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

[Louis Dionne via flang-commits]

================
@@ -0,0 +1,142 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_OPENMP_BACKEND_TRANSFORM_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_OPENMP_BACKEND_TRANSFORM_H
+
+#include <__algorithm/pstl_backends/cpu_backends/backend.h>
+#include <__algorithm/pstl_backends/openmp/backend.h>
+#include <__algorithm/transform.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__type_traits/is_execution_policy.h>
+#include <optional>
+
+#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
+
+namespace __par_backend {
+inline namespace __omp_gpu_backend {
+
+//===----------------------------------------------------------------------===//
+// Templates for two iterators
+//===----------------------------------------------------------------------===//
+
+template <class _Tp, class _DifferenceType, class _Up, class _Function>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__omp_parallel_for_simd(_Tp* __first1, _DifferenceType __n, _Up* __first2, _Function __f) noexcept {
+  __omp_gpu_backend::__omp_map_alloc(__first2, __n);
+  __omp_gpu_backend::__omp_map_to(__first1, __n);
+#  pragma omp target teams distribute parallel for simd
+  for (_DifferenceType __i = 0; __i < __n; ++__i)
+    *(__first2 + __i) = __f(*(__first1 + __i));
+  __omp_gpu_backend::__omp_map_from(__first2, __n);
+  __omp_gpu_backend::__omp_map_free(__first1, __n);
+  return __empty{};
+}
+
+// Extracting the underlying pointer
+
+template <class _Iterator1, class _DifferenceType, class _Iterator2, class _Function>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__parallel_for_simd(_Iterator1 __first1, _DifferenceType __n, _Iterator2 __first2, _Function __f) noexcept {
+  return __omp_gpu_backend::__omp_parallel_for_simd(
+      __omp_gpu_backend::__omp_extract_base_ptr(__first1),
+      __n,
+      __omp_gpu_backend::__omp_extract_base_ptr(__first2),
+      __f);
+}
+
+//===----------------------------------------------------------------------===//
+// Templates for three iterator
+//===----------------------------------------------------------------------===//
+
+template <class _Tp, class _DifferenceType, class _Up, class _Vp, class _Function>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__omp_parallel_for_simd(_Tp* __first1, _DifferenceType __n, _Up* __first2, _Vp* __first3, _Function __f) noexcept {
----------------
ldionne wrote:

I would call this something like `__omp_transform`, since this is really implementing `transform`, not just `for`. WDYT? This could also apply to some of the other helper functions in this patch.

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