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

[Louis Dionne via llvm-commits]

================
@@ -179,10 +180,17 @@ struct __select_backend<std::execution::parallel_policy> {
   using type = __cpu_backend_tag;
 };
 
+#    if defined(_LIBCPP_PSTL_GPU_OFFLOAD)
----------------
ldionne wrote:

I think I would do this here:

```
#  if defined(_LIBCPP_PSTL_BACKEND_SERIAL) || defined(_LIBCPP_PSTL_BACKEND_THREAD) ||                          \
      defined(_LIBCPP_PSTL_BACKEND_LIBDISPATCH)

template <>
struct __select_backend<std::execution::parallel_policy> {
  using type = __cpu_backend_tag;
};

template <>
struct __select_backend<std::execution::parallel_unsequenced_policy> {
  using type = __cpu_backend_tag;
};

#elif defined(_LIBCPP_PSTL_BACKEND_OPENMP)

template <>
struct __select_backend<std::execution::parallel_policy> {
  using type = __cpu_backend_tag;
};

template <>
struct __select_backend<std::execution::parallel_unsequenced_policy> {
  using type = __openmp_backend_tag;
};

#  else

// ...New vendors[...]

#endif
```

Then, from `libcxx/include/__algorithm/pstl_backends/cpu_backends/backend.h` we should do this instead:

```
#if defined(_LIBCPP_PSTL_BACKEND_SERIAL) || defined(_LIBCPP_PSTL_BACKEND_OPENMP)
#  include <__algorithm/pstl_backends/cpu_backends/serial.h>
// etc...
```

We'll refactor this eventually. I get the feeling that we'll want to remove the notion of a multi-level backend selection but we can cross that bridge when we get to it.


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