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

[Joseph Huber via flang-commits]

================
@@ -145,6 +145,15 @@
 #undef RT_GPU_TARGET
 #endif
 
+/*
+ * RT_THIN_IO leaves out the I/O paths that need descriptor-io.cpp and the
+ * external unit machinery, which the native GPU builds don't include. The
+ * CUDA PTX library also sets it, from CMake.
+ */
+#if RT_GPU_TARGET && !defined(RT_DEVICE_COMPILATION)
----------------
jhuber6 wrote:

I thought we deleted the OpenMP offload build completely. Hopefully that's not still hanging around in the fork.

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