Add support for cuda 11.1 and compute 8.6

arrayfire · Oct 5, 2020 · db3c333 · db3c333
1 parent c903e57
commit db3c333
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 2 deletions.
diff --git a/CMakeModules/select_compute_arch.cmake b/CMakeModules/select_compute_arch.cmake
@@ -95,13 +95,22 @@ endif()
 
 if(CUDA_VERSION VERSION_GREATER "11.0" OR CUDA_VERSION VERSION_EQUAL "11.0")
   list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Ampere")
-  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.0" "8.0+PTX")
+  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.0")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.0")
 
   list(REMOVE_ITEM CUDA_KNOWN_GPU_ARCHITECTURES "Kepler+Tegra")
   list(REMOVE_ITEM CUDA_KNOWN_GPU_ARCHITECTURES "Kepler")
   list(REMOVE_ITEM CUDA_COMMON_GPU_ARCHITECTURES "3.0" "3.2")
 
+  if(CUDA_VERSION VERSION_GREATER "11.1" OR CUDA_VERSION VERSION_EQUAL "11.1")
+    list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.6")
+  endif()
+
+  if(CUDA_VERSION VERSION_GREATER "11.1" OR CUDA_VERSION VERSION_EQUAL "11.1"
+      AND CUDA_VERSION VERSION_LESS "12.0")
+    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.0+PTX")
+  endif()
+
   if(CUDA_VERSION VERSION_LESS "12.0")
     set(CUDA_LIMIT_GPU_ARCHITECTURE "9.0")
   endif()

diff --git a/src/backend/cuda/device_manager.cpp b/src/backend/cuda/device_manager.cpp
@@ -97,6 +97,7 @@ static const int jetsonComputeCapabilities[] = {
 
 // clang-format off
 static const cuNVRTCcompute Toolkit2MaxCompute[] = {
+    {11010, 8, 0, 0},
     {11000, 8, 0, 0},
     {10020, 7, 5, 2},
     {10010, 7, 5, 2},
@@ -115,6 +116,7 @@ static const cuNVRTCcompute Toolkit2MaxCompute[] = {
 // clang-format off
 static const ToolkitDriverVersions
     CudaToDriverVersion[] = {
+        {11010, 455.23f, 456.38f},
         {11000, 450.51f, 451.48f},
         {10020, 440.33f, 441.22f},
         {10010, 418.39f, 418.96f},
@@ -237,7 +239,8 @@ static inline int compute2cores(unsigned major, unsigned minor) {
         {0x10, 8},   {0x11, 8},   {0x12, 8},   {0x13, 8},   {0x20, 32},
         {0x21, 48},  {0x30, 192}, {0x32, 192}, {0x35, 192}, {0x37, 192},
         {0x50, 128}, {0x52, 128}, {0x53, 128}, {0x60, 64},  {0x61, 128},
-        {0x62, 128}, {0x70, 64},  {0x75, 64},  {0x80, 64},  {-1, -1},
+        {0x62, 128}, {0x70, 64},  {0x75, 64},  {0x80, 64},  {0x86, 128},
+        {-1, -1},
     };
 
     for (int i = 0; gpus[i].compute != -1; ++i) {