fix(nccl-tests): limit CUDA gencode to sm_70+ (CUDA 13 dropped Pascal)

This commit is contained in:
2026-03-26 23:48:40 +03:00
parent d2f384b6eb
commit 4fddaba9c5

View File

@@ -96,10 +96,16 @@ SRC_DIR=$(ls -d nccl-tests-* 2>/dev/null | head -1)
cd "$SRC_DIR"
echo "=== building all_reduce_perf ==="
# CUDA 13.0 dropped support for compute_60 (Pascal); target Volta+ only.
GENCODE="-gencode=arch=compute_70,code=sm_70 \
-gencode=arch=compute_80,code=sm_80 \
-gencode=arch=compute_86,code=sm_86 \
-gencode=arch=compute_90,code=sm_90"
make MPI=0 \
NVCC="$NVCC" \
CUDA_HOME="$CUDA_HOME" \
NCCL_HOME="$NCCL_INCLUDE_DIR/.." \
NVCC_GENCODE="$GENCODE" \
BUILDDIR="./build"
[ -f "./build/all_reduce_perf" ] || { echo "ERROR: all_reduce_perf not found after build"; exit 1; }