Skip to content

Respect TORCH_CUDA_ARCH_LIST to speed up builds #100

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
d3banjan wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Respect TORCH_CUDA_ARCH_LIST to speed up builds #100

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
51 changes: 33 additions & 18 deletions setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -173,27 +173,42 @@ def append_nvcc_threads(nvcc_extra_args):
"Note: make sure nvcc has a supported version by running nvcc -V."
)

cc_flag.append("-gencode")
cc_flag.append("arch=compute_75,code=sm_75")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_80,code=sm_80")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_87,code=sm_87")
if bare_metal_version >= Version("11.8"):
cc_flag.append("-gencode")
cc_flag.append("arch=compute_90,code=sm_90")
if bare_metal_version >= Version("12.8"):
cc_flag.append("-gencode")
cc_flag.append("arch=compute_100,code=sm_100")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_120,code=sm_120")
if bare_metal_version >= Version("13.0"):
# Allow overriding CUDA architectures via TORCH_CUDA_ARCH_LIST
# (standard PyTorch convention, used by flash-attention, xformers, etc.)
cuda_arch_list = os.environ.get("TORCH_CUDA_ARCH_LIST", None)
if cuda_arch_list:
for arch in cuda_arch_list.replace(" ", "").split(";"):
arch = arch.strip()
if not arch:
continue
if "+" in arch: # e.g. "8.6+PTX"
arch = arch.split("+")[0]
major, minor = arch.split(".")
cc = f"{major}{minor}"
cc_flag.append("-gencode")
cc_flag.append(f"arch=compute_{cc},code=sm_{cc}")
else:
cc_flag.append("-gencode")
cc_flag.append("arch=compute_103,code=sm_103")
cc_flag.append("arch=compute_75,code=sm_75")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_110,code=sm_110")
cc_flag.append("arch=compute_80,code=sm_80")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_121,code=sm_121")
cc_flag.append("arch=compute_87,code=sm_87")
if bare_metal_version >= Version("11.8"):
cc_flag.append("-gencode")
cc_flag.append("arch=compute_90,code=sm_90")
if bare_metal_version >= Version("12.8"):
cc_flag.append("-gencode")
cc_flag.append("arch=compute_100,code=sm_100")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_120,code=sm_120")
if bare_metal_version >= Version("13.0"):
cc_flag.append("-gencode")
cc_flag.append("arch=compute_103,code=sm_103")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_110,code=sm_110")
cc_flag.append("-gencode")
cc_flag.append("arch=compute_121,code=sm_121")

# HACK: The compiler flag -D_GLIBCXX_USE_CXX11_ABI is set to be the same as
# torch._C._GLIBCXX_USE_CXX11_ABI
Expand Down