You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by "krishnaraj36 (via GitHub)" <gi...@apache.org> on 2023/01/24 12:24:23 UTC
[GitHub] [tvm] krishnaraj36 opened a new pull request, #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
krishnaraj36 opened a new pull request, #13834:
URL: https://github.com/apache/tvm/pull/13834
Enable the DepthToSpace and Resize bilinear operator in CLML runtime and bug fix in concat layer.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] srkreddy1238 merged pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "srkreddy1238 (via GitHub)" <gi...@apache.org>.
srkreddy1238 merged PR #13834:
URL: https://github.com/apache/tvm/pull/13834
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] krishnaraj36 commented on pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "krishnaraj36 (via GitHub)" <gi...@apache.org>.
krishnaraj36 commented on PR #13834:
URL: https://github.com/apache/tvm/pull/13834#issuecomment-1402174988
@srkreddy1238
Please review the PR, let me know feedback.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] tvm-bot commented on pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "tvm-bot (via GitHub)" <gi...@apache.org>.
tvm-bot commented on PR #13834:
URL: https://github.com/apache/tvm/pull/13834#issuecomment-1401852745
<!---bot-comment-->
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from [Reviewers](https://github.com/apache/incubator-tvm/blob/master/CONTRIBUTORS.md#reviewers) by @-ing them in a comment.
<!--bot-comment-ccs-start-->
* cc @areusch <sub>See [#10317](https://github.com/apache/tvm/issues/10317) for details</sub><!--bot-comment-ccs-end-->
<sub>Generated by [tvm-bot](https://github.com/apache/tvm/blob/main/ci/README.md#github-actions)</sub>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] srkreddy1238 commented on a diff in pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "srkreddy1238 (via GitHub)" <gi...@apache.org>.
srkreddy1238 commented on code in PR #13834:
URL: https://github.com/apache/tvm/pull/13834#discussion_r1087433807
##########
src/runtime/contrib/clml/clml_runtime.cc:
##########
@@ -1151,13 +1159,14 @@ class CLMLRuntime : public JSONRuntimeBase {
cl_arithmetic_mode_qcom cl_arithmetic_mode = MakeCLArithMode(cl_dtype);
int inputSize = input_.size();
auto output = MakeCLMLTensorFromJSONNode(node, CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
+ cl_uint axis = std::stoi(node.GetAttr<std::vector<std::string>>("axis")[0]);
cl_ml_tensor_qcom* concatInputs = new cl_ml_tensor_qcom[inputSize];
for (int i = 0; i < inputSize; i++) {
auto input = MakeCLMLTensorFromJSONEntry(node.GetInputs()[i], {},
CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
concatInputs[i] = input->tensor;
}
- cl_ml_op_concat_desc_qcom concatDesc = {1, (cl_uint)inputSize, cl_arithmetic_mode};
+ cl_ml_op_concat_desc_qcom concatDesc = {axis, (cl_uint)inputSize, cl_arithmetic_mode};
Review Comment:
Till them probably we can limit it in frontend ```clml.py``` and let it take TVM path for ```axis != 1``` instead of failing at run time.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] echuraev commented on a diff in pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "echuraev (via GitHub)" <gi...@apache.org>.
echuraev commented on code in PR #13834:
URL: https://github.com/apache/tvm/pull/13834#discussion_r1086246706
##########
src/runtime/contrib/clml/clml_runtime.cc:
##########
@@ -1151,13 +1159,14 @@ class CLMLRuntime : public JSONRuntimeBase {
cl_arithmetic_mode_qcom cl_arithmetic_mode = MakeCLArithMode(cl_dtype);
int inputSize = input_.size();
auto output = MakeCLMLTensorFromJSONNode(node, CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
+ cl_uint axis = std::stoi(node.GetAttr<std::vector<std::string>>("axis")[0]);
cl_ml_tensor_qcom* concatInputs = new cl_ml_tensor_qcom[inputSize];
for (int i = 0; i < inputSize; i++) {
auto input = MakeCLMLTensorFromJSONEntry(node.GetInputs()[i], {},
CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
concatInputs[i] = input->tensor;
}
- cl_ml_op_concat_desc_qcom concatDesc = {1, (cl_uint)inputSize, cl_arithmetic_mode};
+ cl_ml_op_concat_desc_qcom concatDesc = {axis, (cl_uint)inputSize, cl_arithmetic_mode};
Review Comment:
Could you please update `concat` test and check how it works for different axes?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] krishnaraj36 commented on a diff in pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "krishnaraj36 (via GitHub)" <gi...@apache.org>.
krishnaraj36 commented on code in PR #13834:
URL: https://github.com/apache/tvm/pull/13834#discussion_r1086837349
##########
tests/python/contrib/test_clml/test_ops.py:
##########
@@ -574,5 +574,107 @@ def _verify(out, params, inputs):
_verify(*(_get_model((1, 16), relay.nn.relu)))
+@pytest.mark.parametrize("dtype", ["float32"])
+@tvm.testing.requires_openclml
+def test_depth_to_space(device, dtype):
+ def _get_model(a_shape, block_size):
+ a = relay.var("a", shape=(a_shape), dtype=dtype)
+ out = relay.nn.depth_to_space(a, block_size)
+ inputs = {"a": tvm.nd.array(np.random.uniform(-1, 1, a_shape).astype(dtype))}
+ params = {}
+ return out, params, inputs
+
+ def _verify(out, params, inputs):
+ mod = IRModule.from_expr(out)
+ opencl_out = build_and_run(mod, inputs, 1, params, device, enable_clml=False)[0]
+ clml_out = build_and_run(mod, inputs, 1, params, device, enable_clml=True)[0]
+ tvm.testing.assert_allclose(
+ clml_out[0].asnumpy(), opencl_out[0].asnumpy(), rtol=1e-3, atol=1e-3
+ )
+
+ # Check to make sure these ops are offloaded to CLML instead of TVM.
+ exp_codegen = [
+ {
+ "attrs": {
+ "dtype": [[dtype]],
+ "shape": [[list(inputs["a"].shape)]],
+ },
+ "name": "",
+ "op": "input",
+ },
+ {
+ "attrs": {
+ "block_size": [[str(int(out.attrs.block_size))]],
+ "layout": [["NCHW"]],
+ "mode": [["DCR"]],
+ "dtype": [[dtype]],
+ "num_inputs": "1",
+ "num_outputs": "1",
+ "shape": [[list(clml_out[0].shape)]],
+ },
+ "inputs": [[0, 0, 0]],
+ "name": "nn.depth_to_space",
+ "op": "kernel",
+ },
+ ]
+ verify_codegen(out, exp_codegen, device, params)
+
+ _verify(*(_get_model((1, 64, 8, 8), 4)))
+ _verify(*(_get_model((1, 64, 8, 8), 8)))
+
+
+@pytest.mark.parametrize("dtype", ["float32"])
Review Comment:
Yes I have added float16 test case as well.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] krishnaraj36 commented on a diff in pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "krishnaraj36 (via GitHub)" <gi...@apache.org>.
krishnaraj36 commented on code in PR #13834:
URL: https://github.com/apache/tvm/pull/13834#discussion_r1086836460
##########
src/runtime/contrib/clml/clml_runtime.cc:
##########
@@ -1151,13 +1159,14 @@ class CLMLRuntime : public JSONRuntimeBase {
cl_arithmetic_mode_qcom cl_arithmetic_mode = MakeCLArithMode(cl_dtype);
int inputSize = input_.size();
auto output = MakeCLMLTensorFromJSONNode(node, CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
+ cl_uint axis = std::stoi(node.GetAttr<std::vector<std::string>>("axis")[0]);
cl_ml_tensor_qcom* concatInputs = new cl_ml_tensor_qcom[inputSize];
for (int i = 0; i < inputSize; i++) {
auto input = MakeCLMLTensorFromJSONEntry(node.GetInputs()[i], {},
CL_TENSOR_LAYOUT_OPTIMAL_QCOM, cl_dtype);
concatInputs[i] = input->tensor;
}
- cl_ml_op_concat_desc_qcom concatDesc = {1, (cl_uint)inputSize, cl_arithmetic_mode};
+ cl_ml_op_concat_desc_qcom concatDesc = {axis, (cl_uint)inputSize, cl_arithmetic_mode};
Review Comment:
We have known issue for Concat layer in CLML pipeline when axis != 1, for which team is working on. Once it is resolved we will add more test cases on axis attributes.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] srkreddy1238 commented on a diff in pull request #13834: [CLML][RUNTIME] Enable more ops in CLML runtime
Posted by "srkreddy1238 (via GitHub)" <gi...@apache.org>.
srkreddy1238 commented on code in PR #13834:
URL: https://github.com/apache/tvm/pull/13834#discussion_r1086232498
##########
tests/python/contrib/test_clml/test_ops.py:
##########
@@ -574,5 +574,107 @@ def _verify(out, params, inputs):
_verify(*(_get_model((1, 16), relay.nn.relu)))
+@pytest.mark.parametrize("dtype", ["float32"])
+@tvm.testing.requires_openclml
+def test_depth_to_space(device, dtype):
+ def _get_model(a_shape, block_size):
+ a = relay.var("a", shape=(a_shape), dtype=dtype)
+ out = relay.nn.depth_to_space(a, block_size)
+ inputs = {"a": tvm.nd.array(np.random.uniform(-1, 1, a_shape).astype(dtype))}
+ params = {}
+ return out, params, inputs
+
+ def _verify(out, params, inputs):
+ mod = IRModule.from_expr(out)
+ opencl_out = build_and_run(mod, inputs, 1, params, device, enable_clml=False)[0]
+ clml_out = build_and_run(mod, inputs, 1, params, device, enable_clml=True)[0]
+ tvm.testing.assert_allclose(
+ clml_out[0].asnumpy(), opencl_out[0].asnumpy(), rtol=1e-3, atol=1e-3
+ )
+
+ # Check to make sure these ops are offloaded to CLML instead of TVM.
+ exp_codegen = [
+ {
+ "attrs": {
+ "dtype": [[dtype]],
+ "shape": [[list(inputs["a"].shape)]],
+ },
+ "name": "",
+ "op": "input",
+ },
+ {
+ "attrs": {
+ "block_size": [[str(int(out.attrs.block_size))]],
+ "layout": [["NCHW"]],
+ "mode": [["DCR"]],
+ "dtype": [[dtype]],
+ "num_inputs": "1",
+ "num_outputs": "1",
+ "shape": [[list(clml_out[0].shape)]],
+ },
+ "inputs": [[0, 0, 0]],
+ "name": "nn.depth_to_space",
+ "op": "kernel",
+ },
+ ]
+ verify_codegen(out, exp_codegen, device, params)
+
+ _verify(*(_get_model((1, 64, 8, 8), 4)))
+ _verify(*(_get_model((1, 64, 8, 8), 8)))
+
+
+@pytest.mark.parametrize("dtype", ["float32"])
Review Comment:
Can we cover "float16" also here if supported on CLML & TVM too ?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org