cpp_wrapper: Precompile device-specific header files #146928
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/146928
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit ef90af2 with merge base c73c72b ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@desertfire has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@desertfire has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
This saves us about a second per compilation, which is _massive_ for the OpInfo tests. Total OpInfo test runtime is down about 2x from this change alone. Relands #144002, with changes needed by fbcode internals. ghstack-source-id: 51921ab Pull Request resolved: pytorch/pytorch#146928
|
@desertfire has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
I've spent a while wrestling with it, and I think I'm going to give up and disable precompilation of headers on Windows. That's not our main platform for testing anyway, and we get the majority of our benefit from this when running tests. The reason why I'm going this way is that MSVC deals with precompiled headers weirdly, where you have to both use the precompiled header and link in a resulting object file, and cpp_builder is currently pretty tied to the model that one build == one target file, not two. |
|
@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 saves us about a second per compilation, which is _massive_ for the OpInfo tests. Total OpInfo test runtime is down about 2x from this change alone. Relands #144002, with changes needed by fbcode internals. ghstack-source-id: efdd6b6 Pull Request resolved: pytorch/pytorch#146928
This saves us about a second per compilation, which is _massive_ for the OpInfo tests. Total OpInfo test runtime is down about 2x from this change alone. Relands #144002, with changes needed by fbcode internals. ghstack-source-id: aa55832 Pull Request resolved: pytorch/pytorch#146928
Stack from ghstack (oldest at bottom):
This saves us about a second per compilation, which is massive for the OpInfo tests. Total OpInfo test runtime is down about 2x from this change alone.
Relands #144002, with changes needed by fbcode internals.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov