@@ -14,14 +14,25 @@ SET(CMAKE_CXX_STANDARD 11)
1414SET (CMAKE_CUDA_STANDARD 11)
1515# nvcc -o libdeepmd_op_cuda.so -I/usr/local/cub-1.8.0 -rdc=true -DHIGH_PREC=true -gencode arch=compute_61,code=sm_61 -shared -Xcompiler -fPIC deepmd_op.cu -L/usr/local/cuda/lib64 -lcudadevrt
1616# very important here! Include path to cub.
17- include_directories (cub)
18- # nvcc flags
19- set (CUDA_NVCC_FLAGS -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
20- -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
21- -gencode arch=compute_70,code=sm_70; # Volta - GV100/Tesla V100, GTX 1180 (GV104)
22- -gencode arch=compute_75,code=sm_75; # Turing - RTX 2080, Titan RTX, Quadro R8000
23- -O3; -Xcompiler -fPIC;
24- )
17+ include_directories (cub)
18+ if (${CUDA_VERSION_MAJOR} STREQUAL "10" )
19+ message (STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR} )
20+ # nvcc flags
21+ set (CUDA_NVCC_FLAGS -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
22+ -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
23+ -gencode arch=compute_70,code=sm_70; # Volta - GV100/Tesla V100, GTX 1180 (GV104)
24+ -gencode arch=compute_75,code=sm_75; # Turing - RTX 2080, Titan RTX, Quadro R8000
25+ -O3; -Xcompiler -fPIC;
26+ )
27+ else (${CUDA_VERSION_MAJOR} STREQUAL "10" )
28+ message (STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR} )
29+ # nvcc flags
30+ set (CUDA_NVCC_FLAGS -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
31+ -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
32+ -gencode arch=compute_70,code=sm_70; # Volta - GV100/Tesla V100, GTX 1180 (GV104)
33+ -O3; -Xcompiler -fPIC;
34+ )
35+ endif ()
2536
2637set (SOURCE_FILES
2738 descrpt_se_a.cu descrpt_se_r.cu prod_force_se_a.cu prod_force_se_r.cu prod_virial_se_a.cu prod_virial_se_r.cu
0 commit comments