Dispatch to mv rather than mm in the case that tensor1.ndim == 1 and tensor2.ndim == 2 by lezcano · Pull Request #75195 · pytorch/pytorch

lezcano · 2022-04-04T16:21:41Z

Stack from ghstack:

This should hopefully be faster, it makes the calling code simpler, and
it solves a bug when using matmul with the out= parameter (before it
would throw an incorrect error).

…tensor2.ndim == 2 This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). [ghstack-poisoned]

facebook-github-bot · 2022-04-04T16:21:51Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/75195
📄 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 our office hours

💊 CI failures summary and remediations

As of commit debf0bf (more details on the Dr. CI page):

Expand to see more

1/1 failures introduced in this PR

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages

pull / pytorch-xla-linux-bionic-py3.7-clang8 / test (xla, 1, 1, linux.2xlarge) (1/1)

Step: "Test" (full log | diagnosis details | 🔁 rerun)

2022-05-04T20:36:24.5572937Z �[0;31m[ FAILED ] �[mAtenXlaTensorTest.TestMatmul_1x2

2022-05-04T20:36:24.5513452Z �[0;32m[ RUN      ] �[mXlaUtilCacheTest.BasicTest
2022-05-04T20:36:24.5514102Z �[0;32m[       OK ] �[mXlaUtilCacheTest.BasicTest (0 ms)
2022-05-04T20:36:24.5517062Z �[0;32m[----------] �[m1 test from XlaUtilCacheTest (0 ms total)
2022-05-04T20:36:24.5517392Z 
2022-05-04T20:36:24.5570100Z �[0;32m[----------] �[mGlobal test environment tear-down
2022-05-04T20:36:24.5570786Z �[0;32m[==========] �[m623 tests from 8 test suites ran. (512555 ms total)
2022-05-04T20:36:24.5571192Z �[0;32m[  PASSED  ] �[m621 tests.
2022-05-04T20:36:24.5571554Z �[0;32m[  SKIPPED ] �[m1 test, listed below:
2022-05-04T20:36:24.5572040Z �[0;32m[  SKIPPED ] �[mAtenXlaTensorTest.TestGroupNormBackward
2022-05-04T20:36:24.5572501Z �[0;31m[  FAILED  ] �[m1 test, listed below:
2022-05-04T20:36:24.5572937Z �[0;31m[  FAILED  ] �[mAtenXlaTensorTest.TestMatmul_1x2
2022-05-04T20:36:24.5573171Z 
2022-05-04T20:36:24.5573268Z  1 FAILED TEST
2022-05-04T20:36:24.7276308Z + cleanup
2022-05-04T20:36:24.7276526Z + retcode=1
2022-05-04T20:36:24.7276685Z + set +x
2022-05-04T20:36:24.7569431Z ##[error]Process completed with exit code 1.
2022-05-04T20:36:24.7765995Z ##[group]Run pytorch/pytorch/.github/actions/get-workflow-job-id@master
2022-05-04T20:36:24.7766245Z with:
2022-05-04T20:36:24.7766611Z   github-token: ***
2022-05-04T20:36:24.7766792Z env:

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

…m == 1 and tensor2.ndim == 2" This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). [ghstack-poisoned]

lezcano · 2022-04-04T20:35:23Z

@ezyang This PR shows (or will show, once the CI of the previous PRs is green) that this is the line that's causing trouble.

We get this error: https://github.com/pytorch/pytorch/runs/5820894454?check_suite_focus=true in the distributed build. I have tried to track down what could be causing it, and following the execution of the trace that it throws, I don't see how that state could be reached. It's also particularly odd that this just happens in some random test in a distributed setting...

…m == 1 and tensor2.ndim == 2" This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). [ghstack-poisoned]

ezyang · 2022-04-05T00:03:30Z

Can you repro it locally?

I looked over the patch and it looks... fine-ish? The contiguous changes look a little skeevy

lezcano · 2022-04-05T02:37:09Z

I could not reproduce it locally. This PR should just be one line. I pushed some further changes to start debugging it via CI, but this is certainly not the best approach.

Once we had access to sshing into the jobs, but J believe that's not working atm...

…m == 1 and tensor2.ndim == 2" This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). [ghstack-poisoned]

ngimel · 2022-05-04T23:20:25Z

@pytorchbot merge this

…tensor2.ndim == 2 (#75195) Summary: This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). Pull Request resolved: #75195 Approved by: https://github.com/ezyang Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/4baf7c0899a2fa9c3630613f37d5fc65971db21c Reviewed By: malfet Differential Revision: D36171145 fbshipit-source-id: f6054f25940e6a8ad9de0e7576def098b15e9c3f

…= 1 and tensor2.ndim == 2 (#75195)" This reverts commit d024d26. ghstack-source-id: d5daa0c Pull Request resolved: #82484

…= 1 and tensor2.ndim == 2 (#75195)" This reverts commit d024d26. [ghstack-poisoned]

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Why is this a partial revert? - the out= tests for nn.functional.linear fail on a complete revert - the profiler tests fail on the revert (so I updated the expecttests for the profiler tests) Test Plan: - test offline that the functorch regression was fixed [ghstack-poisoned]

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Why is this a partial revert? - the out= tests for nn.functional.linear fail on a complete revert - the profiler tests fail on the revert (so I updated the expecttests for the profiler tests) Test Plan: - test offline that the functorch regression was fixed ghstack-source-id: 1332851 Pull Request resolved: #82504

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed [ghstack-poisoned]

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed ghstack-source-id: ff73cb6 Pull Request resolved: #82504

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed [ghstack-poisoned]

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed ghstack-source-id: f127f9b Pull Request resolved: #82504

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed [ghstack-poisoned]

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed ghstack-source-id: ea72d0f Pull Request resolved: #82504

This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Test Plan: - test offline that the functorch regression was fixed Pull Request resolved: #82504 Approved by: https://github.com/ngimel, https://github.com/ezyang, https://github.com/atalman

Summary: This is a short-term fix for a serious regression in functorch (pytorch/functorch#989). Additional things this PR does: - the out= tests for nn.functional.linear fail after the revert. I added some xfails. These xfails were present in the original PR (#75195). - the profiler tests fail on the revert, so I updated the expecttests for the profiler tests Pull Request resolved: #82504 Approved by: https://github.com/ngimel, https://github.com/ezyang, https://github.com/atalman Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/8f86e361918e3c8a0ee2b569be6c82dfbf32d705 Test plan from GitHub: - test offline that the functorch regression was fixed Reviewed By: kit1980 Differential Revision: D38394573 Pulled By: zou3519 fbshipit-source-id: f9185d9cb447fb439d8e402712f2f2617f73b8cc

Dispatch to mv rather than mm in the case that tensor1.ndim == 1 and …

2657049

…tensor2.ndim == 2 This should hopefully be faster, it makes the calling code simpler, and it solves a bug when using matmul with the out= parameter (before it would throw an incorrect error). [ghstack-poisoned]

lezcano requested review from IvanYashchuk and nikitaved as code owners April 4, 2022 16:21

facebook-github-bot added the cla signed label Apr 4, 2022

This was referenced Apr 4, 2022

Improved matmul tests #75193

Closed

Refactor the API of the matmul implementation #75194

Closed

Go through the dispatcher in matmul_out for the 1D-1D case #75196

Closed

Micro-optimisations for matmul 2.0: Electric boogaloo #75197

Closed

lezcano removed request for IvanYashchuk and nikitaved April 4, 2022 16:24

lezcano added module: linear algebra Issues related to specialized linear algebra operations in PyTorch; includes matrix multiply matmul release notes: performance_as_product topic: performance topic category labels Apr 4, 2022

pytorchbot added the open source label Apr 4, 2022

lezcano mentioned this pull request Apr 4, 2022

Fix addmm_cpu for int64 #75200

Closed

lezcano mentioned this pull request Apr 4, 2022

Make mv and addmv support torch.float16 #75220

Closed

lezcano requested a review from ezyang April 4, 2022 20:32

lezcano added the with-ssh label Apr 4, 2022

lezcano added 2 commits April 5, 2022 18:03

lezcano mentioned this pull request Apr 5, 2022

Fix mv/addmv on CUDA when dealing with vectors of size=1 and stride=0 #75279

Closed

ezyang approved these changes Apr 6, 2022

View reviewed changes

lezcano mentioned this pull request May 4, 2022

Avoid copies in matmul #76828

Closed

pytorchmergebot closed this in 4baf7c0 May 4, 2022

facebook-github-bot deleted the gh/Lezcano/57/head branch May 8, 2022 14:16

zou3519 mentioned this pull request Jul 29, 2022

25% Performance regression from v0.1.1 to 0.2.0 when calculating hessian pytorch/functorch#989

Open

zou3519 added a commit that referenced this pull request Jul 29, 2022

Revert "Dispatch to mv rather than mm in the case that tensor1.ndim =…

c5e8548

…= 1 and tensor2.ndim == 2 (#75195)" This reverts commit d024d26. ghstack-source-id: d5daa0c Pull Request resolved: #82484

zou3519 added a commit that referenced this pull request Jul 29, 2022

Revert "Dispatch to mv rather than mm in the case that tensor1.ndim =…

4e81b6c

…= 1 and tensor2.ndim == 2 (#75195)" This reverts commit d024d26. [ghstack-poisoned]

zou3519 mentioned this pull request Aug 1, 2022

Revert #75195 #82504

Closed

zou3519 mentioned this pull request Aug 2, 2022

Revert #75195 (#82504) #82662

Merged

lezcano mentioned this pull request Feb 22, 2023

Don't create large intermediary tensors in the backward of matmul #95261

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Dispatch to mv rather than mm in the case that tensor1.ndim == 1 and tensor2.ndim == 2#75195

Dispatch to mv rather than mm in the case that tensor1.ndim == 1 and tensor2.ndim == 2#75195
lezcano wants to merge 10 commits intogh/Lezcano/57/basefrom
gh/Lezcano/57/head

lezcano commented Apr 4, 2022 •

edited

Loading

Uh oh!

facebook-github-bot commented Apr 4, 2022 •

edited

Loading

🕵️ 1 new failure recognized by patterns

pull / pytorch-xla-linux-bionic-py3.7-clang8 / test (xla, 1, 1, linux.2xlarge) (1/1)

Uh oh!

lezcano commented Apr 4, 2022

Uh oh!

ezyang commented Apr 5, 2022

Uh oh!

lezcano commented Apr 5, 2022

Uh oh!

ngimel commented May 4, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

Conversation

lezcano commented Apr 4, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Apr 4, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

🕵️ 1 new failure recognized by patterns

pull / pytorch-xla-linux-bionic-py3.7-clang8 / test (xla, 1, 1, linux.2xlarge) (1/1)

Uh oh!

lezcano commented Apr 4, 2022

Uh oh!

ezyang commented Apr 5, 2022

Uh oh!

lezcano commented Apr 5, 2022

Uh oh!

ngimel commented May 4, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

lezcano commented Apr 4, 2022 •

edited

Loading

facebook-github-bot commented Apr 4, 2022 •

edited

Loading