + 32)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[7] = (conv2d_nchw_1[7] + (pad_temp.shared_1[(threadIdx.x + 48)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[9] = (conv2d_nchw_1[9] + (pad_temp.shared_1[(threadIdx.x + 64)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[11] = (conv2d_nchw_1[11] + (pad_temp.shared_1[(threadIdx.x + 80)]*placeholder.shared_1[1]))
-            conv2d_nchw_1[13] = (conv2d_nchw_1[13] + (pad_temp.shared_1[(threadIdx.x + 96)]*placeholder.shared_1[1]))
-          }
-        }
-        compute_1: Buffer(compute, float32, [501760], [])[((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x)] = max(conv2d_nchw_1[0], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 16)] = max(conv2d_nchw_1[2], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 32)] = max(conv2d_nchw_1[4], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 48)] = max(conv2d_nchw_1[6], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 64)] = max(conv2d_nchw_1[8], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 80)] = max(conv2d_nchw_1[10], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 96)] = max(conv2d_nchw_1[12], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50176)] = max(conv2d_nchw_1[1], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50192)] = max(conv2d_nchw_1[3], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50208)] = max(conv2d_nchw_1[5], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50224)] = max(conv2d_nchw_1[7], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50240)] = max(conv2d_nchw_1[9], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50256)] = max(conv2d_nchw_1[11], 0f32)
-        compute_1[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50272)] = max(conv2d_nchw_1[13], 0f32)
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(placeholder: T.handle, placeholder_1: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            placeholder_2 = T.match_buffer(placeholder, (1, 3, 224, 224))
+            placeholder_3 = T.match_buffer(placeholder_1, (10, 3, 5, 5))
+            compute = T.allocate([501760], "float32", "global")
+            blockIdx_z = T.env_thread("blockIdx.z")
+            T.launch_thread(blockIdx_z, 5)
+            conv2d_nchw = T.allocate([14], "float32", "local")
+            pad_temp_shared = T.allocate([112], "float32", "shared")
+            placeholder_shared = T.allocate([2], "float32", "shared")
+            blockIdx_y = T.env_thread("blockIdx.y")
+            T.launch_thread(blockIdx_y, 224)
+            blockIdx_x = T.env_thread("blockIdx.x")
+            T.launch_thread(blockIdx_x, 2)
+            threadIdx_z = T.env_thread("threadIdx.z")
+            T.launch_thread(threadIdx_z, 1)
+            threadIdx_y = T.env_thread("threadIdx.y")
+            T.launch_thread(threadIdx_y, 1)
+            threadIdx_x = T.env_thread("threadIdx.x")
+            T.launch_thread(threadIdx_x, 16)
+            conv2d_nchw_1 = T.buffer_decl((4,), data=conv2d_nchw, scope="local", align=8)
+            conv2d_nchw_1[0] = T.float32(0)
+            conv2d_nchw_1[2] = T.float32(0)
+            conv2d_nchw_1[4] = T.float32(0)
+            conv2d_nchw_1[6] = T.float32(0)
+            conv2d_nchw_1[8] = T.float32(0)
+            conv2d_nchw_1[10] = T.float32(0)
+            conv2d_nchw_1[12] = T.float32(0)
+            conv2d_nchw_1[1] = T.float32(0)
+            conv2d_nchw_1[3] = T.float32(0)
+            conv2d_nchw_1[5] = T.float32(0)
+            conv2d_nchw_1[7] = T.float32(0)
+            conv2d_nchw_1[9] = T.float32(0)
+            conv2d_nchw_1[11] = T.float32(0)
+            conv2d_nchw_1[13] = T.float32(0)
+            for rc_outer, ry_outer in T.grid(3, 5):
+                threadIdx_z_1 = T.env_thread("threadIdx.z")
+                threadIdx_y_1 = T.env_thread("threadIdx.y")
+                threadIdx_x_1 = T.env_thread("threadIdx.x")
+                pad_temp_shared_1 = T.buffer_decl((112,), data=pad_temp_shared, scope="shared")
+                placeholder_4 = T.buffer_decl((150528,), data=placeholder_2.data)
+                with T.launch_thread(threadIdx_z_1, 1):
+                    T.launch_thread(threadIdx_y_1, 1)
+                    T.launch_thread(threadIdx_x_1, 16)
+                    pad_temp_shared_1[threadIdx_x_1 * 7] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and 1 <= blockIdx_x * 56 + threadIdx_x_1 * 7 // 2, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 450], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 1] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and 1 <= blockIdx_x * 56 + (threadIdx_x_1 * 7 + 1) // 2, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 449], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 2] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 448], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 3] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 447], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 4] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 446], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 5] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 445], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 6] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 444], T.float32(0))
+                threadIdx_z_2 = T.env_thread("threadIdx.z")
+                threadIdx_y_2 = T.env_thread("threadIdx.y")
+                threadIdx_x_2 = T.env_thread("threadIdx.x")
+                placeholder_shared_1 = T.buffer_decl((2,), data=placeholder_shared, scope="shared", align=8)
+                placeholder_5 = T.buffer_decl((750,), data=placeholder_3.data)
+                with T.launch_thread(threadIdx_z_2, 1):
+                    T.launch_thread(threadIdx_y_2, 1)
+                    T.launch_thread(threadIdx_x_2, 16)
+                    if T.likely(threadIdx_x_2 < 2):
+                        placeholder_shared_1[threadIdx_x_2] = placeholder_5[blockIdx_z * 150 + threadIdx_x_2 * 75 + rc_outer * 25 + ry_outer * 5]
+                conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[0]
+                conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[0]
+                conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[0]
+                conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[0]
+                conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[0]
+                conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[0]
+                conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[0]
+                conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[1]
+                conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[1]
+                conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[1]
+                conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[1]
+                conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[1]
+                conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[1]
+                conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[1]
+                with T.launch_thread(threadIdx_z_1, 1):
+                    T.launch_thread(threadIdx_y_1, 1)
+                    T.launch_thread(threadIdx_x_1, 16)
+                    pad_temp_shared_1[threadIdx_x_1 * 7] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and 1 <= blockIdx_x * 56 + (threadIdx_x_1 * 7 + 1) // 2, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 449], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 1] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 448], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 2] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 447], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 3] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 446], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 4] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 445], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 5] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 444], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 6] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 443], T.float32(0))
+                with T.launch_thread(threadIdx_z_2, 1):
+                    T.launch_thread(threadIdx_y_2, 1)
+                    T.launch_thread(threadIdx_x_2, 16)
+                    if T.likely(threadIdx_x_2 < 2):
+                        placeholder_shared_1[threadIdx_x_2] = placeholder_5[blockIdx_z * 150 + threadIdx_x_2 * 75 + rc_outer * 25 + ry_outer * 5 + 1]
+                conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[0]
+                conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[0]
+                conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[0]
+                conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[0]
+                conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[0]
+                conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[0]
+                conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[0]
+                conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[1]
+                conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[1]
+                conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[1]
+                conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[1]
+                conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[1]
+                conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[1]
+                conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[1]
+                with T.launch_thread(threadIdx_z_1, 1):
+                    T.launch_thread(threadIdx_y_1, 1)
+                    T.launch_thread(threadIdx_x_1, 16)
+                    pad_temp_shared_1[threadIdx_x_1 * 7] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 448], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 1] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 447], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 2] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 446], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 3] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 445], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 4] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 444], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 5] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 443], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 6] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 442], T.float32(0))
+                with T.launch_thread(threadIdx_z_2, 1):
+                    T.launch_thread(threadIdx_y_2, 1)
+                    T.launch_thread(threadIdx_x_2, 16)
+                    if T.likely(threadIdx_x_2 < 2):
+                        placeholder_shared_1[threadIdx_x_2] = placeholder_5[blockIdx_z * 150 + threadIdx_x_2 * 75 + rc_outer * 25 + ry_outer * 5 + 2]
+                conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[0]
+                conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[0]
+                conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[0]
+                conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[0]
+                conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[0]
+                conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[0]
+                conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[0]
+                conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[1]
+                conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[1]
+                conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[1]
+                conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[1]
+                conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[1]
+                conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[1]
+                conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[1]
+                with T.launch_thread(threadIdx_z_1, 1):
+                    T.launch_thread(threadIdx_y_1, 1)
+                    T.launch_thread(threadIdx_x_1, 16)
+                    pad_temp_shared_1[threadIdx_x_1 * 7] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 447], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 1] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 446], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 2] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 445], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 3] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 444], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 4] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 443], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 5] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 442], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 6] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and blockIdx_x * 56 + (threadIdx_x_1 * 7 + 9) // 2 < 113, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 441], T.float32(0))
+                with T.launch_thread(threadIdx_z_2, 1):
+                    T.launch_thread(threadIdx_y_2, 1)
+                    T.launch_thread(threadIdx_x_2, 16)
+                    if T.likely(threadIdx_x_2 < 2):
+                        placeholder_shared_1[threadIdx_x_2] = placeholder_5[blockIdx_z * 150 + threadIdx_x_2 * 75 + rc_outer * 25 + ry_outer * 5 + 3]
+                conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[0]
+                conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[0]
+                conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[0]
+                conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[0]
+                conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[0]
+                conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[0]
+                conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[0]
+                conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[1]
+                conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[1]
+                conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[1]
+                conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[1]
+                conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[1]
+                conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[1]
+                conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[1]
+                with T.launch_thread(threadIdx_z_1, 1):
+                    T.launch_thread(threadIdx_y_1, 1)
+                    T.launch_thread(threadIdx_x_1, 16)
+                    pad_temp_shared_1[threadIdx_x_1 * 7] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 446], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 1] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 445], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 2] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 444], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 3] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 443], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 4] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 442], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 5] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and blockIdx_x * 56 + (threadIdx_x_1 * 7 + 9) // 2 < 113, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 441], T.float32(0))
+                    pad_temp_shared_1[threadIdx_x_1 * 7 + 6] = T.if_then_else(2 <= blockIdx_y + ry_outer and blockIdx_y + ry_outer < 226 and blockIdx_x * 56 + threadIdx_x_1 * 7 // 2 < 108, placeholder_4[rc_outer * 50176 + blockIdx_y * 224 + ry_outer * 224 + blockIdx_x * 112 + threadIdx_x_1 * 7 - 440], T.float32(0))
+                with T.launch_thread(threadIdx_z_2, 1):
+                    T.launch_thread(threadIdx_y_2, 1)
+                    T.launch_thread(threadIdx_x_2, 16)
+                    if T.likely(threadIdx_x_2 < 2):
+                        placeholder_shared_1[threadIdx_x_2] = placeholder_5[blockIdx_z * 150 + threadIdx_x_2 * 75 + rc_outer * 25 + ry_outer * 5 + 4]
+                conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[0]
+                conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[0]
+                conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[0]
+                conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[0]
+                conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[0]
+                conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[0]
+                conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[0]
+                conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[threadIdx_x] * placeholder_shared_1[1]
+                conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[threadIdx_x + 16] * placeholder_shared_1[1]
+                conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[threadIdx_x + 32] * placeholder_shared_1[1]
+                conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[threadIdx_x + 48] * placeholder_shared_1[1]
+                conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[threadIdx_x + 64] * placeholder_shared_1[1]
+                conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[threadIdx_x + 80] * placeholder_shared_1[1]
+                conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[threadIdx_x + 96] * placeholder_shared_1[1]
+            compute_1 = T.buffer_decl((501760,), data=compute)
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x] = T.max(conv2d_nchw_1[0], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 16] = T.max(conv2d_nchw_1[2], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 32] = T.max(conv2d_nchw_1[4], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 48] = T.max(conv2d_nchw_1[6], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 64] = T.max(conv2d_nchw_1[8], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 80] = T.max(conv2d_nchw_1[10], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 96] = T.max(conv2d_nchw_1[12], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50176] = T.max(conv2d_nchw_1[1], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50192] = T.max(conv2d_nchw_1[3], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50208] = T.max(conv2d_nchw_1[5], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50224] = T.max(conv2d_nchw_1[7], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50240] = T.max(conv2d_nchw_1[9], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50256] = T.max(conv2d_nchw_1[11], T.float32(0))
+            compute_1[blockIdx_z * 100352 + blockIdx_y * 224 + blockIdx_x * 112 + threadIdx_x + 50272] = T.max(conv2d_nchw_1[13], T.float32(0))
 
 
 
diff --git a/docs/_sources/tutorial/sg_execution_times.rst.txt b/docs/_sources/tutorial/sg_execution_times.rst.txt
index deeeef33e6..3d2846468d 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
 =================
-**15:00.963** total execution time for **tutorial** files:
+**15:03.977** total execution time for **tutorial** files:
 
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_autotvm_relay_x86.py` (``autotvm_relay_x86.py``)                 | 11:32.342 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_autotvm_relay_x86.py` (``autotvm_relay_x86.py``)                 | 11:48.544 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_auto_scheduler_matmul_x86.py` (``auto_scheduler_matmul_x86.py``) | 01:28.083 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_auto_scheduler_matmul_x86.py` (``auto_scheduler_matmul_x86.py``) | 01:13.404 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_tensor_expr_get_started.py` (``tensor_expr_get_started.py``)     | 00:59.012 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_tensor_expr_get_started.py` (``tensor_expr_get_started.py``)     | 00:58.876 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_relay_quick_start.py` (``relay_quick_start.py``)                 | 00:35.118 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_relay_quick_start.py` (``relay_quick_start.py``)                 | 00:35.284 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_autotvm_matmul_x86.py` (``autotvm_matmul_x86.py``)               | 00:24.066 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_autotvm_matmul_x86.py` (``autotvm_matmul_x86.py``)               | 00:25.613 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_tensor_ir_blitz_course.py` (``tensor_ir_blitz_course.py``)       | 00:01.325 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_tensor_ir_blitz_course.py` (``tensor_ir_blitz_course.py``)       | 00:01.257 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_intro_topi.py` (``intro_topi.py``)                               | 00:00.842 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_intro_topi.py` (``intro_topi.py``)                               | 00:00.833 | 0.0 MB |
 +------------------------------------------------------------------------------------------+-----------+--------+
-| :ref:`sphx_glr_tutorial_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``) | 00:00.174 | 0.0 MB |
+| :ref:`sphx_glr_tutorial_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``) | 00:00.166 | 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_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_tvmc_command_line_driver.py` (``tvmc_command_line_driver.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 81bf5cf3c5..bf63322802 100644
--- a/docs/_sources/tutorial/tensor_expr_get_started.rst.txt
+++ b/docs/_sources/tutorial/tensor_expr_get_started.rst.txt
@@ -343,18 +343,23 @@ return a readable C-style statement.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i: int32, 0, n) "parallel" {
-        C_3: Buffer(C_2, float32, [(stride_2*n)], [], type="auto")[(i*stride_2)] = (A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[(i*stride)] + B_3: Buffer(B_2, float32, [(stride_1*n)], [], type="auto")[(i*stride_1)])
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (n,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (n,), strides=(stride_2,), type="auto")
+            for i in T.parallel(n):
+                C_2 = T.buffer_decl((stride_2 * n,), data=C_1.data, type="auto")
+                A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                B_2 = T.buffer_decl((stride_1 * n,), data=B_1.data, type="auto")
+                C_2[i * stride_2] = A_2[i * stride] + B_2[i * stride_1]
 
 
 
@@ -439,24 +444,27 @@ factor to be the number of threads on your CPU.
 
  .. code-block:: none
 
-    vector: 0.000025
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [n: int32], [stride: int32], type="auto"),
-                 B: Buffer(B_2: Pointer(float32), float32, [n], [stride_1: int32], type="auto"),
-                 C: Buffer(C_2: Pointer(float32), float32, [n], [stride_2: int32], type="auto")}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (i.outer: int32, 0, floordiv((n + 3), 4)) "parallel" {
-        for (i.inner.s: int32, 0, 4) {
-          if @tir.likely((((i.outer*4) + i.inner.s) < n), dtype=bool) {
-            let cse_var_1: int32 = ((i.outer*4) + i.inner.s)
-            C_3: Buffer(C_2, float32, [(stride_2*n)], [], type="auto")[(cse_var_1*stride_2)] = (A_3: Buffer(A_2, float32, [(stride*n)], [], type="auto")[(cse_var_1*stride)] + B_3: Buffer(B_2, float32, [(stride_1*n)], [], type="auto")[(cse_var_1*stride_1)])
-          }
-        }
-      }
-    }
-
-
+    vector: 0.000026
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            n = T.var("int32")
+            stride = T.var("int32")
+            A_1 = T.match_buffer(A, (n,), strides=(stride,), type="auto")
+            stride_1 = T.var("int32")
+            B_1 = T.match_buffer(B, (n,), strides=(stride_1,), type="auto")
+            stride_2 = T.var("int32")
+            C_1 = T.match_buffer(C, (n,), strides=(stride_2,), type="auto")
+            for i_outer in T.parallel((n + 3) // 4):
+                for i_inner_s in range(4):
+                    if T.likely(i_outer * 4 + i_inner_s < n):
+                        C_2 = T.buffer_decl((stride_2 * n,), data=C_1.data, type="auto")
+                        A_2 = T.buffer_decl((stride * n,), data=A_1.data, type="auto")
+                        B_2 = T.buffer_decl((stride_1 * n,), data=B_1.data, type="auto")
+                        cse_var_1: T.int32 = i_outer * 4 + i_inner_s
+                        C_2[cse_var_1 * stride_2] = A_2[cse_var_1 * stride] + B_2[cse_var_1 * stride_1]
 
 
 
@@ -490,10 +498,10 @@ We can now compare the different schedules
  .. code-block:: none
 
                 Operator                  Timing             Performance
-                   numpy    6.782519999433135e-06                    1.0
-                   naive              6.6863e-06      0.9858135325157645
-                parallel    7.101999999999999e-06       1.04710343656835
-                  vector             2.47586e-05      3.6503541459618636
+                   numpy    6.613810000999365e-06                    1.0
+                   naive    6.733000000000001e-06     1.0180213823775743
+                parallel    6.7131999999999995e-06    1.0150276465434622
+                  vector    2.5874599999999998e-05     3.912207940066357
 
 
 
@@ -914,7 +922,7 @@ matrix multiplication.
 
  .. code-block:: none
 
-    Numpy running time: 0.018397
+    Numpy running time: 0.017937
 
 
 
@@ -972,7 +980,7 @@ optimizations.
 
  .. code-block:: none
 
-    none: 3.255350
+    none: 3.260486
 
 
 
@@ -999,25 +1007,23 @@ nested loops over the indices of the A and B matrices.
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (x: int32, 0, 1024) {
-        for (y: int32, 0, 1024) {
-          C_3: Buffer(C_2, float32, [1048576], [])[((x*1024) + y)] = 0f32
-          for (k: int32, 0, 1024) {
-            let cse_var_2: int32 = (x*1024)
-            let cse_var_1: int32 = (cse_var_2 + y)
-            C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [1048576], [])[(cse_var_2 + k)]*B_3: Buffer(B_2, float32, [1048576], [])[((k*1024) + y)]))
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for x, y in T.grid(1024, 1024):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                C_2[x * 1024 + y] = T.float32(0)
+                for k in range(1024):
+                    cse_var_2: T.int32 = x * 1024
+                    cse_var_1: T.int32 = cse_var_2 + y
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[cse_var_2 + k] * B_2[k * 1024 + y]
 
 
 
@@ -1074,7 +1080,7 @@ schedule.
 
  .. code-block:: none
 
-    blocking: 0.300131
+    blocking: 0.297954
 
 
 
@@ -1100,36 +1106,25 @@ internal representation and compare it to the original:
 
  .. code-block:: none
 
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (x.outer: int32, 0, 32) {
-        for (y.outer: int32, 0, 32) {
-          for (x.inner.init: int32, 0, 32) {
-            for (y.inner.init: int32, 0, 32) {
-              C_3: Buffer(C_2, float32, [1048576], [])[((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)) + y.inner.init)] = 0f32
-            }
-          }
-          for (k.outer: int32, 0, 256) {
-            for (k.inner: int32, 0, 4) {
-              for (x.inner: int32, 0, 32) {
-                for (y.inner: int32, 0, 32) {
-                  let cse_var_3: int32 = (y.outer*32)
-                  let cse_var_2: int32 = ((x.outer*32768) + (x.inner*1024))
-                  let cse_var_1: int32 = ((cse_var_2 + cse_var_3) + y.inner)
-                  C_3[cse_var_1] = (C_3[cse_var_1] + (A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)]*B_3: Buffer(B_2, float32, [1048576], [])[((((k.outer*4096) + (k.inner*1024)) + cse_var_3) + y.inner)]))
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for x_outer, y_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for x_inner_init, y_inner_init in T.grid(32, 32):
+                    C_2[x_outer * 32768 + x_inner_init * 1024 + y_outer * 32 + y_inner_init] = T.float32(0)
+                for k_outer, k_inner, x_inner, y_inner in T.grid(256, 4, 32, 32):
+                    cse_var_3: T.int32 = y_outer * 32
+                    cse_var_2: T.int32 = x_outer * 32768 + x_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3 + y_inner
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1] = C_2[cse_var_1] + A_2[cse_var_2 + k_outer * 4 + k_inner] * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3 + y_inner]
 
 
 
@@ -1169,33 +1164,26 @@ already cache friendly from our previous optimizations.
 
  .. code-block:: none
 
-    vectorization: 0.334208
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (x.outer: int32, 0, 32) {
-        for (y.outer: int32, 0, 32) {
-          for (x.inner.init: int32, 0, 32) {
-            C_3: Buffer(C_2, float32, [1048576], [])[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
-          }
-          for (k.outer: int32, 0, 256) {
-            for (k.inner: int32, 0, 4) {
-              for (x.inner: int32, 0, 32) {
-                let cse_var_3: int32 = (y.outer*32)
-                let cse_var_2: int32 = ((x.outer*32768) + (x.inner*1024))
-                let cse_var_1: int32 = (cse_var_2 + cse_var_3)
-                C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)], 32)*B_3: Buffer(B_2, float32, [1048576], [])[ramp((((k.outer*4096) + (k.inner*1024)) + cse_var_3), 1, 32)]))
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    vectorization: 0.329291
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for x_outer, y_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for x_inner_init in range(32):
+                    C_2[x_outer * 32768 + x_inner_init * 1024 + y_outer * 32:x_outer * 32768 + x_inner_init * 1024 + y_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, k_inner, x_inner in T.grid(256, 4, 32):
+                    cse_var_3: T.int32 = y_outer * 32
+                    cse_var_2: T.int32 = x_outer * 32768 + x_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_2 + k_outer * 4 + k_inner], 32) * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3:k_outer * 4096 + k_inner * 1024 + cse_var_3 + 32]
 
 
 
@@ -1242,33 +1230,26 @@ more cache friendly.
 
  .. code-block:: none
 
-    loop permutation: 0.119802
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      for (x.outer: int32, 0, 32) {
-        for (y.outer: int32, 0, 32) {
-          for (x.inner.init: int32, 0, 32) {
-            C_3: Buffer(C_2, float32, [1048576], [])[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
-          }
-          for (k.outer: int32, 0, 256) {
-            for (x.inner: int32, 0, 32) {
-              for (k.inner: int32, 0, 4) {
-                let cse_var_3: int32 = (y.outer*32)
-                let cse_var_2: int32 = ((x.outer*32768) + (x.inner*1024))
-                let cse_var_1: int32 = (cse_var_2 + cse_var_3)
-                C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_2 + (k.outer*4)) + k.inner)], 32)*B_3: Buffer(B_2, float32, [1048576], [])[ramp((((k.outer*4096) + (k.inner*1024)) + cse_var_3), 1, 32)]))
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    loop permutation: 0.116738
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            for x_outer, y_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for x_inner_init in range(32):
+                    C_2[x_outer * 32768 + x_inner_init * 1024 + y_outer * 32:x_outer * 32768 + x_inner_init * 1024 + y_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, x_inner, k_inner in T.grid(256, 32, 4):
+                    cse_var_3: T.int32 = y_outer * 32
+                    cse_var_2: T.int32 = x_outer * 32768 + x_inner * 1024
+                    cse_var_1: T.int32 = cse_var_2 + cse_var_3
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_2 + k_outer * 4 + k_inner], 32) * B_2[k_outer * 4096 + k_inner * 1024 + cse_var_3:k_outer * 4096 + k_inner * 1024 + cse_var_3 + 32]
 
 
 
@@ -1340,40 +1321,31 @@ optimized schedule.
 
  .. code-block:: none
 
-    array packing: 0.111110
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global {
-        for (x: int32, 0, 32) "parallel" {
-          for (y: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((x*1024) + y)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((y*1024) + (x*32)), 1, 32)]
-          }
-        }
-        for (x.outer: int32, 0, 32) {
-          for (y.outer: int32, 0, 32) {
-            for (x.inner.init: int32, 0, 32) {
-              C_3: Buffer(C_2, float32, [1048576], [])[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (x.inner: int32, 0, 32) {
-                for (k.inner: int32, 0, 4) {
-                  let cse_var_3: int32 = ((x.outer*32768) + (x.inner*1024))
-                  let cse_var_2: int32 = (k.outer*4)
-                  let cse_var_1: int32 = (cse_var_3 + (y.outer*32))
-                  C_3[ramp(cse_var_1, 1, 32)] = (C_3[ramp(cse_var_1, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[((cse_var_3 + cse_var_2) + k.inner)], 32)*packedB_1[(((y.outer*1024) + cse_var_2) + k.inner)]))
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    array packing: 0.108993
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for x in T.parallel(32):
+                for y in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[x * 1024 + y] = B_2[y * 1024 + x * 32:y * 1024 + x * 32 + 32]
+            for x_outer, y_outer in T.grid(32, 32):
+                C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                for x_inner_init in range(32):
+                    C_2[x_outer * 32768 + x_inner_init * 1024 + y_outer * 32:x_outer * 32768 + x_inner_init * 1024 + y_outer * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, x_inner, k_inner in T.grid(256, 32, 4):
+                    cse_var_3: T.int32 = x_outer * 32768 + x_inner * 1024
+                    cse_var_2: T.int32 = k_outer * 4
+                    cse_var_1: T.int32 = cse_var_3 + y_outer * 32
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    C_2[cse_var_1:cse_var_1 + 32] = C_2[cse_var_1:cse_var_1 + 32] + T.Broadcast(A_2[cse_var_3 + cse_var_2 + k_inner], 32) * packedB_1[y_outer * 1024 + cse_var_2 + k_inner]
 
 
 
@@ -1432,50 +1404,39 @@ to `C` when all the block results are ready.
 
  .. code-block:: none
 
-    block caching: 0.111965
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global;
-      allocate(C.global: Pointer(global float32), float32, [1024]), storage_scope = global {
-        for (x: int32, 0, 32) "parallel" {
-          for (y: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((x*1024) + y)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((y*1024) + (x*32)), 1, 32)]
-          }
-        }
-        for (x.outer: int32, 0, 32) {
-          for (y.outer: int32, 0, 32) {
-            for (x.c.init: int32, 0, 32) {
-              C.global_1: Buffer(C.global, float32, [1024], [])[ramp((x.c.init*32), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (x.c: int32, 0, 32) {
-                let cse_var_4: int32 = (k.outer*4)
-                let cse_var_3: int32 = (x.c*32)
-                let cse_var_2: int32 = ((y.outer*1024) + cse_var_4)
-                let cse_var_1: int32 = (((x.outer*32768) + (x.c*1024)) + cse_var_4)
-                 {
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[cse_var_1], 32)*packedB_1[cse_var_2]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 1)], 32)*packedB_1[(cse_var_2 + 1)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 2)], 32)*packedB_1[(cse_var_2 + 2)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 3)], 32)*packedB_1[(cse_var_2 + 3)]))
-                }
-              }
-            }
-            for (x.inner: int32, 0, 32) {
-              for (y.inner: int32, 0, 32) {
-                C_3: Buffer(C_2, float32, [1048576], [])[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = C.global_1[((x.inner*32) + y.inner)]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    block caching: 0.109743
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            C_global = T.allocate([1024], "float32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for x in T.parallel(32):
+                for y in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[x * 1024 + y] = B_2[y * 1024 + x * 32:y * 1024 + x * 32 + 32]
+            for x_outer, y_outer in T.grid(32, 32):
+                C_global_1 = T.buffer_decl((1024,), data=C_global)
+                for x_c_init in range(32):
+                    C_global_1[x_c_init * 32:x_c_init * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                for k_outer, x_c in T.grid(256, 32):
+                    cse_var_4: T.int32 = k_outer * 4
+                    cse_var_3: T.int32 = x_c * 32
+                    cse_var_2: T.int32 = y_outer * 1024 + cse_var_4
+                    cse_var_1: T.int32 = x_outer * 32768 + x_c * 1024 + cse_var_4
+                    A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1], 32) * packedB_1[cse_var_2]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 1], 32) * packedB_1[cse_var_2 + 1]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 2], 32) * packedB_1[cse_var_2 + 2]
+                    C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 3], 32) * packedB_1[cse_var_2 + 3]
+                for x_inner, y_inner in T.grid(32, 32):
+                    C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                    C_2[x_outer * 32768 + x_inner * 1024 + y_outer * 32 + y_inner] = C_global_1[x_inner * 32 + y_inner]
 
 
 
@@ -1517,50 +1478,40 @@ of thread-level parallelization.
 
  .. code-block:: none
 
-    parallelization: 0.146416
-    @main = primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
-      attr = {"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True}
-      buffers = {A: Buffer(A_2: Pointer(float32), float32, [1024, 1024], []),
-                 B: Buffer(B_2: Pointer(float32), float32, [1024, 1024], []),
-                 C: Buffer(C_2: Pointer(float32), float32, [1024, 1024], [])}
-      buffer_map = {A_1: A, B_1: B, C_1: C} {
-      allocate(packedB: Pointer(global float32x32), float32x32, [32768]), storage_scope = global {
-        for (x: int32, 0, 32) "parallel" {
-          for (y: int32, 0, 1024) {
-            packedB_1: Buffer(packedB, float32x32, [32768], [])[((x*1024) + y)] = B_3: Buffer(B_2, float32, [1048576], [])[ramp(((y*1024) + (x*32)), 1, 32)]
-          }
-        }
-        for (x.outer: int32, 0, 32) "parallel" {
-          allocate(C.global: Pointer(global float32), float32, [1024]), storage_scope = global;
-          for (y.outer: int32, 0, 32) {
-            for (x.c.init: int32, 0, 32) {
-              C.global_1: Buffer(C.global, float32, [1024], [])[ramp((x.c.init*32), 1, 32)] = broadcast(0f32, 32)
-            }
-            for (k.outer: int32, 0, 256) {
-              for (x.c: int32, 0, 32) {
-                let cse_var_4: int32 = (k.outer*4)
-                let cse_var_3: int32 = (x.c*32)
-                let cse_var_2: int32 = ((y.outer*1024) + cse_var_4)
-                let cse_var_1: int32 = (((x.outer*32768) + (x.c*1024)) + cse_var_4)
-                 {
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3: Buffer(A_2, float32, [1048576], [])[cse_var_1], 32)*packedB_1[cse_var_2]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 1)], 32)*packedB_1[(cse_var_2 + 1)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 2)], 32)*packedB_1[(cse_var_2 + 2)]))
-                  C.global_1[ramp(cse_var_3, 1, 32)] = (C.global_1[ramp(cse_var_3, 1, 32)] + (broadcast(A_3[(cse_var_1 + 3)], 32)*packedB_1[(cse_var_2 + 3)]))
-                }
-              }
-            }
-            for (x.inner: int32, 0, 32) {
-              for (y.inner: int32, 0, 32) {
-                C_3: Buffer(C_2, float32, [1048576], [])[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = C.global_1[((x.inner*32) + y.inner)]
-              }
-            }
-          }
-        }
-      }
-    }
-
-
+    parallelization: 0.143840
+    @I.ir_module
+    class Module:
+        @T.prim_func
+        def main(A: T.handle, B: T.handle, C: T.handle):
+            T.func_attr({"from_legacy_te_schedule": True, "global_symbol": "main", "tir.noalias": True})
+            A_1 = T.match_buffer(A, (1024, 1024))
+            B_1 = T.match_buffer(B, (1024, 1024))
+            C_1 = T.match_buffer(C, (1024, 1024))
+            packedB = T.allocate([32768], "float32x32", "global")
+            packedB_1 = T.buffer_decl((32768,), "float32x32", data=packedB)
+            for x in T.parallel(32):
+                for y in range(1024):
+                    B_2 = T.buffer_decl((1048576,), data=B_1.data)
+                    packedB_1[x * 1024 + y] = B_2[y * 1024 + x * 32:y * 1024 + x * 32 + 32]
+            for x_outer in T.parallel(32):
+                C_global = T.allocate([1024], "float32", "global")
+                for y_outer in range(32):
+                    C_global_1 = T.buffer_decl((1024,), data=C_global)
+                    for x_c_init in range(32):
+                        C_global_1[x_c_init * 32:x_c_init * 32 + 32] = T.Broadcast(T.float32(0), 32)
+                    for k_outer, x_c in T.grid(256, 32):
+                        cse_var_4: T.int32 = k_outer * 4
+                        cse_var_3: T.int32 = x_c * 32
+                        cse_var_2: T.int32 = y_outer * 1024 + cse_var_4
+                        cse_var_1: T.int32 = x_outer * 32768 + x_c * 1024 + cse_var_4
+                        A_2 = T.buffer_decl((1048576,), data=A_1.data)
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1], 32) * packedB_1[cse_var_2]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 1], 32) * packedB_1[cse_var_2 + 1]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 2], 32) * packedB_1[cse_var_2 + 2]
+                        C_global_1[cse_var_3:cse_var_3 + 32] = C_global_1[cse_var_3:cse_var_3 + 32] + T.Broadcast(A_2[cse_var_1 + 3], 32) * packedB_1[cse_var_2 + 3]
+                    for x_inner, y_inner in T.grid(32, 32):
+                        C_2 = T.buffer_decl((1048576,), data=C_1.data)
+                        C_2[x_outer * 32768 + x_inner * 1024 + y_outer * 32 + y_inner] = C_global_1[x_inner * 32 + y_inner]
 
 
 
@@ -1597,13 +1548,13 @@ working, we can compare the results.
  .. code-block:: none
 
                 Operator                  Timing             Performance
-                    none            3.2553497041                     1.0
-                blocking            0.3001310018     0.09219623975328839
-           vectorization     0.33420826950000004     0.10266432177135265
-        loop permutation     0.11980180809999999      0.0368015171915674
-           array packing     0.11110955859999998    0.034131374107077145
-           block caching            0.1119645329    0.034394010805961814
-         parallelization            0.1464159081     0.04497701365711777
+                    none            3.2604858586                     1.0
+                blocking             0.297953824     0.09138325909744523
+           vectorization     0.32929113600000004     0.10099449906566757
+        loop permutation     0.11673757879999999    0.035803737192138974
+           array packing            0.1089929865     0.03342844938661989
+           block caching            0.1097431324      0.0336585212018438
+         parallelization            0.1438398794     0.04411608749064243
 
 
 
diff --git a/docs/_sources/tutorial/tensor_ir_blitz_course.rst.txt b/docs/_sources/tutorial/tensor_ir_blitz_course.rst.txt
index 6b5ad68cdf..3fb6484e44 100644
--- a/docs/_sources/tutorial/tensor_ir_blitz_course.rst.txt
+++ b/docs/_sources/tutorial/tensor_ir_blitz_course.rst.txt
@@ -126,23 +126,22 @@ Following is a simple example for vector addition.
  .. code-block:: none
 
     <class 'tvm.ir.module.IRModule'>
-    # from tvm.script import tir as T
-    @tvm.script.ir_module
+    @I.ir_module
     class Module:
         @T.prim_func
-        def main(A: T.Buffer[8, "float32"], B: T.Buffer[8, "float32"]):
-            # function attr dict
-            T.func_attr({"tir.noalias": True, "global_symbol": "main"})
-            # body
-            # with T.block("root")
-            for i in T.serial(8):
-                with T.block("B"):
-                    vi = T.axis.spatial(8, i)
-                    T.reads(A[vi])
-                    T.writes(B[vi])
-                    B[vi] = A[vi] + T.float32(1)
-    
-
+        def main(a: T.handle, b: T.handle):
+            T.func_attr({"global_symbol": "main", "tir.noalias": True})
+            A = T.match_buffer(a, (8,))
+            B = T.match_buffer(b, (8,))
+            with T.block("root"):
+                T.reads()
+                T.writes()
+                for i in range(8):
+                    with T.block("B"):
+                        vi = T.axis.spatial(8, i)
+                        T.reads(A[vi])
+                        T.writes(B[vi])
+                        B[vi] = A[vi] + T.float32(1)
 
 
 
@@ -175,23 +174,22 @@ to an IRModule.
 
  .. code-block:: none
 
-    # from tvm.script import tir as T
-    @tvm.script.ir_module
+    @I.ir_module
     class Module:
         @T.prim_func
-        def main(A: T.Buffer[8, "float32"], B: T.Buffer[8, "float32"]):
-            # function attr dict
+        def main(var_A: T.handle, var_B: T.handle):
             T.func_attr({"global_symbol": "main", "tir.noalias": True})
-            # body
-            # with T.block("root")
-            for i0 in T.serial(8):
-                with T.block("B"):
-                    v_i0 = T.axis.spatial(8, i0)
-                    T.reads(A[v_i0])
-                    T.writes(B[v_i0])
-                    B[v_i0] = A[v_i0] + T.float32(1)
-    
-
+            A = T.match_buffer(var_A, (8,))
+            B = T.match_buffer(var_B, (8,))
+            with T.block("root"):
+                T.reads()
+                T.writes()
+                for i0 in range(8):
+                    with T.block("B"):
+                        v_i0 = T.axis.spatial(8, i0)
+                        T.reads(A[v_i0])
+                        T.writes(B[v_i0])
+                        B[v_i0] = A[v_i0] + T.float32(1)
 
 
 
@@ -322,23 +320,22 @@ Tile the loop into 3 loops and print the result.
 
  .. code-block:: none
 
-    # from tvm.script import tir as T
-    @tvm.script.ir_module
+    @I.ir_module
     class Module:
         @T.prim_func
-        def main(A: T.Buffer[8, "float32"], B: T.Buffer[8, "float32"]):
-            # function attr dict
-            T.func_attr({"tir.noalias": True, "global_symbol": "main"})
-            # body
-            # with T.block("root")
-            for i_0, i_1, i_2 in T.grid(2, 2, 2):
-                with T.block("B"):
-                    vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
-                    T.reads(A[vi])
-                    T.writes(B[vi])
-                    B[vi] = A[vi] + T.float32(1)
-    
-
+        def main(a: T.handle, b: T.handle):
+            T.func_attr({"global_symbol": "main", "tir.noalias": True})
+            A = T.match_buffer(a, (8,))
+            B = T.match_buffer(b, (8,))
+            with T.block("root"):
+                T.reads()
+                T.writes()
+                for i_0, i_1, i_2 in T.grid(2, 2, 2):
+                    with T.block("B"):
+                        vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
+                        T.reads(A[vi])
+                        T.writes(B[vi])
+                        B[vi] = A[vi] + T.float32(1)
 
 
 
@@ -363,23 +360,22 @@ We can also reorder the loops. Now we move loop `i_2` to outside of `i_1`.
 
  .. code-block:: none
 
-    # from tvm.script import tir as T
-    @tvm.script.ir_module
+    @I.ir_module
     class Module:
         @T.prim_func
-        def main(A: T.Buffer[8, "float32"], B: T.Buffer[8, "float32"]):
-            # function attr dict
-            T.func_attr({"tir.noalias": True, "global_symbol": "main"})
-            # body
-            # with T.block("root")
-            for i_0, i_2, i_1 in T.grid(2, 2, 2):
-                with T.block("B"):
-                    vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
-                    T.reads(A[vi])
-                    T.writes(B[vi])
-                    B[vi] = A[vi] + T.float32(1)
-    
-
+        def main(a: T.handle, b: T.handle):
+            T.func_attr({"global_symbol": "main", "tir.noalias": True})
+            A = T.match_buffer(a, (8,))
+            B = T.match_buffer(b, (8,))
+            with T.block("root"):
+                T.reads()
+                T.writes()
+                for i_0, i_2, i_1 in T.grid(2, 2, 2):
+                    with T.block("B"):
+                        vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
+                        T.reads(A[vi])
+                        T.writes(B[vi])
+                        B[vi] = A[vi] + T.float32(1)
 
 
 
@@ -410,25 +406,24 @@ also use primitives and do incrementally transformation.
 
  .. code-block:: none
 
-    # from tvm.script import tir as T
-    @tvm.script.ir_module
+    @I.ir_module
     class Module:
         @T.prim_func
-        def main(A: T.Buffer[8, "float32"], B: T.Buffer[8, "float32"]):
-            # function attr dict
-            T.func_attr({"tir.noalias": True, "global_symbol": "main"})
-            # body
-            # with T.block("root")
-            for i_0 in T.thread_binding(2, thread="blockIdx.x"):
-                for i_2 in T.thread_binding(2, thread="threadIdx.x"):
-                    for i_1 in T.serial(2):
-                        with T.block("B"):
-                            vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
-                            T.reads(A[vi])
-                            T.writes(B[vi])
-                            B[vi] = A[vi] + T.float32(1)
-    
-
+        def main(a: T.handle, b: T.handle):
+            T.func_attr({"global_symbol": "main", "tir.noalias": True})
+            A = T.match_buffer(a, (8,))
+            B = T.match_buffer(b, (8,))
+            with T.block("root"):
+                T.reads()
+                T.writes()
+                for i_0 in T.thread_binding(2, thread="blockIdx.x"):
+                    for i_2 in T.thread_binding(2, thread="threadIdx.x"):
+                        for i_1 in range(2):
+                            with T.block("B"):
+                                vi = T.axis.spatial(8, i_0 * 4 + i_1 * 2 + i_2)
+                                T.reads(A[vi])
+                                T.writes(B[vi])
+                                B[vi] = A[vi] + T.float32(1)
 
 
 
diff --git a/docs/commit_hash b/docs/commit_hash
index 98c0c7e5c5..38111def60 100644
--- a/docs/commit_hash
+++ b/docs/commit_hash
@@ -1 +1 @@
-2e2ae26898bf6c651222b2f88cf166945b6b1d9a
+7e3dc45fed385e12f8cfd9c9284b5551ada82f95
diff --git a/docs/genindex.html b/docs/genindex.html
index 00e80ff623..1e57b29243 100644
--- a/docs/genindex.html
+++ b/docs/genindex.html
@@ -3209,6 +3209,8 @@
 </li>
       </ul></li>
       <li><a href="reference/api/python/relay/testing.html#tvm.relay.testing.Prelude">Prelude (class in tvm.relay.testing)</a>
+</li>
+      <li><a href="reference/api/python/ir.html#tvm.ir.pretty_print">pretty_print() (in module tvm.ir)</a>
 </li>
       <li><a href="reference/api/python/tir.html#tvm.tir.transform.prim_func_pass">prim_func_pass() (in module tvm.tir.transform)</a>
 </li>
@@ -3612,6 +3614,8 @@
         <li><a href="reference/api/python/ir.html#tvm.ir.IRModule.script">(tvm.ir.IRModule method)</a>
 </li>
         <li><a href="reference/api/python/tir.html#tvm.tir.PrimFunc.script">(tvm.tir.PrimFunc method)</a>
+</li>
+        <li><a href="reference/api/python/tir.html#tvm.tir.Stmt.script">(tvm.tir.Stmt method)</a>
 </li>
       </ul></li>
       <li><a href="reference/api/python/relay/analysis.html#tvm.relay.analysis.search_fc_transpose">search_fc_transpose() (in module tvm.relay.analysis)</a>
@@ -3708,6 +3712,8 @@
         <li><a href="reference/api/python/tir.html#tvm.tir.PrimFunc.show">(tvm.tir.PrimFunc method)</a>
 </li>
         <li><a href="reference/api/python/tir.html#tvm.tir.Schedule.show">(tvm.tir.Schedule method)</a>
+</li>
+        <li><a href="reference/api/python/tir.html#tvm.tir.Stmt.show">(tvm.tir.Stmt method)</a>
 </li>
       </ul></li>
       <li><a href="reference/api/python/tir.html#tvm.tir.Shuffle">Shuffle (class in tvm.tir)</a>
@@ -3726,10 +3732,10 @@
         <li><a href="reference/api/python/topi.html#tvm.topi.nn.simplify">(in module tvm.topi.nn)</a>
 </li>
       </ul></li>
-      <li><a href="reference/api/python/relay/transform.html#tvm.relay.transform.SimplifyExpr">SimplifyExpr() (in module tvm.relay.transform)</a>
-</li>
   </ul></td>
   <td style="width: 33%; vertical-align: top;"><ul>
+      <li><a href="reference/api/python/relay/transform.html#tvm.relay.transform.SimplifyExpr">SimplifyExpr() (in module tvm.relay.transform)</a>
+</li>
       <li><a href="reference/api/python/relay/transform.html#tvm.relay.transform.SimplifyFCTranspose">SimplifyFCTranspose() (in module tvm.relay.transform)</a>
 </li>
       <li><a href="reference/api/python/relay/transform.html#tvm.relay.transform.SimplifyInference">SimplifyInference() (in module tvm.relay.transform)</a>
diff --git a/docs/how_to/compile_models/from_darknet.html b/docs/how_to/compile_models/from_darknet.html
index 9b24324d73..cfc32fda28 100644
--- a/docs/how_to/compile_models/from_darknet.html
+++ b/docs/how_to/compile_models/from_darknet.html
@@ -585,7 +585,7 @@ class:[&#39;truck 0.9266&#39;] left:471 top:83 right:689 bottom:169
 class:[&#39;bicycle 0.9984&#39;] 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  18.798 seconds)</p>
+<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 1 minutes  13.059 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 73fe2e6288..8fa956abd8 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 1000ms/step
+1/1 [==============================] - 1s 952ms/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 24fea6308b..8a00bf0ccd 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">&quot;x&quot;</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.zipfaabf753-aa08-47b4-a3a6-91f5abd6fdd6 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.zipf850238c-fdb0-49a7-83f7-d5e376bb9e0e 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 cf3c2c37e3..89421012a5 100644
--- a/docs/how_to/compile_models/from_oneflow.html
+++ b/docs/how_to/compile_models/from_oneflow.html
@@ -449,13 +449,13 @@ 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: &quot;https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/ResNet/resnet18.zip&quot; to /workspace/.oneflow/flowvision_cache/resnet18.zip
 
   0%|          | 0.00/41.5M [00:00&lt;?, ?B/s]
- 19%|#9        | 7.99M/41.5M [00:00&lt;00:00, 43.6MB/s]
- 39%|###8      | 16.0M/41.5M [00:00&lt;00:00, 56.4MB/s]
- 54%|#####3    | 22.3M/41.5M [00:00&lt;00:00, 59.9MB/s]
- 68%|######8   | 28.3M/41.5M [00:00&lt;00:00, 56.5MB/s]
- 82%|########1 | 33.8M/41.5M [00:00&lt;00:00, 55.8MB/s]
- 95%|#########4| 39.2M/41.5M [00:00&lt;00:00, 52.5MB/s]
-100%|##########| 41.5M/41.5M [00:00&lt;00:00, 52.8MB/s]
+ 15%|#5        | 6.33M/41.5M [00:00&lt;00:00, 61.2MB/s]
+ 29%|##9       | 12.2M/41.5M [00:00&lt;00:00, 52.0MB/s]
+ 48%|####7     | 19.8M/41.5M [00:00&lt;00:00, 58.6MB/s]
+ 61%|######1   | 25.4M/41.5M [00:00&lt;00:00, 49.5MB/s]
+ 77%|#######7  | 32.0M/41.5M [00:00&lt;00:00, 48.2MB/s]
+ 96%|#########6| 40.0M/41.5M [00:00&lt;00:00, 56.5MB/s]
+100%|##########| 41.5M/41.5M [00:00&lt;00:00, 55.5MB/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 cd1cccc4ac..7f836c7848 100644
--- a/docs/how_to/compile_models/from_pytorch.html
+++ b/docs/how_to/compile_models/from_pytorch.html
@@ -432,11 +432,10 @@ be unstable.</p>
 Downloading: &quot;https://download.pytorch.org/models/resnet18-f37072fd.pth&quot; to /workspace/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth
 
   0%|          | 0.00/44.7M [00:00&lt;?, ?B/s]
- 21%|##        | 9.23M/44.7M [00:00&lt;00:00, 96.8MB/s]
- 41%|####1     | 18.5M/44.7M [00:00&lt;00:00, 53.0MB/s]
- 69%|######9   | 31.0M/44.7M [00:00&lt;00:00, 77.8MB/s]
- 90%|########9 | 40.1M/44.7M [00:00&lt;00:00, 78.7MB/s]
-100%|##########| 44.7M/44.7M [00:00&lt;00:00, 77.3MB/s]
+ 34%|###4      | 15.4M/44.7M [00:00&lt;00:00, 161MB/s]
+ 69%|######8   | 30.7M/44.7M [00:00&lt;00:00, 119MB/s]
+ 96%|#########5| 42.7M/44.7M [00:00&lt;00:00, 111MB/s]
+100%|##########| 44.7M/44.7M [00:00&lt;00:00, 115MB/s]
 </pre></div>
 </div>
 </div>
... 386549 lines suppressed ...