[ez][inductor] add a few outer dimension reduction cases for LOAF #162028
Conversation
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/162028
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 3dcc3f9 with merge base 8171d60 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/inductor
module: inductor
topic: not user facing
This was referenced Sep 3, 2025
jansel
approved these changes
Sep 3, 2025
eellison
approved these changes
Sep 3, 2025
This was referenced Sep 4, 2025
|
@pytorchbot merge |
Merge startedYour 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 |
This was referenced Sep 5, 2025
pytorchmergebot
pushed a commit
that referenced
this pull request
Sep 6, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: #162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: #162028
pytorchmergebot
pushed a commit
that referenced
this pull request
Sep 6, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: #162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: #162028, #162221
pytorchmergebot
pushed a commit
that referenced
this pull request
Sep 7, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: #162311 Approved by: https://github.com/jansel ghstack dependencies: #162028, #162221, #162303
daisyden
pushed a commit
to daisyden/pytorch
that referenced
this pull request
Sep 8, 2025
…torch#162028) For the not able to fuse issue reported here: pytorch#93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue. Pull Request resolved: pytorch#162028 Approved by: https://github.com/jansel, https://github.com/eellison
daisyden
pushed a commit
to daisyden/pytorch
that referenced
this pull request
Sep 8, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: pytorch#162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: pytorch#162028
daisyden
pushed a commit
to daisyden/pytorch
that referenced
this pull request
Sep 8, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: pytorch#162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221
daisyden
pushed a commit
to daisyden/pytorch
that referenced
this pull request
Sep 8, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: pytorch#162311 Approved by: https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221, pytorch#162303
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
…torch#162028) For the not able to fuse issue reported here: pytorch#93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue. Pull Request resolved: pytorch#162028 Approved by: https://github.com/jansel, https://github.com/eellison
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: pytorch#162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: pytorch#162028
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: pytorch#162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: pytorch#162311 Approved by: https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221, pytorch#162303
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…torch#162028) For the not able to fuse issue reported here: pytorch#93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue. Pull Request resolved: pytorch#162028 Approved by: https://github.com/jansel, https://github.com/eellison
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: pytorch#162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: pytorch#162028
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: pytorch#162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: pytorch#162311 Approved by: https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221, pytorch#162303
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…torch#162028) For the not able to fuse issue reported here: pytorch#93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue. Pull Request resolved: pytorch#162028 Approved by: https://github.com/jansel, https://github.com/eellison
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: pytorch#162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: pytorch#162028
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: pytorch#162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: pytorch#162311 Approved by: https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221, pytorch#162303
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…torch#162028) For the not able to fuse issue reported here: pytorch#93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue. Pull Request resolved: pytorch#162028 Approved by: https://github.com/jansel, https://github.com/eellison
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
Make sure TemplateBuffer & ComputedBuffer have the same dependencies prefix. Pull Request resolved: pytorch#162221 Approved by: https://github.com/jansel, https://github.com/eellison ghstack dependencies: pytorch#162028
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
Skiping renaming cause wrong dependencies when mutations are involved. Test: CUDA_VISIBLE_DEVICES=4,5,6 TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/distributed/test_compute_comm_reordering.py TestComputeCommReorderingMultiProc.test_reorder_compute_for_overlap Both all-reduce and wait-tensor ir node contains a MutationBuffer for this test. Pull Request resolved: pytorch#162303 Approved by: https://github.com/eellison, https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
LOAF previously may skip these fusion opportunities and cause some tests fail. Test: - TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=1 python test/inductor/test_torchinductor_strided_blocks.py TritonBlockPointerTestGPU.test_2d_reduction_odd_shapes_view_size4_num_block_pointers_1_num_triton_kernels_1_reduction_op4_cuda Pull Request resolved: pytorch#162311 Approved by: https://github.com/jansel ghstack dependencies: pytorch#162028, pytorch#162221, pytorch#162303
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
Stack from ghstack (oldest at bottom):
For the not able to fuse issue reported here: #93718 , LOAF can fuse the outer dimension softmax into a single kernel and brings 1.87x speedup for the example shape mentioned in the issue.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben