fix bug of "unsupported architecture " when use some version's CUDA toolkit

Author: Lu

source/op/cuda/CMakeLists.txt

@@ -14,24 +14,77 @@ SET(CMAKE_CXX_STANDARD 11)
 SET(CMAKE_CUDA_STANDARD 11)
 # 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
 # very important here! Include path to cub.
+# for searching device compute capability, https://developer.nvidia.com/cuda-gpus
 include_directories(cub)
-if (${CUDA_VERSION_MAJOR} STREQUAL "10")
+
+if (${CUDA_VERSION_MAJOR} GREATER "10")
+    message(STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR})
+        # nvcc flags
+        set(CUDA_NVCC_FLAGS -gencode arch=compute_50,code=sm_50;
+                            -gencode arch=compute_52,code=sm_52; # Tesla M40, Tesla M40, Quadro M6000...
+                            -gencode arch=compute_53,code=sm_53; 
+                            -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
+                            -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
+                            -gencode arch=compute_70,code=sm_70; # Volta  - GV100/Tesla V100, GTX 1180 (GV104)
+                            -gencode arch=compute_75,code=sm_75; # Turing - RTX 2080, Titan RTX, Quadro R8000
+                            -O3; -Xcompiler -fPIC;
+            )
+elseif (${CUDA_VERSION_MAJOR} STREQUAL "10")
     message(STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR})
     # nvcc flags
-    set(CUDA_NVCC_FLAGS -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
+    set(CUDA_NVCC_FLAGS -gencode arch=compute_30,code=sm_30; # Tesla K10, Quadro K600 K420 K410,
+                        -gencode arch=compute_35,code=sm_35; # Tesla K20 K40, TITAN Z Black, GTX 780Ti 780
+                        -gencode arch=compute_37,code=sm_37; # Tesla K80
+                        -gencode arch=compute_50,code=sm_50; # Quadro 620 1200
+                        -gencode arch=compute_52,code=sm_52; # Tesla M40 M40, Quadro M6000 M5000 M4000 M2000, TITAN X, GTX 980Ti 980 970 960 950
+                        -gencode arch=compute_53,code=sm_53; # Jetson TX1, Tegra X1
+                        -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
                         -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
                         -gencode arch=compute_70,code=sm_70; # Volta  - GV100/Tesla V100, GTX 1180 (GV104)
                         -gencode arch=compute_75,code=sm_75; # Turing - RTX 2080, Titan RTX, Quadro R8000
                         -O3; -Xcompiler -fPIC;
         )
-else (${CUDA_VERSION_MAJOR} STREQUAL "10")
+elseif (${CUDA_VERSION_MAJOR} STREQUAL "9")
     message(STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR})
     # nvcc flags
-    set(CUDA_NVCC_FLAGS -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
+    set(CUDA_NVCC_FLAGS -gencode arch=compute_30,code=sm_30;
+                        -gencode arch=compute_35,code=sm_35;
+                        -gencode arch=compute_37,code=sm_37;
+                        -gencode arch=compute_50,code=sm_50;
+                        -gencode arch=compute_52,code=sm_52; # Tesla M40, Tesla M40, Quadro M6000...
+                        -gencode arch=compute_53,code=sm_53; 
+                        -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
                         -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
                         -gencode arch=compute_70,code=sm_70; # Volta  - GV100/Tesla V100, GTX 1180 (GV104)
+                        -gencode arch=compute_72,code=sm_72;
+                        -O3; -Xcompiler -fPIC;
+        )
+elseif (${CUDA_VERSION_MAJOR} STREQUAL "8")
+    message(STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR})
+    # nvcc flags
+    set(CUDA_NVCC_FLAGS -gencode arch=compute_30,code=sm_30;
+                        -gencode arch=compute_35,code=sm_35;
+                        -gencode arch=compute_37,code=sm_37;
+                        -gencode arch=compute_50,code=sm_50;
+                        -gencode arch=compute_52,code=sm_52; # Tesla M40, Tesla M40, Quadro M6000...
+                        -gencode arch=compute_53,code=sm_53; 
+                        -gencode arch=compute_60,code=sm_60; # Pascal – GP100/Tesla P100 – DGX-1 (Generic Pascal)
+                        -gencode arch=compute_61,code=sm_61; # Pascal - GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4, Discrete GPU on the NVIDIA Drive PX2
+                        -O3; -Xcompiler -fPIC;
+        )
+elseif (${CUDA_VERSION_MAJOR} STREQUAL "7")
+    message(STATUS "CUDA major version is " ${CUDA_VERSION_MAJOR})
+    # nvcc flags
+    set(CUDA_NVCC_FLAGS -gencode arch=compute_30,code=sm_30;
+                        -gencode arch=compute_35,code=sm_35;
+                        -gencode arch=compute_37,code=sm_37;
+                        -gencode arch=compute_50,code=sm_50;
+                        -gencode arch=compute_52,code=sm_52; # Tesla M40, Tesla M40, Quadro M6000...
+                        -gencode arch=compute_53,code=sm_53; 
                         -O3; -Xcompiler -fPIC;
         )
+else () 
+    message(FATAL_ERROR "unsupported CUDA_VERSION " ${CUDA_VERSION} ", please use a newer version (>=7.0) of CUDA toolkit!")
 endif()
 
 set (SOURCE_FILES