[flang-commits] [flang] Adding Separate OpenMP Offloading Backend to `libcxx/include/__algorithm/pstl_backends` (PR #66968)

[Louis Dionne via flang-commits]

================
@@ -0,0 +1,349 @@
+//===----------------------------------------------------------------------===//
----------------
ldionne wrote:

Not attached to this line:

We need to discuss how exceptions are handled when we offload to the GPU. @AntonRydahl mentioned that `throw` was compiled to `__builtin_trap()`. This is going to both fail our tests and make us non-conforming. Our preference would be that the OpenMP backend doesn't try to offload if there are exceptions in the code path we're considering offloading.

Needs more discussion.

https://github.com/llvm/llvm-project/pull/66968