[clang] [llvm] [openmp] [OpenMP] Use generic IR for the OpenMP DeviceRTL (PR #119091)

[Michał Górny via cfe-commits]

================
@@ -288,18 +258,11 @@ function(compileDeviceRTLLibrary target_cpu target_name target_triple)
   endif()
 endfunction()
 
-# Generate a Bitcode library for all the gpu architectures the user requested.
-add_custom_target(omptarget.devicertl.nvptx)
 add_custom_target(omptarget.devicertl.amdgpu)
-foreach(gpu_arch ${LIBOMPTARGET_DEVICE_ARCHITECTURES})
-  if("${gpu_arch}" IN_LIST all_amdgpu_architectures)
-    compileDeviceRTLLibrary(${gpu_arch} amdgpu amdgcn-amd-amdhsa -Xclang -mcode-object-version=none)
-  elseif("${gpu_arch}" IN_LIST all_nvptx_architectures)
-    compileDeviceRTLLibrary(${gpu_arch} nvptx nvptx64-nvidia-cuda --cuda-feature=+ptx63)
-  else()
-    message(FATAL_ERROR "Unknown GPU architecture '${gpu_arch}'")
-  endif()
-endforeach()
+compileDeviceRTLLibrary(amdgpu amdgcn-amd-amdhsa -Xclang -mcode-object-version=none)
+
+add_custom_target(omptarget.devicertl.nvptx)
+compileDeviceRTLLibrary(nvptx nvptx64-nvidia-cuda --cuda-feature=+ptx63)
----------------
mgorny wrote:

Ok, so it's not accidental/obsolete — that's all I needed to know :-).

Another thing I've noticed that clang needs `libomptarget.devicertl.a` even when no GPU target is actually used. I've worked around that for now by creating an empty `.a` file locally, but perhaps that should also be handled better somehow (I can make a PR later if you tell me how you'd like it handled).

https://github.com/llvm/llvm-project/pull/119091