[inductor][addmm] incorporate into new get_mm_configs properly#161534
Closed
coconutruben wants to merge 33 commits intogh/coconutruben/55/basefrom
Closed
[inductor][addmm] incorporate into new get_mm_configs properly#161534coconutruben wants to merge 33 commits intogh/coconutruben/55/basefrom
coconutruben wants to merge 33 commits intogh/coconutruben/55/basefrom
Conversation
\# why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic \# what - expand KernelInputs to also store views of specific nodes, by names - use that view (inp, the unexpanded version) in the heuristics to adjust it depending on whether we're in max-autotune or not \# testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` [ghstack-poisoned]
This was referenced Aug 26, 2025
This was referenced Aug 26, 2025
Contributor
Author
|
@coconutruben has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
Contributor
Author
|
@coconutruben has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
jansel
approved these changes
Sep 9, 2025
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
Contributor
Author
|
@coconutruben has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
jansel
requested changes
Sep 11, 2025
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
Contributor
Author
|
@coconutruben has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
coconutruben
added a commit
that referenced
this pull request
Sep 13, 2025
\# why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic \# what - remove the view from inp when not in max-autotune for addmm aten \# testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` ghstack-source-id: 4399549 Pull Request resolved: #161534
…erly" # why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic # what - pass unexpanded bias (inp) - let template (heuristics) that it to be expanded (ATen in not max-autotune, Triton always) expand it # testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov Differential Revision: [D81520581](https://our.internmc.facebook.com/intern/diff/D81520581) [ghstack-poisoned]
coconutruben
added a commit
that referenced
this pull request
Sep 13, 2025
\# why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic \# what - remove the view from inp when not in max-autotune for addmm aten \# testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` ghstack-source-id: 208c906 Pull Request resolved: #161534
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
Khanaksahu
pushed a commit
to Khanaksahu/pytorch-fork
that referenced
this pull request
Nov 17, 2025
\# why - addmm aten running with an expanded version of bias vs the regular bias sometimes causes numerics differences - to avoid this for now, we make addmm aten use inp vs inp_expanded depending on if we're in max-autotune or not, matching the previous logic \# what - expand KernelInputs to also store views of specific nodes, by names - use that view (inp, the unexpanded version) in the heuristics to adjust it depending on whether we're in max-autotune or not \# testing ``` python3 -bb -m pytest test/inductor/test_aot_inductor.py::AOTInductorTestABICompatibleCpu::test_aoti_debug_printer_codegen_cpu ``` ghstack-source-id: 1182f39 Pull Request resolved: pytorch/pytorch#161534
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
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):
why
bias sometimes causes numerics differences
depending on if we're in max-autotune or not, matching the previous
logic
what
testing
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @jataylo @chenyang78
Differential Revision: D81520581