[flang-commits] [flang] [llvm] [flang-rt] Use thin I/O in the native GPU builds (PR #226307)

[Joseph Huber via flang-commits]

================
@@ -81,7 +81,7 @@ macro(enable_cuda_compilation name files)
       endforeach ()
     endif ()
     foreach (tgt IN ITEMS "obj.${name}PTX")
-      target_compile_definitions(${tgt} PRIVATE RT_CUDA_THIN_IO=1)
+      target_compile_definitions(${tgt} PRIVATE RT_THIN_IO=1)
----------------
jhuber6 wrote:

If it's a compilation through NVCC or clang, the following should still work https://godbolt.org/z/xEWscn1jb. So I'm confused why we can't just let the preprocessor handle this entirely.

https://github.com/llvm/llvm-project/pull/226307