enable TF remapper optimizer #1418
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
TF supports a remapper optimizer which remaps subgraphs onto more efficient implementations by replacing commonly occuring subgraphs with optimized fused monolithic kernels. However, its support is limited: (1) MatMul + BiasAdd (not Add) + Activation; (2) Float32 (but not float64); (3) Activation is Tanh; (4) MKL is built and used. This commit replaces Add by BiasAdd in the NN. The speed of a single op can be improved by about 20% when TF is using MKL and precision is set to float32. One can find `_MklNativeFusedMatMul` op in the profiler. See also: - https://www.tensorflow.org/guide/graph_optimization - https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/grappler/optimizers/remapper.cc (cherry picked from commit 8f2dc44)
Codecov Report
@@ Coverage Diff @@
## devel #1418 +/- ##
==========================================
+ Coverage 74.55% 75.72% +1.17%
==========================================
Files 92 92
Lines 7623 7650 +27
==========================================
+ Hits 5683 5793 +110
+ Misses 1940 1857 -83
Continue to review full report at Codecov.
|
|
This PR only introduces The support for tanh + MKL is introduced in tensorflow/tensorflow#42173 (v2.4). I don't know why only FP32 is supported. |
|
I found that Intel OneDNN does not support fp64 at all... |
wanghan-iapcm
approved these changes
Jan 16, 2022
denghuilu
approved these changes
Jan 16, 2022
|
I have some other ideas: we can customize a remapper optimizer. TF has provided the interface. |
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Outdated suggestions cannot be applied.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.
TF supports a remapper optimizer which remaps subgraphs onto more efficient implementations by replacing commonly occuring subgraphs with optimized fused monolithic kernels. However, its support is limited: (1)
MatMul+BiasAdd(notAdd) + Activation; (2) Float32 (but not float64); (3) Activation is Tanh; (4) MKL is built and used.This commit replaces
AddbyBiasAddin the NN. The speed of a single op can be improved by about 20% when TF is using MKL and precision is set to float32. One can find_MklNativeFusedMatMulop in the profiler.Original graph. Ops include MklMatMul, AddV2, and Tanh.
New graph.
_MklNativeFusedMatMulis used here.See also:
(cherry picked from commit 8f2dc44)