CUDA: Update arch selection, restructure for maintainability

Fixes: #21712

1 files changed, 32 insertions, 46 deletions

diff --git a/Modules/FindCUDA/select_compute_arch.cmake b/Modules/FindCUDA/select_compute_arch.cmake
index b583308..a35b3f8 100644
--- a/Modules/FindCUDA/select_compute_arch.cmake
+++ b/Modules/FindCUDA/select_compute_arch.cmake
@@ -25,44 +25,26 @@ if(CMAKE_CUDA_COMPILER_LOADED) # CUDA as a language
 endif()
 
 # See: https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#gpu-feature-list
+# Additions, deprecations, and removals can be found in the release notes:
+# https://developer.nvidia.com/cuda-toolkit-archive
 
-# This list will be used for CUDA_ARCH_NAME = All option
-set(CUDA_KNOWN_GPU_ARCHITECTURES  "Fermi" "Kepler" "Maxwell")
+# The initial status here is for CUDA 7.0
+set(CUDA_KNOWN_GPU_ARCHITECTURES  "Fermi" "Kepler" "Maxwell" "Kepler+Tegra" "Kepler+Tesla" "Maxwell+Tegra")
+set(CUDA_COMMON_GPU_ARCHITECTURES "2.0" "2.1" "3.0" "3.5" "5.0" "5.3")
+set(CUDA_LIMIT_GPU_ARCHITECTURE "6.0")
+set(CUDA_ALL_GPU_ARCHITECTURES "2.0" "2.1" "3.0" "3.2" "3.5" "3.7" "5.0" "5.2" "5.3")
+set(_CUDA_MAX_COMMON_ARCHITECTURE "5.2+PTX")
 
-# This list will be used for CUDA_ARCH_NAME = Common option (enabled by default)
-set(CUDA_COMMON_GPU_ARCHITECTURES "3.5" "5.0")
-# 3.0 is removed in CUDA 11, see:
-# https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#deprecated-features
-if(CUDA_VERSION VERSION_LESS "11.0")
-  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "3.0")
-endif()
-
-if(CUDA_VERSION VERSION_LESS "7.0")
-  set(CUDA_LIMIT_GPU_ARCHITECTURE "5.2")
-endif()
-
-# This list is used to filter CUDA archs when autodetecting
-set(CUDA_ALL_GPU_ARCHITECTURES "3.0" "3.2" "3.5" "5.0")
-
-if(CUDA_VERSION VERSION_GREATER_EQUAL "7.0")
-  list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Kepler+Tegra" "Kepler+Tesla" "Maxwell+Tegra")
-  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "5.2")
-
-  if(CUDA_VERSION VERSION_LESS "8.0")
-    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "5.2+PTX")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "6.0")
-  endif()
-endif()
 
 if(CUDA_VERSION VERSION_GREATER_EQUAL "8.0")
   list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Pascal")
   list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "6.0" "6.1")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "6.0" "6.1" "6.2")
 
-  if(CUDA_VERSION VERSION_LESS "9.0")
-    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "6.2+PTX")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "7.0")
-  endif()
+  set(_CUDA_MAX_COMMON_ARCHITECTURE "6.2+PTX")
+  set(CUDA_LIMIT_GPU_ARCHITECTURE "7.0")
+
+  list(REMOVE_ITEM CUDA_COMMON_GPU_ARCHITECTURES "2.0" "2.1")
 endif ()
 
 if(CUDA_VERSION VERSION_GREATER_EQUAL "9.0")
@@ -70,10 +52,11 @@ if(CUDA_VERSION VERSION_GREATER_EQUAL "9.0")
   list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "7.0")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "7.0" "7.2")
 
-  if(CUDA_VERSION VERSION_LESS "10.0")
-    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "7.2+PTX")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
-  endif()
+  set(_CUDA_MAX_COMMON_ARCHITECTURE "7.2+PTX")
+  set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
+
+  list(REMOVE_ITEM CUDA_KNOWN_GPU_ARCHITECTURES "Fermi")
+  list(REMOVE_ITEM CUDA_ALL_GPU_ARCHITECTURES "2.0" "2.1")
 endif()
 
 if(CUDA_VERSION VERSION_GREATER_EQUAL "10.0")
@@ -81,32 +64,35 @@ if(CUDA_VERSION VERSION_GREATER_EQUAL "10.0")
   list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "7.5")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "7.5")
 
-  if(CUDA_VERSION VERSION_LESS "11.0")
-    list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "7.5+PTX")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
-  endif()
+  set(_CUDA_MAX_COMMON_ARCHITECTURE "7.5+PTX")
+  set(CUDA_LIMIT_GPU_ARCHITECTURE "8.0")
+
+  list(REMOVE_ITEM CUDA_COMMON_GPU_ARCHITECTURES "3.0")
 endif()
 
+# https://docs.nvidia.com/cuda/archive/11.0/cuda-toolkit-release-notes/index.html#cuda-general-new-features
+# https://docs.nvidia.com/cuda/archive/11.0/cuda-toolkit-release-notes/index.html#deprecated-features
 if(CUDA_VERSION VERSION_GREATER_EQUAL "11.0")
   list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Ampere")
   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()
+  set(_CUDA_MAX_COMMON_ARCHITECTURE "8.0+PTX")
+  set(CUDA_LIMIT_GPU_ARCHITECTURE "8.6")
+
+  list(REMOVE_ITEM CUDA_COMMON_GPU_ARCHITECTURES "3.5" "5.0")
+  list(REMOVE_ITEM CUDA_ALL_GPU_ARCHITECTURES "3.0" "3.2")
 endif()
 
 if(CUDA_VERSION VERSION_GREATER_EQUAL "11.1")
-  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.6" "8.6+PTX")
+  list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.6")
   list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.6")
 
-  if(CUDA_VERSION VERSION_LESS "12.0")
-    set(CUDA_LIMIT_GPU_ARCHITECTURE "9.0")
-  endif()
+  set(_CUDA_MAX_COMMON_ARCHITECTURE "8.6+PTX")
+  set(CUDA_LIMIT_GPU_ARCHITECTURE "9.0")
 endif()
 
+list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "${_CUDA_MAX_COMMON_ARCHITECTURE}")
 
 # Check with: cmake -DCUDA_VERSION=7.0 -P select_compute_arch.cmake
 if(DEFINED CMAKE_SCRIPT_MODE_FILE)