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

[Louis Dionne via flang-commits]

================
@@ -0,0 +1,50 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_GPU_BACKNEDS_FOR_EACH_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_GPU_BACKNEDS_FOR_EACH_H
+
+#include <__algorithm/for_each.h>
+#include <__algorithm/pstl_backends/cpu_backends/backend.h>
+#include <__algorithm/pstl_backends/gpu_backends/backend.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/terminate_on_exception.h>
+#include <stdio.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+template <class _ExecutionPolicy, class _ForwardIterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void
+__pstl_for_each(__gpu_backend_tag, _ForwardIterator __first, _ForwardIterator __last, _Functor __func) {
+  // It is only safe to execute for_each on the GPU, it the execution policy is
+  // parallel unsequenced, as it is the only execution policy prohibiting throwing
+  // exceptions and allowing SIMD instructions
+  if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+                __has_random_access_iterator_category_or_concept<_ForwardIterator>::value &&
+                __libcpp_is_contiguous_iterator<_ForwardIterator>::value) {
+    std::__par_backend::__parallel_for_simd_1(__first, __last - __first, __func);
+  }
+  // Else we fall back to the GPU backend
+  else {
+    std::__pstl_for_each<_ExecutionPolicy>(__cpu_backend_tag{}, __first, __last, __func);
+  }
----------------
ldionne wrote:

```suggestion
                __libcpp_is_contiguous_iterator<_ForwardIterator>::value) {
    std::__par_backend::__parallel_for_simd_1(__first, __last - __first, __func);
  }
  // Else we fall back to the CPU backend
  else {
    std::__pstl_for_each<_ExecutionPolicy>(__cpu_backend_tag{}, __first, __last, __func);
  }
```

Checking for `__has_random_access_iterator_category_or_concept` is redundant since you already check for `__libcpp_is_contiguous_iterator`.

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