Support for CUDA compute architecture 8.6

talhakavuncu · Nov 22, 2020 · 45e7f8b · 45e7f8b
1 parent d0aa70e
commit 45e7f8b
Showing 1 changed file with 14 additions and 4 deletions.
diff --git a/cmake/SelectCudaComputeArch.cmake b/cmake/SelectCudaComputeArch.cmake
@@ -38,7 +38,7 @@
 #      ARCH_AND_PTX : NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
 #      NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal Volta Turing Ampere
 #      NUM: Any number. Only those pairs are currently accepted by NVCC though:
-#            2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 7.0 7.2 7.5 8.0
+#            2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 7.0 7.2 7.5 8.0 8.6
 #      Returns LIST of flags to be added to CUDA_NVCC_FLAGS in ${out_variable}
 #      Additionally, sets ${out_variable}_readable to the resulting numeric list
 #      Example:
@@ -113,16 +113,26 @@ if(CUDA_VERSION VERSION_EQUAL "10.0" OR CUDA_VERSION VERSION_GREATER "10.0")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "7.5")
 
   if(CUDA_VERSION VERSION_LESS "11.0")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
     list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "7.5+PTX")
+    set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
   endif()
 endif()
 
 if(CUDA_VERSION VERSION_EQUAL "11.0" OR CUDA_VERSION VERSION_GREATER "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")
 
+  if(CUDA_VERSION VERSION_LESS "11.1")
+    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.0+PTX")
+    set(CUDA_LIMIT_GPU_ARCHITECTURE "8.6")
+  endif()
+endif()
+
+if(CUDA_VERSION VERSION_EQUAL "11.1" OR CUDA_VERSION VERSION_GREATER "11.1")
+  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.6" "8.6+PTX")
+  list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.6")
+
   if(CUDA_VERSION VERSION_LESS "12.0")
     set(CUDA_LIMIT_GPU_ARCHITECTURE "9.0")
   endif()
@@ -186,7 +196,7 @@ function(CUDA_DETECT_INSTALLED_GPUS OUT_VARIABLE)
     set(CUDA_GPU_DETECT_OUTPUT_FILTERED "")
     separate_arguments(CUDA_GPU_DETECT_OUTPUT)
     foreach(ITEM IN ITEMS ${CUDA_GPU_DETECT_OUTPUT})
-        if(CUDA_LIMIT_GPU_ARCHITECTURE AND (ITEM VERSION_EQUAL CUDA_LIMIT_GPU_ARCHITECTURE OR ITEM VERSION_GREATER CUDA_LIMIT_GPU_ARCHITECTURE))
+        if(CUDA_LIMIT_GPU_ARCHITECTURE AND ITEM VERSION_GREATER_EQUAL CUDA_LIMIT_GPU_ARCHITECTURE)
         list(GET CUDA_COMMON_GPU_ARCHITECTURES -1 NEWITEM)
         string(APPEND CUDA_GPU_DETECT_OUTPUT_FILTERED " ${NEWITEM}")
       else()