[hipDNN] Fix backward activation reference input order#7736
Open
SamuelReeder wants to merge 1 commit into
Open
[hipDNN] Fix backward activation reference input order#7736SamuelReeder wants to merge 1 commit into
SamuelReeder wants to merge 1 commit into
Conversation
ece9af8 to
2dc7dd9
Compare
adickin-amd
approved these changes
May 25, 2026
CMiservaAMD
approved these changes
May 25, 2026
2dc7dd9 to
1d687ab
Compare
1d687ab to
1d8f4aa
Compare
Contributor
Author
|
Editing MIOpen provider too... |
1d8f4aa to
37a4b6e
Compare
f8f89b4 to
3efd97c
Compare
CMiservaAMD
approved these changes
May 25, 2026
3efd97c to
21d16fe
Compare
EwanC
approved these changes
May 26, 2026
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR fixes hipDNN backward activation reference and graph construction to consume upstream gradient
dybefore forward inputx. It updates direct pointwise, CPU graph executor, flatbuffer graph utility, hip-kernel-provider and MIOpen-provider fused batchnorm handling, and fused BN inference + DReLU + BN backward sample call sites so backward activation input ordering is consistent.Risk Assessment
Low risk. This is a narrow CPU reference/test SDK/sample/provider ordering fix with targeted local coverage and no public API, schema, or build configuration changes; PR CI is pending.
Testing Summary
hipdnn_test_sdk_teststo verify affected test SDK sources compile after the ordering change.Testing Checklist
cmake --build build --target hipdnn_test_sdk_tests- Status: Passedcmake --build build- Status: Passedcmake --build build-hipkernel-pr --target hip_kernel_provider_tests hip_kernel_provider_integration_tests- Status: Passedcmake --build build-miopen-pr --target miopen_plugin_tests miopen_plugin_integration_tests- Status: Passed./build-hipkernel-pr/bin/hip_kernel_provider_tests- Status: Passed./build-hipkernel-pr/bin/hip_kernel_provider_tests --gtest_filter='TestBatchnormPlanBuilder.IsApplicableReturnsTrueForValidThreeNodeGraph:TestBatchnormPlanBuilder.BuildPlanSetsPlanForFusedBackwardGraph:TestBatchnormBwdParams.InitializesFusedActivationWithAllTensors:TestBatchnormValidator.ValidInferenceActivationBackward'- Status: Passed./build-miopen-pr/bin/miopen_plugin_tests --gtest_filter='*FusedThreeNodeGraph*:*FusedGraph*:*BatchnormBwdParams*Fused*:*FusedMode*:*FusedBackward*:*ActivationBackward*:*ActivationMissingIn1*:*ActivationOutputDoesNotMatchBnBackwardDy*:*BnInferenceOutputDoesNotMatchActivationInput*'- Status: Passed./build/bin/hipdnn_test_sdk_tests --gtest_filter='*PointwiseSignatureKey*:*PlanRegistrySignatureKey*:*ReluBackward*:*ParameterizedReluBackward*:*SigmoidBackward*:*TanhBackward*'- Status: Passedctest --test-dir build -R hipdnn_sample_fused_bn_inference_drelu_bn_backward --output-on-failure- Status: Passedctest --test-dir build --output-on-failure --parallel 8- Status: Passedclang-format-18 --dry-run -Werror <changed files>- Status: Passedgit diff --check -- <changed files>- Status: PassedTechnical Changes
ReluBackward,ParameterizedReluBackward,SigmoidBackward, andTanhBackwardfunctor signatures to takedybeforexwhile preserving downstream gradient calculations.