GPU CMake: Improve architecture auto-detection

Author: davidrohr

dependencies/FindO2GPU.cmake

@@ -10,7 +10,12 @@
 # or submit itself to any jurisdiction.
 
 # NOTE!!!! - Whenever this file is changed, move it over to alidist/resources
-# FindO2GPU.cmake Version 8
+# FindO2GPU.cmake Version 9
+
+set(CUDA_COMPUTETARGET_DEFAULT_FULL 80-real 86-real 89-real 120-real 75-virtual)
+set(HIP_AMDGPUTARGET_DEFAULT_FULL gfx906;gfx908)
+set(CUDA_COMPUTETARGET_DEFAULT_MINIMAL 75-virtual)
+set(HIP_AMDGPUTARGET_DEFAULT_MINIMAL gfx906)
 
 if(NOT DEFINED ENABLE_CUDA)
   set(ENABLE_CUDA "AUTO")
@@ -32,11 +37,11 @@ if(CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
 endif()
 
 if(CUDA_COMPUTETARGET AND CUDA_COMPUTETARGET STREQUAL "default")
-  set(CUDA_COMPUTETARGET 80-real 86-real 89-real 120-real 75-virtual)
+  set(CUDA_COMPUTETARGET ${CUDA_COMPUTETARGET_DEFAULT_FULL})
 endif()
 
 if(HIP_AMDGPUTARGET AND HIP_AMDGPUTARGET STREQUAL "default")
-  set(HIP_AMDGPUTARGET gfx906;gfx908)
+  set(HIP_AMDGPUTARGET ${HIP_AMDGPUTARGET_DEFAULT_FULL})
 endif()
 
 function(set_target_cuda_arch target)
@@ -112,7 +117,7 @@ if(ENABLE_CUDA)
   if(CUDA_COMPUTETARGET)
     set(CMAKE_CUDA_ARCHITECTURES ${CUDA_COMPUTETARGET})
   else()
-    set(CMAKE_CUDA_ARCHITECTURES 75-virtual)
+    set(O2_GPU_CUDA_UPDATE_NATIVE_ARCHITECTURE 1)
   endif()
   set(CMAKE_CUDA_STANDARD ${CMAKE_CXX_STANDARD})
   set(CMAKE_CUDA_STANDARD_REQUIRED TRUE)
@@ -156,6 +161,13 @@ if(ENABLE_CUDA)
       set(CMAKE_CUDA_COMPILER OFF)
     endif()
   endif()
+  if(NOT CMAKE_CUDA_ARCHITECTURES OR O2_GPU_CUDA_UPDATE_NATIVE_ARCHITECTURE)
+    if(NOT CMAKE_CUDA_ARCHITECTURES_NATIVE STREQUAL "")
+      set(CMAKE_CUDA_ARCHITECTURES ${CMAKE_CUDA_ARCHITECTURES_NATIVE})
+    else()
+      set(CMAKE_CUDA_ARCHITECTURES ${CUDA_COMPUTETARGET_DEFAULT_MINIMAL})
+    endif()
+  endif()
   if(CMAKE_CUDA_COMPILER)
     set(CMAKE_CUDA_FLAGS "-Xcompiler \"${O2_GPU_CMAKE_CXX_FLAGS_NOSTD}\" ${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr --extended-lambda -Xcompiler -Wno-attributes -Wno-deprecated-gpu-targets ${GPUCA_CUDA_DENORMALS_FLAGS}")
     set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "-Xcompiler \"${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}\" ${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
@@ -184,7 +196,7 @@ if(ENABLE_CUDA)
     endif()
 
     set(CUDA_ENABLED ON)
-    message(STATUS "CUDA found (Version ${CMAKE_CUDA_COMPILER_VERSION})")
+    message(STATUS "CUDA found (Version ${CMAKE_CUDA_COMPILER_VERSION}, Architectures ${CMAKE_CUDA_ARCHITECTURES})")
   elseif(NOT ENABLE_CUDA STREQUAL "AUTO")
     message(FATAL_ERROR "CUDA not found (Compiler: ${CMAKE_CUDA_COMPILER})")
   else()
@@ -305,7 +317,6 @@ if(ENABLE_HIP)
   if(hip_FOUND AND hipcub_FOUND AND rocthrust_FOUND AND rocprim_FOUND AND hip_HIPCC_EXECUTABLE AND hip_HIPIFY_PERL_EXECUTABLE)
     set(HIP_ENABLED ON)
     set_target_properties(roc::rocthrust PROPERTIES IMPORTED_GLOBAL TRUE)
-    message(STATUS "HIP Found (${hip_HIPCC_EXECUTABLE} version ${hip_VERSION})")
     set(CMAKE_HIP_FLAGS "${O2_GPU_CMAKE_CXX_FLAGS_NOSTD} ${CMAKE_HIP_FLAGS} ${GPUCA_HIP_DENORMALS_FLAGS}")
     set(CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
     string(APPEND CMAKE_HIP_FLAGS " -fgpu-defer-diag -mllvm -amdgpu-enable-lower-module-lds=false -mllvm -amdgpu-function-calls=true -Wno-invalid-command-line-argument -Wno-unused-command-line-argument -Wno-invalid-constexpr -Wno-ignored-optimization-argument -Wno-unused-private-field -Wno-pass-failed ")
@@ -321,6 +332,7 @@ if(ENABLE_HIP)
     if(HIP_AMDGPUTARGET)
       set(CMAKE_HIP_ARCHITECTURES "${HIP_AMDGPUTARGET}")
     endif()
+    message(STATUS "HIP Found (${hip_HIPCC_EXECUTABLE} version ${hip_VERSION}, Architectures ${CMAKE_HIP_ARCHITECTURES})")
   else()
     set(HIP_ENABLED OFF)
   endif()