CUDA: Refactor PTX compilation flag into compiler modules

Modules/CMakeCUDAInformation.cmake

@@ -138,7 +138,7 @@ endif()
 #Specify how to compile when ptx has been requested
 if(NOT CMAKE_CUDA_COMPILE_PTX_COMPILATION)
   set(CMAKE_CUDA_COMPILE_PTX_COMPILATION
-    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> ${_CMAKE_COMPILE_AS_CUDA_FLAG} -ptx <SOURCE> -o <OBJECT>")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> ${_CMAKE_COMPILE_AS_CUDA_FLAG} ${_CMAKE_CUDA_PTX_FLAG} <SOURCE> -o <OBJECT>")
 endif()
 
 #Specify how to compile when separable compilation has been requested

Modules/Compiler/NVIDIA-CUDA.cmake

@@ -5,6 +5,7 @@ set(CMAKE_CUDA_VERBOSE_FLAG "-v")
 set(CMAKE_CUDA_VERBOSE_COMPILE_FLAG "-Xcompiler=-v")
 
 set(_CMAKE_COMPILE_AS_CUDA_FLAG "-x cu")
+set(_CMAKE_CUDA_PTX_FLAG "-ptx")
 
 if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 10.2.89)
   # The -forward-unknown-to-host-compiler flag was only