Use native GPU architecture when NVIDIA GPU is detected; otherwise fall back to multi-arch build. (#732)

This change makes MSCCL++ automatically select CUDA architectures based
on the build environment. If an NVIDIA GPU is detected, the build
targets the native GPU architecture for optimal performance; otherwise,
it falls back to building for multiple architectures for portability.
When building for the native architecture, FP8 support is automatically
enabled for “a-series” GPUs (e.g., sm_100a), allowing the appropriate
optimized code paths to be picked up.

Author: mahdiehghazim

CMakeLists.txt

@@ -106,18 +106,22 @@ if(MSCCLPP_GPU_ARCHS)
         message(FATAL_ERROR "MSCCLPP_GPU_ARCHS is empty. Specify GPU architectures or leave unset.")
     endif()
 elseif(MSCCLPP_USE_CUDA)
-    if(CUDAToolkit_VERSION VERSION_LESS "11.8")
-        message(FATAL_ERROR "CUDA 11.8 or higher required, found ${CUDAToolkit_VERSION}")
-    endif()
-    set(MSCCLPP_GPU_ARCHS 80)
-    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.0")
-        list(APPEND MSCCLPP_GPU_ARCHS 90)
-    endif()
-    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
-        list(APPEND MSCCLPP_GPU_ARCHS 100)
-    endif()
-    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.9")
-        list(APPEND MSCCLPP_GPU_ARCHS 120)
+    if(NVIDIA_FOUND)
+        set(MSCCLPP_GPU_ARCHS "native")
+    else()
+	    if(CUDAToolkit_VERSION VERSION_LESS "11.8")
+	        message(FATAL_ERROR "CUDA 11.8 or higher required, found ${CUDAToolkit_VERSION}")
+	    endif()
+	    set(MSCCLPP_GPU_ARCHS 80)
+	    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.0")
+	        list(APPEND MSCCLPP_GPU_ARCHS 90)
+	    endif()
+	    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
+	        list(APPEND MSCCLPP_GPU_ARCHS 100)
+	    endif()
+	    if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.9")
+	        list(APPEND MSCCLPP_GPU_ARCHS 120)
+	    endif()
     endif()
 elseif(MSCCLPP_USE_ROCM)
     set(MSCCLPP_GPU_ARCHS gfx90a gfx941 gfx942)