Repository navigation
feat(hipdnn): add GPU batchnorm backward reference - #12520
Merged
BalintCsala merged 10 commits intoOct 9, 2026
Merged
BalintCsala merged 10 commits into
BalintCsala merged 10 commits into
Conversation
BalintCsala
added
organization: streamhpc
✅ All Checks Passed — Ready for Review
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
|
🎉 All checks passed! This PR is ready for review. |
BalintCsala
force-pushed
the
users/BalintCsala/batchnorm-backward-gpu-ref
branch
from
September 24, 2026 10:43
8e3935f to
446477c
Compare
brentmaas
requested review from
saikubairkota
and
a balanced review from Copilot
September 24, 2026 12:35
Contributor
There was a problem hiding this comment.
Copilot review overview
🟢 Approval recommended
The only unresolved finding is a non-blocking test-coverage nit.
Review effort: Balanced
Findings: None
What changed in this PR
Adds a HipRTC GPU reference implementation for batch normalization backward, supporting FP32, FP16, BFP16, and saved or calculated statistics.
Changes:
- Adds the backward kernel, launch API, arguments, and validation.
- Adds CPU-comparison integration tests and build registration.
- Supports multiple tensor ranks and layouts.
| File | Description |
|---|---|
dnn-providers/integration-tests/tests/gpu-ref/TestGpuFpReferenceBatchnormBwd.cpp |
Adds CPU/GPU correctness tests; larger calculated-statistics coverage is suggested. |
dnn-providers/integration-tests/tests/CMakeLists.txt |
Registers the new tests. |
dnn-providers/integration-tests/gpu-ref/src/GpuFpReferenceBatchnorm.cpp |
Implements kernel compilation and launch. |
dnn-providers/integration-tests/gpu-ref/kernels/types/GpuRefBatchnormArgs.h |
Defines backward kernel arguments. |
dnn-providers/integration-tests/gpu-ref/kernels/CMakeLists.txt |
Packages the new kernel. |
dnn-providers/integration-tests/gpu-ref/kernels/batchnorm/GpuRefBatchnormBwd.cpp |
Implements backward calculations and reductions. |
dnn-providers/integration-tests/gpu-ref/include/hipdnn-gpu-ref/GpuFpReferenceBatchnorm.hpp |
Adds the backward API and validation. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
adickin-amd
reviewed
Sep 24, 2026
adickin-amd
reviewed
Sep 24, 2026
adickin-amd
reviewed
Sep 24, 2026
adickin-amd
reviewed
Sep 24, 2026
BalintCsala
force-pushed
the
users/BalintCsala/batchnorm-backward-gpu-ref
branch
from
September 28, 2026 14:44
a33924e to
90cc000
Compare
BalintCsala
force-pushed
the
users/BalintCsala/batchnorm-backward-gpu-ref
branch
from
September 30, 2026 15:05
2b977f9 to
2cfcd6b
Compare
saikubairkota
approved these changes
Oct 2, 2026
Pre-commit check failed⛔ pre-commit failed Please run locally:
This repo uses |
BalintCsala
force-pushed
the
users/BalintCsala/batchnorm-backward-gpu-ref
branch
from
October 5, 2026 14:49
49f78d7 to
6211a09
Compare
BalintCsala
force-pushed
the
users/BalintCsala/batchnorm-backward-gpu-ref
branch
from
October 5, 2026 17:05
6211a09 to
a78b447
Compare
adickin-amd
reviewed
Oct 5, 2026
adickin-amd
reviewed
Oct 5, 2026
adickin-amd
approved these changes
Oct 5, 2026
adickin-amd
left a comment
adickin-amd
left a comment
Contributor
There was a problem hiding this comment.
A few things to look at but overall LGTM
BalintCsala
enabled auto-merge (squash)
October 6, 2026 09:12
1 task done
Contributor
|
Cancelling the remaining gfx950 jobs on this run so the TheRock submodule bump PRs can be prioritized on the gfx950 queue. This change does not target gfx950, and gfx950 was the only work still outstanding. Jobs that already finished are unchanged. Re-run failed jobs if you need gfx950 results before merge. |
BalintCsala
disabled auto-merge
October 7, 2026 11:29
BalintCsala
enabled auto-merge (squash)
October 7, 2026 14:45
Contributor
TheRock Submodule Bump ActivityNewest first
|
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.
Motivation
We need a reference batchnorm backward implementation on the GPU for faster tests.
Closes #12519
Technical Details
Kernel is compiled with HipRTC and supports fp32, fp16 and bfp16.
Test Plan
Tested against the CPU implementation.
Test Result
Tests passes locally.
Submission Checklist