【KernelGen】Add cudnn_convolution operator - #1729
Merged
w1120029931-bit merged 13 commits intoMay 14, 2026
Merged
Conversation
Schopenhauer-loves-Hegel
requested review from
0x45f,
huangyiqun,
kiddyjinjin and
zhangpeiyang1
as code owners
February 27, 2026 06:13
Collaborator
Author
|
Closing in favor of grouped PRs to reduce review overhead. |
Schopenhauer-loves-Hegel
changed the title
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
April 15, 2026 20:13
c399d5d to
a852dc8
Compare
|
|
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
April 17, 2026 13:22
a852dc8 to
e8c2058
Compare
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
April 21, 2026 10:26
6d3e9af to
c1b4c93
Compare
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
April 23, 2026 03:41
c1b4c93 to
1dc60b7
Compare
tengqm
previously approved these changes
Apr 23, 2026
factnn
requested review from
bin913,
douxetpur and
w1120029931-bit
as code owners
April 24, 2026 05:58
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
April 24, 2026 10:58
701142d to
194e23a
Compare
tengqm
reviewed
Apr 30, 2026
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
May 9, 2026 04:35
6693478 to
6c4621a
Compare
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
May 11, 2026 09:51
646dc0f to
a88a08e
Compare
tengqm
reviewed
May 11, 2026
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
May 11, 2026 14:49
a88a08e to
25e9596
Compare
w1120029931-bit
requested changes
May 12, 2026
w1120029931-bit
left a comment
w1120029931-bit
left a comment
Collaborator
There was a problem hiding this comment.
please change file test_cudnn_convolution_perf.py to file test_cudnn_convolution.py
Includes operators.yaml entry in alphabetical order. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Qiming Teng <tengqm@outlook.com> Signed-off-by: Zang Peiyu <166481866+factnn@users.noreply.github.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
torch.cudnn_convolution only supports CUDA backend, cannot run on CPU. Use the same GPU tensor for both reference and result. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
torch.cudnn_convolution only runs on CUDA, so ref stays on GPU. gems_assert_close requires ref to be on CPU when TO_CPU is enabled. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
torch.cudnn_convolution is CUDA-only, so both tensors stay on GPU. gems_assert_close requires both on same device; move both to CPU. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
factnn
force-pushed
the
auto-gen/cudnn_convolution
branch
from
May 12, 2026 13:58
f8bc5f1 to
8924dc2
Compare
…tors.yaml Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
w1120029931-bit
approved these changes
May 14, 2026
tengqm
added a commit
that referenced
this pull request
May 17, 2026
* Add cudnn_convolution operator implementation, tests and benchmark Includes operators.yaml entry in alphabetical order. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Apply suggestions from code review Co-authored-by: Qiming Teng <tengqm@outlook.com> Signed-off-by: Zang Peiyu <166481866+factnn@users.noreply.github.com> * fix: apply black formatting to test_cudnn_convolution.py Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: remove to_reference for cudnn_convolution tests (CUDA-only op) torch.cudnn_convolution only supports CUDA backend, cannot run on CPU. Use the same GPU tensor for both reference and result. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: rename benchmark file to remove _perf suffix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: move ref tensor to CPU before gems_assert_close torch.cudnn_convolution only runs on CUDA, so ref stays on GPU. gems_assert_close requires ref to be on CPU when TO_CPU is enabled. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: move both res and ref to CPU before comparison torch.cudnn_convolution is CUDA-only, so both tensors stay on GPU. gems_assert_close requires both on same device; move both to CPU. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: move cudnn_convolution to correct alphabetical position in operators.yaml Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Signed-off-by: Zang Peiyu <166481866+factnn@users.noreply.github.com> Co-authored-by: factnn <1050552884@qq.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Zang Peiyu <166481866+factnn@users.noreply.github.com> Co-authored-by: Qiming Teng <tengqm@outlook.com>
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
5 participants
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 Category
Operator
Type of Change
New Feature
Description
Add
cudnn_convolutionoperator implementation with Triton kernel.N/AIssue
N/A
Progress
Performance
Benchmark data not available.
Generated by auto_gen tool with Claude Code