fix(nccl-tests): pass NCCL_LIB from nccl cache to fix -lnccl link error
This commit is contained in:
@@ -78,6 +78,11 @@ NCCL_H=$(find . -name 'nccl.h' -type f 2>/dev/null | head -1)
|
||||
NCCL_INCLUDE_DIR="$(pwd)/$(dirname "$NCCL_H")"
|
||||
echo "nccl.h: $NCCL_H"
|
||||
|
||||
# libnccl.so comes from the already-built NCCL cache (build-nccl.sh ran first)
|
||||
NCCL_LIB_DIR="${DIST_DIR}/nccl-${NCCL_VERSION}+cuda${NCCL_CUDA_VERSION}/lib"
|
||||
[ -d "$NCCL_LIB_DIR" ] || { echo "ERROR: NCCL lib dir not found at $NCCL_LIB_DIR — run build-nccl.sh first"; exit 1; }
|
||||
echo "nccl lib: $NCCL_LIB_DIR"
|
||||
|
||||
# Download nccl-tests source
|
||||
SRC_TAR="${DOWNLOAD_CACHE_DIR}/nccl-tests-v${NCCL_TESTS_VERSION}.tar.gz"
|
||||
SRC_URL="https://github.com/NVIDIA/nccl-tests/archive/refs/tags/v${NCCL_TESTS_VERSION}.tar.gz"
|
||||
@@ -106,6 +111,7 @@ make MPI=0 \
|
||||
NVCC="$NVCC" \
|
||||
CUDA_HOME="$CUDA_HOME" \
|
||||
NCCL_HOME="$NCCL_INCLUDE_DIR/.." \
|
||||
NCCL_LIB="$NCCL_LIB_DIR" \
|
||||
NVCC_GENCODE="$GENCODE" \
|
||||
BUILDDIR="./build"
|
||||
|
||||
|
||||
Reference in New Issue
Block a user