Skip to content

Fix: CUDA build flags and linker compatibility for Windows and Linux #90

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

Fix: CUDA build flags and linker compatibility for Windows and Linux #90

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
47 changes: 31 additions & 16 deletions xlstm/blocks/slstm/src/cuda_init.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,9 @@ def defines_to_cflags(defines=Union[dict[str, Union[int, str]], Sequence[tuple[s


def load(*, name, sources, extra_cflags=(), extra_cuda_cflags=(), **kwargs):
import platform
system_name = platform.system()

suffix = ""
for flag in extra_cflags:
pref = [st[0] for st in flag[2:].split("=")[0].split("_")]
Expand All @@ -73,6 +76,21 @@ def load(*, name, sources, extra_cflags=(), extra_cuda_cflags=(), **kwargs):
suffix = "_" + suffix
suffix = suffix[:64]

# Determine platform-specific linker flags
extra_ldflags = (
[f"/LIBPATH:{os.environ['CUDA_LIB']}", "cublas.lib"]
if system_name == "Windows"
else [f"-L{os.environ['CUDA_LIB']}", "-lcublas"]
)

# Handle -Xptxas flags platform-specifically
if system_name == "Windows":
xptxas_verbose = ["-Xptxas", "-v"] # must be split for nvcc on Windows
xptxas_opt = ["-Xptxas", "-O3"]
else:
xptxas_verbose = ['-Xptxas="-v"'] # works fine quoted on Linux
xptxas_opt = ['-Xptxas -O3']

extra_cflags = list(extra_cflags) + [
"-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__",
Expand All @@ -86,26 +104,23 @@ def load(*, name, sources, extra_cflags=(), extra_cuda_cflags=(), **kwargs):
extra_cflags.append("-isystem")
extra_cflags.append(eip)

# Compose full set of CUDA flags
extra_cuda_flags = [
*xptxas_verbose,
"-gencode", "arch=compute_80,code=compute_80",
"-res-usage",
"--use_fast_math",
"-O3",
*xptxas_opt,
"--extra-device-vectorization",
]

myargs = {
"verbose": True,
"with_cuda": True,
"extra_ldflags": [f"-L{os.environ['CUDA_LIB']}", "-lcublas"],
"extra_ldflags": extra_ldflags,
"extra_cflags": [*extra_cflags],
"extra_cuda_cflags": [
# "-gencode",
# "arch=compute_70,code=compute_70",
# "-dbg=1",
'-Xptxas="-v"',
"-gencode",
"arch=compute_80,code=compute_80",
"-res-usage",
"--use_fast_math",
"-O3",
"-Xptxas -O3",
"--extra-device-vectorization",
*extra_cflags,
*extra_cuda_cflags,
],
"extra_cuda_cflags": [*extra_cuda_flags, *extra_cflags, *extra_cuda_cflags],
}
print(myargs)
myargs.update(**kwargs)
Expand Down