Commit 1a2f45fc authored by lijian6's avatar lijian6
Browse files

Add compile on gfx938


Signed-off-by: lijian6's avatarlijian <lijian6@sugon.com>
parent 8b97effd
...@@ -89,7 +89,7 @@ if [ "$USE_ROCSHMEM" == "ON" ]; then ...@@ -89,7 +89,7 @@ if [ "$USE_ROCSHMEM" == "ON" ]; then
build_rocshmem build_rocshmem
SHMEM_INSTALL_PREFIX=$(pwd)/third-party/rocshmem_install SHMEM_INSTALL_PREFIX=$(pwd)/third-party/rocshmem_install
COMPILE_OPTIONS=${COMPILE_OPTIONS:= -fPIC -D__HIP_PLATFORM_AMD__=1 -DUSE_ROCM=1 -DHIPBLAS_V2 -DCUDA_HAS_FP16=1 -O3 -fgpu-rdc -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1014"' -DTORCH_EXTENSION_NAME=deep_ep_cpp -D_GLIBCXX_USE_CXX11_ABI=1 --offload-arch=gfx936 -std=c++17 -Wno-return-type} COMPILE_OPTIONS=${COMPILE_OPTIONS:= -fPIC -D__HIP_PLATFORM_AMD__=1 -DUSE_ROCM=1 -DHIPBLAS_V2 -DCUDA_HAS_FP16=1 -O3 -fgpu-rdc -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1014"' -DTORCH_EXTENSION_NAME=deep_ep_cpp -D_GLIBCXX_USE_CXX11_ABI=1 --offload-arch=gfx936 --offload-arch=gfx938 -std=c++17 -Wno-return-type}
if [ "$ROCM_DISABLE_CTX" == "ON" ]; then if [ "$ROCM_DISABLE_CTX" == "ON" ]; then
COMPILE_OPTIONS="-DROCM_DISABLE_CTX $COMPILE_OPTIONS" COMPILE_OPTIONS="-DROCM_DISABLE_CTX $COMPILE_OPTIONS"
fi fi
...@@ -133,7 +133,7 @@ if [ "$USE_NVSHMEM" == "ON" ]; then ...@@ -133,7 +133,7 @@ if [ "$USE_NVSHMEM" == "ON" ]; then
# build_dushmem # build_dushmem
# SHMEM_INSTALL_PREFIX=$(pwd)/third-party/dushmem_install # SHMEM_INSTALL_PREFIX=$(pwd)/third-party/dushmem_install
SHMEM_INSTALL_PREFIX=${ROCM_PATH}/dushmem SHMEM_INSTALL_PREFIX=${ROCM_PATH}/dushmem
COMPILE_OPTIONS=${COMPILE_OPTIONS:= -fPIC -DFORCE_NVSHMEM_API -DHIP_ENABLE_WARP_SYNC_BUILTINS -D__HIP_PLATFORM_AMD__=1 -DUSE_ROCM=1 -DHIPBLAS_V2 -DCUDA_HAS_FP16=1 -O3 -fgpu-rdc -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1014"' -DTORCH_EXTENSION_NAME=deep_ep_cpp -D_GLIBCXX_USE_CXX11_ABI=1 --offload-arch=gfx936 -std=c++17 -Wno-return-type} COMPILE_OPTIONS=${COMPILE_OPTIONS:= -fPIC -DFORCE_NVSHMEM_API -DHIP_ENABLE_WARP_SYNC_BUILTINS -D__HIP_PLATFORM_AMD__=1 -DUSE_ROCM=1 -DHIPBLAS_V2 -DCUDA_HAS_FP16=1 -O3 -fgpu-rdc -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1014"' -DTORCH_EXTENSION_NAME=deep_ep_cpp -D_GLIBCXX_USE_CXX11_ABI=1 --offload-arch=gfx936 --offload-arch=gfx938 -std=c++17 -Wno-return-type}
SHMEM_LINK_OPTIONS="-Wl,-rpath,${SHMEM_INSTALL_PREFIX}/lib/ -l:libnvshmem_device.a -lnvshmem_host" SHMEM_LINK_OPTIONS="-Wl,-rpath,${SHMEM_INSTALL_PREFIX}/lib/ -l:libnvshmem_device.a -lnvshmem_host"
fi fi
# -------------------------- duSHMEM END -------------------------- # # -------------------------- duSHMEM END -------------------------- #
...@@ -147,7 +147,7 @@ hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/kernels/internode.cu -o build_/internode.o ...@@ -147,7 +147,7 @@ hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/kernels/internode.cu -o build_/internode.o
hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/kernels/internode_ll.cu -o build_/internode_ll.o ${COMPILE_OPTIONS} hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/kernels/internode_ll.cu -o build_/internode_ll.o ${COMPILE_OPTIONS}
hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/deep_ep.cu -o build_/deep_ep.o ${COMPILE_OPTIONS} hipcc ${INCLUDE_PATHS} -c $(pwd)/csrc/deep_ep.cu -o build_/deep_ep.o ${COMPILE_OPTIONS}
hipcc -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -g -fwrapv -O2 -shared -Wl,-O1 -Wl,-Bsymbolic-functions build_/internode.o build_/intranode.o build_/runtime.o build_/deep_ep.o build_/layout.o build_/internode_ll.o -L${SHMEM_INSTALL_PREFIX}/lib/ -L/opt/mpi/lib -L/opt/dtk/hip/lib -L/usr/lib/x86_64-linux-gnu -lhipblaslt -lamdhip64 -o deep_ep/deep_ep_cpp.cpython-310-x86_64-linux-gnu.so -Wl,-rpath,/opt/dtk/lib -fgpu-rdc --hip-link --offload-arch=gfx936 -shared -Wl,-soname,deep_ep/deep_ep_cpp.cpython-310-x86_64-linux-gnu.so -L"${llvm_path}/include/../lib/linux" -lclang_rt.builtins-x86_64 /opt/dtk/hip/lib/libgalaxyhip.so ${llvm_path}/lib/linux/libclang_rt.builtins-x86_64.a /opt/hyhal/lib/libhsa-runtime64.so -L${PYTHON_PLATLIB}/torch/lib -L/opt/dtk/lib -L/opt/dtk/hip/lib -L/usr/local/lib -lc10 -ltorch -ltorch_cpu -ltorch_python -lamdhip64 -lc10_hip -ltorch_hip -lrocm-core -lrocm_smi64 ${SHMEM_LINK_OPTIONS} -fgpu-rdc --hip-link -lamdhip64 -lhsa-runtime64 -l:libmpi.so -Wl,-rpath,/opt/mpi/lib/ -libverbs -lmlx5 hipcc -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -g -fwrapv -O2 -shared -Wl,-O1 -Wl,-Bsymbolic-functions build_/internode.o build_/intranode.o build_/runtime.o build_/deep_ep.o build_/layout.o build_/internode_ll.o -L${SHMEM_INSTALL_PREFIX}/lib/ -L/opt/mpi/lib -L/opt/dtk/hip/lib -L/usr/lib/x86_64-linux-gnu -lhipblaslt -lamdhip64 -o deep_ep/deep_ep_cpp.cpython-310-x86_64-linux-gnu.so -Wl,-rpath,/opt/dtk/lib -fgpu-rdc --hip-link --offload-arch=gfx936 --offload-arch=gfx938 -shared -Wl,-soname,deep_ep/deep_ep_cpp.cpython-310-x86_64-linux-gnu.so -L"${llvm_path}/include/../lib/linux" -lclang_rt.builtins-x86_64 /opt/dtk/hip/lib/libgalaxyhip.so ${llvm_path}/lib/linux/libclang_rt.builtins-x86_64.a /opt/hyhal/lib/libhsa-runtime64.so -L${PYTHON_PLATLIB}/torch/lib -L/opt/dtk/lib -L/opt/dtk/hip/lib -L/usr/local/lib -lc10 -ltorch -ltorch_cpu -ltorch_python -lamdhip64 -lc10_hip -ltorch_hip -lrocm-core -lrocm_smi64 ${SHMEM_LINK_OPTIONS} -fgpu-rdc --hip-link -lamdhip64 -lhsa-runtime64 -l:libmpi.so -Wl,-rpath,/opt/mpi/lib/ -libverbs -lmlx5
# build whl # build whl
echo "Using Python: $(which python3)" echo "Using Python: $(which python3)"
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment