EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[NVIDIA] Build CUDA 13 (#11299)

Browse Source 
Co-authored-by: ishandhanani <ishandhanani@gmail.com>
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
This commit is contained in:
Johnny
2025-10-22 20:03:12 -07:00
committed by GitHub
parent 4d4feccbb2
commit e7aa4664b3
6 changed files with 131 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
sgl-kernel/CMakeLists.txt
View File

@@ -224,6 +224,12 @@ if (ENABLE_BELOW_SM90)
"-gencode=arch=compute_80,code=sm_80"
"-gencode=arch=compute_89,code=sm_89"
)
if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_87,code=sm_87"
)
endif()
endif()
if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.8" OR SGL_KERNEL_ENABLE_SM100A)
@@ -231,19 +237,24 @@ if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.8" OR SGL_KERNEL_ENABLE_SM100A)
"-gencode=arch=compute_100a,code=sm_100a"
"-gencode=arch=compute_120a,code=sm_120a"
)
# refer sm_121, sm_110 and sm_101 description https://github.com/pytorch/pytorch/pull/156176
if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "13.0")
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_103a,code=sm_103a"
"-gencode=arch=compute_110a,code=sm_110a"
"-gencode=arch=compute_121a,code=sm_121a"
"--compress-mode=size"
)
if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_110a,code=sm_110a"
"-gencode=arch=compute_121a,code=sm_121a"
)
endif()
else()
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_101a,code=sm_101a"
)
if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_101a,code=sm_101a"
)
endif()
endif()
endif()

7
sgl-kernel/build.sh
View File

@@ -20,7 +20,10 @@ else
BUILDER_NAME="pytorch/manylinux2_28-builder"
fi
if [ ${CUDA_VERSION} = "12.9" ]; then
if [ ${CUDA_VERSION} = "13.0" ]; then
DOCKER_IMAGE="${BUILDER_NAME}:cuda${CUDA_VERSION}"
TORCH_INSTALL="pip install --no-cache-dir torch==2.9.0 --index-url https://download.pytorch.org/whl/cu130"
elif [ ${CUDA_VERSION} = "12.9" ]; then
DOCKER_IMAGE="${BUILDER_NAME}:cuda${CUDA_VERSION}"
TORCH_INSTALL="pip install --no-cache-dir torch==2.8.0 --index-url https://download.pytorch.org/whl/cu129"
elif [ ${CUDA_VERSION} = "12.8" ]; then
@@ -148,6 +151,8 @@ docker run --rm \
export CUDA_VERSION=${CUDA_VERSION} && \
mkdir -p /usr/lib/${ARCH}-linux-gnu/ && \
ln -s /usr/local/cuda-${CUDA_VERSION}/targets/${LIBCUDA_ARCH}-linux/lib/stubs/libcuda.so /usr/lib/${ARCH}-linux-gnu/libcuda.so && \
export CPLUS_INCLUDE_PATH=/usr/local/cuda/include/cccl${CPLUS_INCLUDE_PATH:+:${CPLUS_INCLUDE_PATH}} && \
export C_INCLUDE_PATH=/usr/local/cuda/include/cccl${C_INCLUDE_PATH:+:${C_INCLUDE_PATH}} && \
cd /sgl-kernel && \
ls -la ${PYTHON_ROOT_PATH}/lib/python${PYTHON_VERSION}/site-packages/wheel/ && \