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

[Louis Dionne via libcxx-commits]

================
@@ -0,0 +1,54 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+// This test will fail if the number of devices detected by OpenMP is larger
+// than zero but std::for_each(std::execution::par_unseq,...) is not executed on
+// the device.
+
+// UNSUPPORTED: c++03, c++11, c++14, gcc
+
+// ADDITIONAL_COMPILE_FLAGS: -fno-exceptions --offload-arch=native -L%{lib}/../../lib -lomptarget -L%{lib}/../../projects/openmp/libomptarget/ -lomptarget.devicertl
+
+// REQUIRES: openmp_pstl_backend
+
+#include <algorithm>
+#include <cassert>
+#include <execution>
+#include <vector>
+#include <omp.h>
+
+int main(int, char**) {
+  // We only run the test if a device is detected by OpenMP
+  if (omp_get_num_devices() < 1)
+    return 0;
----------------
ldionne wrote:

This seems wrong to me. We should always be able to run that test, and it might or might not actually use a GPU depending on whether there's one.

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