[Inductor XPU GEMM] Step 1/N: Refactor cutlass configuration. by etaf · Pull Request #160174 · pytorch/pytorch

etaf · 2025-08-08T06:03:48Z

Stack from ghstack (oldest at bottom):

This PR is the first step toward implementing RFC #160175.
Currently, all Cutlass-related Torch Inductor configs are located in torch._inductor.config.cuda. This PR refactors the device-agnostic Cutlass configurations into torch._inductor.config.cutlass, so they can be shared and reused by XPU as well.
The common cutlass configurations are moved from class cuda to class cutlass and the cuda specific configs are still in class cuda.
Besides, to not break the BC, we need to keep the configs moved from class cuda to class cutlass, eg. cuda.cutlass_dir. So we add a wrapper @inherit_fields_from(cutlass) to class cuda and class xpu so they also have configs in class cutlass.

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @jataylo @mlazos @chenyang78

[ghstack-poisoned]

pytorch-bot · 2025-08-08T06:03:51Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/160174

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 654c7ac with merge base af6d994 ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

inductor / unit-test / inductor-test / test (inductor, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (disabled by #168207 but the issue was closed recently and a rebase is needed to make it pass)
test/inductor/test_torchinductor_opinfo.py::TestInductorOpInfoCUDA::test_comprehensive_linalg_lu_factor_ex_cuda_float32

This comment was automatically generated by Dr. CI and updates every 15 minutes.

ghstack-source-id: bc6d53a Pull Request resolved: #160174

…tion." This PR is the second step toward implementing RFC #160175. Currently, all Cutlass-related Torch Inductor configs are located in `torch._inductor.config.cuda`. This PR refactors the device-agnostic Cutlass configurations into a separate module, `torch._inductor.config.cutlass`, so they can be shared and reused by XPU as well. [ghstack-poisoned]

…on." This PR is the first step toward implementing RFC #160175. Currently, all Cutlass-related Torch Inductor configs are located in `torch._inductor.config.cuda`. This PR refactors the device-agnostic Cutlass configurations into `torch._inductor.config.cutlass`, so they can be shared and reused by XPU as well. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben mlazos [ghstack-poisoned]

Summary: Following up a previous change #160174 where we accidentally include cutlass path to config hash and regressed the cache hit rate, we want to have some basic test added to pytorch to guard against accidental changes to configs like this happening again in the future. One way to solve this is to add a fixture test which compare the latest config hash to an existing generated result from previous run. When developer need to update the config, it can simply be done by using ``` EXPECTTEST_ACCEPT=1 pytest test/test_torch_config_hash_fixture.py ``` Test Plan: pytest test/test_torch_config_hash_fixture.py Reviewers: Subscribers: Tasks: Tags:

Summary: Following up a previous change #160174 where we accidentally include cutlass path to config hash and regressed the cache hit rate, we want to have some basic test added to pytorch to guard against accidental changes to configs like this happening again in the future. One way to solve this is to add detection for common patterns like paths, username and hostname in the config values. Test Plan: pytest test/test_torch_config_hash_determinism.py Reviewers: Subscribers: Tasks: Tags:

…pytorch#160174)" This reverts commit eabb7ad. Reverted pytorch#160174 on behalf of https://github.com/huydhn due to Sorry for reverting your change, but it seems to cause a perf regression ([comment](pytorch#160174 (comment)))

Summary: Following up a previous change #160174 where we accidentally include cutlass path to config hash and regressed the cache hit rate, we want to have some basic test added to pytorch to guard against accidental changes to configs like this happening again in the future. One way to solve this is to add detection for common patterns like paths, username and hostname in the config values. Test Plan: pytest test/test_torch_config_hash_determinism.py Reviewers: Subscribers: Tasks: Tags: Fixes #ISSUE_NUMBER Pull Request resolved: #171275 Approved by: https://github.com/bobrenjc93, https://github.com/masnesral

ghstack-source-id: a5f925c Pull Request resolved: pytorch#160174

Summary: Following up a previous change pytorch#160174 where we accidentally include cutlass path to config hash and regressed the cache hit rate, we want to have some basic test added to pytorch to guard against accidental changes to configs like this happening again in the future. One way to solve this is to add detection for common patterns like paths, username and hostname in the config values. Test Plan: pytest test/test_torch_config_hash_determinism.py Reviewers: Subscribers: Tasks: Tags: Fixes #ISSUE_NUMBER Pull Request resolved: pytorch#171275 Approved by: https://github.com/bobrenjc93, https://github.com/masnesral

[ghstack-poisoned]

ghstack-source-id: fa78cb9 Pull Request resolved: pytorch#160174

ghstack-source-id: fa78cb9 Pull Request resolved: pytorch/pytorch#160174

mlazos · 2026-01-20T21:31:23Z

Hi @etaf, it's tricky and error prone to run this internally so we should land, and I can let you know if it passes internally. Sorry for the delay on this.

etaf · 2026-01-21T06:18:07Z

Hi @etaf, it's tricky and error prone to run this internally so we should land, and I can let you know if it passes internally. Sorry for the delay on this.

Thanks for your time!

etaf · 2026-01-27T00:46:04Z

@pytorchbot merge

pytorchmergebot · 2026-01-27T00:48:25Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

pytorchmergebot · 2026-01-27T00:48:58Z

Merge failed

Reason: Command git -C /home/runner/work/pytorch/pytorch cherry-pick -x 4f4c7a5f740ec736a0b4c8c2c6c51bb52735ecac returned non-zero exit code 1

Auto-merging torch/_inductor/codecache.py
Auto-merging torch/_inductor/config.py
Auto-merging torch/_inductor/select_algorithm.py
Auto-merging torch/_inductor/utils.py
CONFLICT (content): Merge conflict in torch/_inductor/utils.py
error: could not apply 4f4c7a5f740... [Inductor XPU GEMM] Step 1/N: Generalize cutlass configuration.
hint: After resolving the conflicts, mark them with
hint: "git add/rm <pathspec>", then run
hint: "git cherry-pick --continue".
hint: You can instead skip this commit with "git cherry-pick --skip".
hint: To abort and get back to the state before "git cherry-pick",
hint: run "git cherry-pick --abort".
hint: Disable this message with "git config set advice.mergeConflict false"

Details for Dev Infra team

Raised by workflow job

…on." This PR is the first step toward implementing RFC #160175. Currently, all Cutlass-related Torch Inductor configs are located in `torch._inductor.config.cuda`. This PR refactors the device-agnostic Cutlass configurations into `torch._inductor.config.cutlass`, so they can be shared and reused by XPU as well. The common cutlass configurations are moved from `class cuda` to `class cutlass` and the cuda specific configs are still in `class cuda`. Besides, to not break the BC, we need to keep the configs moved from `class cuda` to `class cutlass`, eg. cuda.cutlass_dir. So we add a wrapper `inherit_fields_from(cutlass)` to `class cuda` and `class xpu` so they also have configs in `class cutlass`. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy kadeng muchulee8 amjames chauhang aakhundov coconutruben jataylo mlazos chenyang78 [ghstack-poisoned]

etaf · 2026-01-27T08:55:36Z

@pytorchbot merge

pytorchmergebot · 2026-01-27T08:57:34Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

…h#160174) This PR is the first step toward implementing RFC pytorch#160175. Currently, all Cutlass-related Torch Inductor configs are located in `torch._inductor.config.cuda`. This PR refactors the device-agnostic Cutlass configurations into `torch._inductor.config.cutlass`, so they can be shared and reused by XPU as well. The common cutlass configurations are moved from `class cuda` to `class cutlass` and the cuda specific configs are still in `class cuda`. Besides, to not break the BC, we need to keep the configs moved from `class cuda` to `class cutlass`, eg. cuda.cutlass_dir. So we add a wrapper `@inherit_fields_from(cutlass)` to `class cuda` and `class xpu` so they also have configs in `class cutlass`. Pull Request resolved: pytorch#160174 Approved by: https://github.com/EikanWang, https://github.com/mlazos, https://github.com/jansel

Shan19900305 · 2026-02-09T02:21:00Z

@etaf Will this commit continue to be merged?

etaf · 2026-02-09T02:25:49Z

@etaf Will this commit continue to be merged?

It has been merged. And the other PR in the stack will also be merged after get review approval.

[Inductor Intel Cutlass] Step 2/N: Generalize cutlass configuration.

44789dc

[ghstack-poisoned]

etaf mentioned this pull request Aug 8, 2025

[Inductor XPU GEMM] Step 1/N: Add cutlass-sycl repro. #160173

Closed

pytorch-bot bot added ciflow/inductor module: inductor labels Aug 8, 2025

etaf added a commit that referenced this pull request Aug 8, 2025

[Inductor Intel Cutlass] Step 2/N: Generalize cutlass configuration.

35cfe82

ghstack-source-id: bc6d53a Pull Request resolved: #160174

etaf marked this pull request as draft August 8, 2025 06:04

pytorchbot added the open source label Aug 8, 2025

etaf mentioned this pull request Aug 8, 2025

[RFC] Enable cutlass to support Intel GPU into PyTorch Inductor. #160175

Open

11 tasks

etaf requested a review from EikanWang August 8, 2025 06:34

EikanWang changed the title ~~[Inductor Intel Cutlass] Step 2/N: Generalize cutlass configuration.~~ [Inductor XPU GEMM] Step 2/N: Generalize cutlass configuration. Aug 8, 2025

etaf mentioned this pull request Aug 15, 2025

[Inductor XPU GEMM] Step 6/N: Refactor CUDACodeCache. #160706

Closed

etaf mentioned this pull request Aug 15, 2025

[Inductor XPU GEMM] Step 7/N: Refactor CUDABenchmarkRequest #160729

Closed

etaf changed the title ~~[Inductor XPU GEMM] Step 2/N: Generalize cutlass configuration.~~ [Inductor XPU GEMM] Step 1/N: Refactor cutlass configuration. Sep 2, 2025

etaf marked this pull request as ready for review September 2, 2025 03:11

etaf added the release notes: inductor label Sep 2, 2025

etaf added 3 commits September 2, 2025 10:58

etaf added a commit to etaf/pytorch-inductor-xpu that referenced this pull request Jan 12, 2026

[Inductor XPU GEMM] Step 1/N: Generalize cutlass configuration.

c4d2d30

ghstack-source-id: a5f925c Pull Request resolved: pytorch#160174

atalman mentioned this pull request Jan 15, 2026

Release 2.10 validations checklist and cherry-picks #172576

Closed

Update

a1b8947

[ghstack-poisoned]

etaf added a commit to etaf/pytorch-inductor-xpu that referenced this pull request Jan 16, 2026

[Inductor XPU GEMM] Step 1/N: Generalize cutlass configuration.

1fc8603

ghstack-source-id: fa78cb9 Pull Request resolved: pytorch#160174

SergeyTyshkevich pushed a commit to SergeyTyshkevich/chart2 that referenced this pull request Jan 19, 2026

[Inductor XPU GEMM] Step 1/N: Generalize cutlass configuration.

4f4c7a5

ghstack-source-id: fa78cb9 Pull Request resolved: pytorch/pytorch#160174

pytorchmergebot added the merging label Jan 27, 2026

pytorchmergebot removed the merging label Jan 27, 2026

pytorchmergebot added the merging label Jan 27, 2026

pytorchmergebot added the Merged label Jan 27, 2026

pytorchmergebot closed this in b98f4a3 Jan 27, 2026

pytorchmergebot removed the merging label Jan 27, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Inductor XPU GEMM] Step 1/N: Refactor cutlass configuration.#160174

[Inductor XPU GEMM] Step 1/N: Refactor cutlass configuration.#160174
etaf wants to merge 37 commits intogh/etaf/154/basefrom
gh/etaf/154/head

etaf commented Aug 8, 2025 •

edited

Loading

Uh oh!

pytorch-bot bot commented Aug 8, 2025 •

edited

Loading

Uh oh!

mlazos commented Jan 20, 2026

Uh oh!

etaf commented Jan 21, 2026

Uh oh!

etaf commented Jan 27, 2026

Uh oh!

pytorchmergebot commented Jan 27, 2026

Uh oh!

pytorchmergebot commented Jan 27, 2026

Uh oh!

etaf commented Jan 27, 2026

Uh oh!

pytorchmergebot commented Jan 27, 2026

Uh oh!

Shan19900305 commented Feb 9, 2026

Uh oh!

etaf commented Feb 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

Conversation

etaf commented Aug 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Aug 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/160174

✅ You can merge normally! (1 Unrelated Failure)

Uh oh!

mlazos commented Jan 20, 2026

Uh oh!

etaf commented Jan 21, 2026

Uh oh!

etaf commented Jan 27, 2026

Uh oh!

pytorchmergebot commented Jan 27, 2026

Merge started

Uh oh!

pytorchmergebot commented Jan 27, 2026

Merge failed

Uh oh!

etaf commented Jan 27, 2026

Uh oh!

pytorchmergebot commented Jan 27, 2026

Merge started

Uh oh!

Shan19900305 commented Feb 9, 2026

Uh oh!

etaf commented Feb 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

etaf commented Aug 8, 2025 •

edited

Loading

pytorch-bot bot commented Aug 8, 2025 •

edited

Loading