Allow CUDA extension builds to skip generating cuda dependencies during compile time (#119936)
nvcc flag `--generate-dependencies-with-compile` doesn't seem to be supported by `sccache` for now. Builds with this flag enabled will not benefit from sccache.
This PR adds an environment variable that allows users to set this flag and skip those nvcc dependencies to speed up their build with compiler caches. If everything is "fresh build" in CI, we don't care if there are unnecessary recompile during incremental builds.
related: https://github.com/pytorch/pytorch/pull/49344
- [ ] todo: raise an issue to sccache
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119936
Approved by: https://github.com/ezyang
diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
index 46d8776..c5c62f2 100644
--- a/torch/utils/cpp_extension.py
+++ b/torch/utils/cpp_extension.py
@@ -2342,7 +2342,8 @@
cuda_compile_rule = ['rule cuda_compile']
nvcc_gendeps = ''
# --generate-dependencies-with-compile is not supported by ROCm
- if torch.version.cuda is not None:
+ # Nvcc flag `--generate-dependencies-with-compile` is not supported by sccache, which may increase build time.
+ if torch.version.cuda is not None and os.getenv('TORCH_EXTENSION_SKIP_NVCC_GEN_DEPENDENCIES', '0') != '1':
cuda_compile_rule.append(' depfile = $out.d')
cuda_compile_rule.append(' deps = gcc')
# Note: non-system deps with nvcc are only supported
