[oneDNN] Concat refactoring and disabling caching #35002
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
jakpiase
reviewed
Aug 19, 2021
| class ConcatMKLDNNHandler | ||
| : public platform::MKLDNNHandlerNoCachingT<T, dnnl::concat> { | ||
| public: | ||
| ConcatMKLDNNHandler(const paddle::framework::ExecutionContext& ctx, |
There was a problem hiding this comment.
Suggested change
| ConcatMKLDNNHandler(const paddle::framework::ExecutionContext& ctx, | |
| ConcatMKLDNNHandler(const framework::ExecutionContext& ctx, |
| } | ||
|
|
||
| memory::data_type dt = | ||
| paddle::framework::ToMKLDNNDataType(inputs[0]->type()); |
There was a problem hiding this comment.
Suggested change
| paddle::framework::ToMKLDNNDataType(inputs[0]->type()); | |
| framework::ToMKLDNNDataType(inputs[0]->type()); |
| srcs_md.reserve(inputs.size()); | ||
|
|
||
| // Create memory descriptors for each of inputs | ||
| for (size_t i = 0; i < inputs.size(); i++) { |
There was a problem hiding this comment.
Suggested change
| for (size_t i = 0; i < inputs.size(); i++) { | |
| for (size_t i = 0; i < inputs.size(); ++i) { |
| // Create memory descriptors for each of inputs | ||
| for (size_t i = 0; i < inputs.size(); i++) { | ||
| const auto dims = | ||
| paddle::framework::vectorize<int64_t>(inputs[i]->dims()); |
There was a problem hiding this comment.
Suggested change
| paddle::framework::vectorize<int64_t>(inputs[i]->dims()); | |
| framework::vectorize<int64_t>(inputs[i]->dims()); |
| srcs_md.emplace_back(memory::desc(dims, dt, inputs[i]->format())); | ||
| } | ||
|
|
||
| auto dst_dims = paddle::framework::vectorize<int64_t>(output->dims()); |
There was a problem hiding this comment.
Suggested change
| auto dst_dims = paddle::framework::vectorize<int64_t>(output->dims()); | |
| auto dst_dims = framework::vectorize<int64_t>(output->dims()); |
| // we cannot use base AcquireForwardPrimitiveDescriptor | ||
| void AcquireForwardPrimitiveDescriptor( | ||
| const mkldnn::memory::desc& dst_md, const int concat_axis, | ||
| const std::vector<mkldnn::memory::desc>& srcs_md) { |
There was a problem hiding this comment.
Suggested change
| const std::vector<mkldnn::memory::desc>& srcs_md) { | |
| const std::vector<memory::desc>& srcs_md) { |
| // (jczaja) concat oneDNN prim is not having .desc attribute so | ||
| // we cannot use base AcquireForwardPrimitiveDescriptor | ||
| void AcquireForwardPrimitiveDescriptor( | ||
| const mkldnn::memory::desc& dst_md, const int concat_axis, |
There was a problem hiding this comment.
Suggested change
| const mkldnn::memory::desc& dst_md, const int concat_axis, | |
| const memory::desc& dst_md, const int concat_axis, |
| } | ||
|
|
||
| std::shared_ptr<mkldnn::memory> AcquireSrcMemory( | ||
| const framework::Tensor& input, int i) { |
There was a problem hiding this comment.
Suggested change
| const framework::Tensor& input, int i) { | |
| const Tensor& input, int i) { |
| template <typename T> | ||
| class ConcatMKLDNNOpKernel : public paddle::framework::OpKernel<T> { | ||
| public: | ||
| void Compute(const paddle::framework::ExecutionContext& ctx) const override { | ||
| auto& dev_ctx = | ||
| ctx.template device_context<paddle::platform::MKLDNNDeviceContext>(); |
There was a problem hiding this comment.
Suggested change
| ctx.template device_context<paddle::platform::MKLDNNDeviceContext>(); | |
| ctx.template device_context<platform::MKLDNNDeviceContext>(); |
There was a problem hiding this comment.
You have forgotten about this one
|
@jakpiase Please continue your review |
jakpiase
previously approved these changes
Aug 19, 2021
jczaja
force-pushed
the
prv-concat-refactor
branch
from
bc89f59
to
4ad092a
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
Others
PR changes
OPs
Describe
This PR disabled oneDNN primitives caching by PaddlePaddle in favour of oneDNN its own caching mechanism. Also refactoring to Acquire API was made for concat