Skip to content

Adapt CUDA kernels and build system for CUDA 13.x compatibility #7

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

Draft
Copilot wants to merge 2 commits into
base: main
Choose a base branch
from
+12 −1
Draft

Adapt CUDA kernels and build system for CUDA 13.x compatibility #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
abd0315
Initial plan
Copilot Feb 6, 2026
329946b
Adapt paddle_scatter to CUDA 13.x compatibility
Copilot Feb 6, 2026
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
5 changes: 4 additions & 1 deletion csrc/atomics.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,10 @@ static inline __device__ void atomAdd(int64_t *address, int64_t val) {
static inline __device__ void atomAdd(float *address, float val) {
atomicAdd(address, val);
}
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 600 || CUDA_VERSION < 8000)
// CUDA 13.x only supports SM 70+, where native atomicAdd for double is available.
// For older CUDA versions (< 8.0) or architectures without native double atomicAdd
// (SM < 6.0), fall back to CAS-based implementation.
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 600) && (CUDA_VERSION < 13000)
static inline __device__ void atomAdd(double *address, double val) {
AtomicAddDecimalImpl<double, sizeof(double)>()(address, val);
}
Expand Down
8 changes: 8 additions & 0 deletions setup_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,14 @@ def get_extensions():
nvcc_flags = [] if nvcc_flags == "" else nvcc_flags.split(" ")
nvcc_flags += ["-O3"]
nvcc_flags += ["--expt-relaxed-constexpr"]

# CUDA 13.x may have stricter host compiler version checks,
# add --allow-unsupported-compiler to avoid build failures
# with newer or not-yet-certified host compilers.
cuda_major, _ = paddle.version.cuda_version.split(".")
if int(cuda_major) >= 13:
nvcc_flags += ["--allow-unsupported-compiler"]

extra_compile_args["nvcc"] = nvcc_flags

src = get_sources()
Expand Down