Skip to content

Move the logic of mkldnn layout in GetKernelTypeForVar from ActivationOp to OperatorWithKernel #47104

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 19, 2022
Merged

Move the logic of mkldnn layout in GetKernelTypeForVar from ActivationOp to OperatorWithKernel #47104

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions paddle/fluid/framework/operator.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2660,6 +2660,19 @@ OpKernelType OperatorWithKernel::GetKernelTypeForVar(
const std::string& var_name,
const phi::DenseTensor& tensor,
const OpKernelType& expected_kernel_type) const {
#ifdef PADDLE_WITH_MKLDNN
// When the op is first oneDNN op (there was some non oneDNN op
// previously)
// then we also need to rotate shape NHWC -> NCWH
if ((expected_kernel_type.data_layout_ == phi::DataLayout::kMKLDNN) &&
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

此处兼容其他 op 吗？很多 mkldnn op 的 GetKernelTypeForVar 函数中，并没有这部分代码

Do other ops compatible with the code here? Many mkldnn ops do not contain these codes.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

根据PR39097中的comment来看，这应该是一个通用的逻辑，具体是否有影响可能需要CE模型测试做进一步验证

(tensor.layout() != phi::DataLayout::kMKLDNN) &&
paddle::platform::MKLDNNDeviceContext::tls()
.get_cur_paddle_data_layout() == phi::DataLayout::kNHWC) {
return framework::OpKernelType(expected_kernel_type.data_type_,
tensor.place(),
phi::DataLayout::kNHWC);
}
#endif
return OpKernelType(
expected_kernel_type.data_type_, tensor.place(), tensor.layout());
}
Expand Down
21 changes: 0 additions & 21 deletions paddle/fluid/operators/activation_op.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,27 +110,6 @@ class ActivationOp : public framework::OperatorWithKernel {
const framework::ExecutionContext& ctx) const override {
return GetKernelType(ctx, *this, "X");
}

framework::OpKernelType GetKernelTypeForVar(
const std::string& var_name,
const phi::DenseTensor& tensor,
const framework::OpKernelType& expected_kernel_type) const override {
#ifdef PADDLE_WITH_MKLDNN
// When activation is first oneDNN op (there was some non oneDNN op
// previously)
// then we also need to rotate shape NHWC -> NCWH
if ((expected_kernel_type.data_layout_ == phi::DataLayout::kMKLDNN) &&
(tensor.layout() != phi::DataLayout::kMKLDNN) &&
paddle::platform::MKLDNNDeviceContext::tls()
.get_cur_paddle_data_layout() == phi::DataLayout::kNHWC) {
return framework::OpKernelType(expected_kernel_type.data_type_,
tensor.place(),
phi::DataLayout::kNHWC);
}
#endif
return framework::OpKernelType(
expected_kernel_type.data_type_, tensor.place(), tensor.layout());
}
};

class ActivationOpInferVarType
Expand Down