[NVFUSER] refactor nvfuser build#89621
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/89621
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 74e53e1: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
Looks like there's some conflicts now, should I bump this PR to upstream/master? |
|
we can probably wait until next week unless you suspect there will be any new failures that will pop up |
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
1 similar comment
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
Looks like I got a bad upstream commit. I'll keep an eye on the CI hud and grab a clean commit when it gets green again. |
|
recommend using the |
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
Looks like the build failure has been patched.. Sorry that I forgot to update c++14 flag... 😛 |
build_variables.bzl
Outdated
| "torch/csrc/jit/codegen/cuda/runtime/warp.cu", | ||
| "torch/csrc/jit/codegen/cuda/runtime/warp_rocm.cu", | ||
| "torch/csrc/jit/codegen/cuda/runtime/welford.cu", | ||
| "nvfuser/runtime/array.cu", |
There was a problem hiding this comment.
@jjsjann123 i think these need to be updated to third_party/nvfuser ?
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
windows CI seems strangely flaky. |
|
hmmm. can't seem to get the flaky CI to pass... I'll keep trying. |
|
hmm. looks like upstream/viable/strict shows failing tests (inductor) in HUD log. I tried to grab one that at least looks green across the column. 🤞 |
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
I think I need to bump this for import/land, but HUD showing all later commits with some failures (even the commit pointed by |
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
1 similar comment
|
@davidberard98 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
davidberard98
left a comment
There was a problem hiding this comment.
stamp so I can attempt to land internally (don't use pytorchbot to merge this)
So excited!!! 🥳 |
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
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 PR is the first step towards refactors the build for nvfuser in order to have the coegen being a standalone library.
Contents inside this PR:
./nvfuser, from./torch/csrc/jit/codegen/cuda/, except for registration code for integration (interface.h/interface.cpp).sofiles. Currently there are:libnvfuser_codegen.so, which contains the integration, codegen and runtime system of nvfusernvfuser.so, which is nvfuser's python API via pybind. Python frontend is now exposed vianvfuser._C.XXXinstead oftorch._C._nvfusernvfuser_testsCMakeLists.txt, which is undertorch/csrc/jit/codegen/cuda/.libtorch_cuda_xxxany more./CMakeLists.txtat the very end after torch is built.at::DynamicLibrary). This avoids circular dependency in cmake, which will be a nightmare to handle. For details, look attorch/csrc/jit/codegen/cuda/interface.cpp::LoadingNvfuserLibraryFuture work that's scoped in following PR:
cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @EikanWang @kevinstephano @mlazos @soumith @voznesenskym @yanboliang @penguinwu @anijain2305 @Guobing-Chen @chunyuan-w @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @desertfire