[libc-commits] [libc] [libc] Remove CUDA Toolkit dependency for NVPTX build (PR #208497)

[Joseph Huber via libc-commits]

================
@@ -73,12 +73,18 @@ set(LIBC_GPU_TARGET_ARCHITECTURE "${gpu_test_architecture}")
 set(LIBC_GPU_CODE_OBJECT_VERSION "6" CACHE STRING "AMDGPU Code object ABI to use")
 
 if(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
-  # FIXME: This is a hack required to keep the CUDA package from trying to find
-  #        pthreads. We only link the CUDA driver, so this is unneeded.
-  add_library(CUDA::cudart_static_deps IMPORTED INTERFACE)
-
-  find_package(CUDAToolkit QUIET)
-  if(CUDAToolkit_FOUND)
-    get_filename_component(LIBC_CUDA_ROOT "${CUDAToolkit_BIN_DIR}" DIRECTORY ABSOLUTE)
+  # Respect the standard CUDAToolkit_ROOT override if the user provided one,
+  # otherwise ask the compiler which CUDA installation it detected.
+  if(CUDAToolkit_ROOT)
+    set(LIBC_CUDA_ROOT "${CUDAToolkit_ROOT}")
+  else()
+    execute_process(
+      COMMAND ${CMAKE_CXX_COMPILER} -v --print-resource-dir
----------------
jhuber6 wrote:

The NVPTX build is by-contract always an up-to-date clang, which does its own CUDA detection internally, unconditionally.

https://github.com/llvm/llvm-project/pull/208497