mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-06 00:20:18 +01:00
Add CUDA installation script for CUDA 13 (#160201)
Add the almalinux docker for building magma-cuda 13.0 https://github.com/pytorch/pytorch/issues/159779 Also fixed the NVSHMEM download link Pull Request resolved: https://github.com/pytorch/pytorch/pull/160201 Approved by: https://github.com/atalman Co-authored-by: Andrey Talman <atalman@fb.com>
This commit is contained in:
Ting Lu
PyTorch MergeBot
parent
4014672b30
commit
87d6831b2e
|
|
@ -64,6 +64,10 @@ FROM cuda as cuda12.9
|
|||
RUN bash ./install_cuda.sh 12.9
|
||||
ENV DESIRED_CUDA=12.9
|
||||
|
||||
FROM cuda as cuda13.0
|
||||
RUN bash ./install_cuda.sh 13.0
|
||||
ENV DESIRED_CUDA=13.0
|
||||
|
||||
FROM ${ROCM_IMAGE} as rocm
|
||||
ENV PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
|
||||
ADD ./common/install_mkl.sh install_mkl.sh
|
||||
|
|
@ -79,6 +83,7 @@ FROM base as all_cuda
|
|||
COPY --from=cuda12.6 /usr/local/cuda-12.6 /usr/local/cuda-12.6
|
||||
COPY --from=cuda12.8 /usr/local/cuda-12.8 /usr/local/cuda-12.8
|
||||
COPY --from=cuda12.9 /usr/local/cuda-12.9 /usr/local/cuda-12.9
|
||||
COPY --from=cuda13.0 /usr/local/cuda-13.0 /usr/local/cuda-13.0
|
||||
|
||||
# Final step
|
||||
FROM ${BASE_TARGET} as final
|
||||
|
|
|
|||
1
.ci/docker/ci_commit_pins/nccl-cu13.txt
Normal file
1
.ci/docker/ci_commit_pins/nccl-cu13.txt
Normal file
|
|
@ -0,0 +1 @@
|
|||
v2.27.7-1
|
||||
|
|
@ -128,74 +128,6 @@ function install_129 {
|
|||
ldconfig
|
||||
}
|
||||
|
||||
function prune_124 {
|
||||
echo "Pruning CUDA 12.4"
|
||||
#####################################################################################
|
||||
# CUDA 12.4 prune static libs
|
||||
#####################################################################################
|
||||
export NVPRUNE="/usr/local/cuda-12.4/bin/nvprune"
|
||||
export CUDA_LIB_DIR="/usr/local/cuda-12.4/lib64"
|
||||
|
||||
export GENCODE="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
|
||||
export GENCODE_CUDNN="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
|
||||
|
||||
if [[ -n "$OVERRIDE_GENCODE" ]]; then
|
||||
export GENCODE=$OVERRIDE_GENCODE
|
||||
fi
|
||||
if [[ -n "$OVERRIDE_GENCODE_CUDNN" ]]; then
|
||||
export GENCODE_CUDNN=$OVERRIDE_GENCODE_CUDNN
|
||||
fi
|
||||
|
||||
# all CUDA libs except CuDNN and CuBLAS
|
||||
ls $CUDA_LIB_DIR/ | grep "\.a" | grep -v "culibos" | grep -v "cudart" | grep -v "cudnn" | grep -v "cublas" | grep -v "metis" \
|
||||
| xargs -I {} bash -c \
|
||||
"echo {} && $NVPRUNE $GENCODE $CUDA_LIB_DIR/{} -o $CUDA_LIB_DIR/{}"
|
||||
|
||||
# prune CuDNN and CuBLAS
|
||||
$NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublas_static.a -o $CUDA_LIB_DIR/libcublas_static.a
|
||||
$NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublasLt_static.a -o $CUDA_LIB_DIR/libcublasLt_static.a
|
||||
|
||||
#####################################################################################
|
||||
# CUDA 12.4 prune visual tools
|
||||
#####################################################################################
|
||||
export CUDA_BASE="/usr/local/cuda-12.4/"
|
||||
rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2024.1.0 $CUDA_BASE/nsight-systems-2023.4.4/
|
||||
}
|
||||
|
||||
function prune_126 {
|
||||
echo "Pruning CUDA 12.6"
|
||||
#####################################################################################
|
||||
# CUDA 12.6 prune static libs
|
||||
#####################################################################################
|
||||
export NVPRUNE="/usr/local/cuda-12.6/bin/nvprune"
|
||||
export CUDA_LIB_DIR="/usr/local/cuda-12.6/lib64"
|
||||
|
||||
export GENCODE="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
|
||||
export GENCODE_CUDNN="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
|
||||
|
||||
if [[ -n "$OVERRIDE_GENCODE" ]]; then
|
||||
export GENCODE=$OVERRIDE_GENCODE
|
||||
fi
|
||||
if [[ -n "$OVERRIDE_GENCODE_CUDNN" ]]; then
|
||||
export GENCODE_CUDNN=$OVERRIDE_GENCODE_CUDNN
|
||||
fi
|
||||
|
||||
# all CUDA libs except CuDNN and CuBLAS
|
||||
ls $CUDA_LIB_DIR/ | grep "\.a" | grep -v "culibos" | grep -v "cudart" | grep -v "cudnn" | grep -v "cublas" | grep -v "metis" \
|
||||
| xargs -I {} bash -c \
|
||||
"echo {} && $NVPRUNE $GENCODE $CUDA_LIB_DIR/{} -o $CUDA_LIB_DIR/{}"
|
||||
|
||||
# prune CuDNN and CuBLAS
|
||||
$NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublas_static.a -o $CUDA_LIB_DIR/libcublas_static.a
|
||||
$NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublasLt_static.a -o $CUDA_LIB_DIR/libcublasLt_static.a
|
||||
|
||||
#####################################################################################
|
||||
# CUDA 12.6 prune visual tools
|
||||
#####################################################################################
|
||||
export CUDA_BASE="/usr/local/cuda-12.6/"
|
||||
rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2024.3.2 $CUDA_BASE/nsight-systems-2024.5.1/
|
||||
}
|
||||
|
||||
function install_128 {
|
||||
CUDNN_VERSION=9.8.0.87
|
||||
echo "Installing CUDA 12.8.1 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
|
||||
|
|
@ -214,18 +146,39 @@ function install_128 {
|
|||
ldconfig
|
||||
}
|
||||
|
||||
function install_130 {
|
||||
CUDNN_VERSION=9.12.0.46
|
||||
NVSHMEM_VERSION=3.3.20
|
||||
echo "Installing CUDA 13.0 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
|
||||
# install CUDA 13.0 in the same container
|
||||
install_cuda 13.0.0 cuda_13.0.0_580.65.06_linux
|
||||
|
||||
# cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
|
||||
install_cudnn 13 $CUDNN_VERSION
|
||||
|
||||
install_nvshmem 13 $NVSHMEM_VERSION
|
||||
|
||||
CUDA_VERSION=13.0 bash install_nccl.sh
|
||||
|
||||
CUDA_VERSION=13.0 bash install_cusparselt.sh
|
||||
|
||||
ldconfig
|
||||
}
|
||||
|
||||
# idiomatic parameter and option handling in sh
|
||||
while test $# -gt 0
|
||||
do
|
||||
case "$1" in
|
||||
12.4) install_124; prune_124
|
||||
12.4) install_124;
|
||||
;;
|
||||
12.6|12.6.*) install_126; prune_126
|
||||
12.6|12.6.*) install_126;
|
||||
;;
|
||||
12.8|12.8.*) install_128;
|
||||
;;
|
||||
12.9|12.9.*) install_129;
|
||||
;;
|
||||
13.0|13.0.*) install_130;
|
||||
;;
|
||||
*) echo "bad argument $1"; exit 1
|
||||
;;
|
||||
esac
|
||||
|
|
|
|||
|
|
@ -5,7 +5,15 @@ set -ex
|
|||
# cuSPARSELt license: https://docs.nvidia.com/cuda/cusparselt/license.html
|
||||
mkdir tmp_cusparselt && cd tmp_cusparselt
|
||||
|
||||
if [[ ${CUDA_VERSION:0:4} =~ ^12\.[5-9]$ ]]; then
|
||||
if [[ ${CUDA_VERSION:0:4} =~ "13" ]]; then
|
||||
arch_path='sbsa'
|
||||
export TARGETARCH=${TARGETARCH:-$(uname -m)}
|
||||
if [ ${TARGETARCH} = 'amd64' ] || [ "${TARGETARCH}" = 'x86_64' ]; then
|
||||
arch_path='x86_64'
|
||||
fi
|
||||
CUSPARSELT_NAME="libcusparse_lt-linux-${arch_path}-0.8.0.4_cuda13-archive"
|
||||
curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-${arch_path}/${CUSPARSELT_NAME}.tar.xz
|
||||
elif [[ ${CUDA_VERSION:0:4} =~ ^12\.[5-9]$ ]]; then
|
||||
arch_path='sbsa'
|
||||
export TARGETARCH=${TARGETARCH:-$(uname -m)}
|
||||
if [ ${TARGETARCH} = 'amd64' ] || [ "${TARGETARCH}" = 'x86_64' ]; then
|
||||
|
|
|
|||
|
|
@ -7,6 +7,8 @@ if [[ ${CUDA_VERSION:0:2} == "11" ]]; then
|
|||
NCCL_VERSION=$(cat ci_commit_pins/nccl-cu11.txt)
|
||||
elif [[ ${CUDA_VERSION:0:2} == "12" ]]; then
|
||||
NCCL_VERSION=$(cat ci_commit_pins/nccl-cu12.txt)
|
||||
elif [[ ${CUDA_VERSION:0:2} == "13" ]]; then
|
||||
NCCL_VERSION=$(cat ci_commit_pins/nccl-cu13.txt)
|
||||
else
|
||||
echo "Unexpected CUDA_VERSION ${CUDA_VERSION}"
|
||||
exit 1
|
||||
|
|
|
|||
2
.github/workflows/build-almalinux-images.yml
vendored
2
.github/workflows/build-almalinux-images.yml
vendored
|
|
@ -36,7 +36,7 @@ jobs:
|
|||
runs-on: linux.9xlarge.ephemeral
|
||||
strategy:
|
||||
matrix:
|
||||
tag: ["cuda12.6", "cuda12.8", "cuda12.9", "rocm6.3", "rocm6.4", "cpu"]
|
||||
tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm6.3", "rocm6.4", "cpu"]
|
||||
steps:
|
||||
- name: Build docker image
|
||||
uses: pytorch/pytorch/.github/actions/binary-docker-build@main
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user