[Openmp-commits] [openmp] [OpenMP] Add interface for flushing with memory order and scope (PR #133325)

[Hansang Bae via Openmp-commits]

https://github.com/hansangbae created https://github.com/llvm/llvm-project/pull/133325

Current specification allows user input to the flush construct via
clauses for memory order and scope.

>From 7bcdea6fe418cd804c1e017065ce5bc34365d919 Mon Sep 17 00:00:00 2001
From: Hansang Bae <hansang.bae at intel.com>
Date: Thu, 27 Mar 2025 15:00:52 -0500
Subject: [PATCH] [OpenMP] Add interface for flushing with memory order and
 scope

Current specification allows user input to the flush construct via
clauses for memory order and scope.
---
 openmp/runtime/src/dllexports                 |  1 +
 openmp/runtime/src/kmp.h                      |  2 +
 openmp/runtime/src/kmp_csupport.cpp           | 33 +++++++++++
 .../test/flush/omp_flush_acquire_release.c    | 58 +++++++++++++++++++
 4 files changed, 94 insertions(+)
 create mode 100644 openmp/runtime/test/flush/omp_flush_acquire_release.c

diff --git a/openmp/runtime/src/dllexports b/openmp/runtime/src/dllexports
index 0667d53c35a18..fae1a381ee7ec 100644
--- a/openmp/runtime/src/dllexports
+++ b/openmp/runtime/src/dllexports
@@ -221,6 +221,7 @@
 #    __kmpc_end_taskq                        128
 #    __kmpc_end_taskq_task                   129
     __kmpc_flush                            130
+    __kmpc_flush_explicit
     __kmpc_for_static_fini                  135
     __kmpc_for_static_init_4                136
     __kmpc_for_static_init_8                137
diff --git a/openmp/runtime/src/kmp.h b/openmp/runtime/src/kmp.h
index 3d34513491154..6d48f9faafca8 100644
--- a/openmp/runtime/src/kmp.h
+++ b/openmp/runtime/src/kmp.h
@@ -4204,6 +4204,8 @@ KMP_EXPORT void __kmpc_serialized_parallel(ident_t *, kmp_int32 global_tid);
 KMP_EXPORT void __kmpc_end_serialized_parallel(ident_t *, kmp_int32 global_tid);
 
 KMP_EXPORT void __kmpc_flush(ident_t *);
+KMP_EXPORT void __kmpc_flush_explicit(ident_t *, kmp_int32 order,
+                                      kmp_int32 scope);
 KMP_EXPORT void __kmpc_barrier(ident_t *, kmp_int32 global_tid);
 KMP_EXPORT kmp_int32 __kmpc_master(ident_t *, kmp_int32 global_tid);
 KMP_EXPORT void __kmpc_end_master(ident_t *, kmp_int32 global_tid);
diff --git a/openmp/runtime/src/kmp_csupport.cpp b/openmp/runtime/src/kmp_csupport.cpp
index fdbf9ff45e354..d505f4f18405b 100644
--- a/openmp/runtime/src/kmp_csupport.cpp
+++ b/openmp/runtime/src/kmp_csupport.cpp
@@ -801,6 +801,39 @@ void __kmpc_flush(ident_t *loc) {
 #endif
 }
 
+/*!
+ at ingroup SYNCHRONIZATION
+ at param loc source location information.
+ at param order memory order input from user.
+ at param scope memory scope input from user.
+Perform memory fence with explicit memory semantics.
+*/
+void __kmpc_flush_explicit(ident_t *loc, kmp_int32 order, kmp_int32 scope) {
+  // `scope` is not used on the initial device.
+  switch (order) {
+  case std::memory_order_relaxed:
+    [[fallthrough]];
+  case std::memory_order_acquire:
+    [[fallthrough]];
+  case std::memory_order_release:
+    [[fallthrough]];
+  case std::memory_order_acq_rel:
+    [[fallthrough]];
+  case std::memory_order_seq_cst:
+    std::atomic_thread_fence(static_cast<std::memory_order>(order));
+    break;
+  default:
+    KMP_BUILTIN_UNREACHABLE;
+  }
+
+#if OMPT_SUPPORT && OMPT_OPTIONAL
+  if (ompt_enabled.ompt_callback_flush) {
+    ompt_callbacks.ompt_callback(ompt_callback_flush)(
+        __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
+  }
+#endif
+}
+
 /* -------------------------------------------------------------------------- */
 /*!
 @ingroup SYNCHRONIZATION
diff --git a/openmp/runtime/test/flush/omp_flush_acquire_release.c b/openmp/runtime/test/flush/omp_flush_acquire_release.c
new file mode 100644
index 0000000000000..b3dd42d888e92
--- /dev/null
+++ b/openmp/runtime/test/flush/omp_flush_acquire_release.c
@@ -0,0 +1,58 @@
+// RUN: %libomp-compile-and-run
+// REQUIRES: clang
+
+// Test is based on OpenMP API Example (omp_5.0) acquire_release3.c
+// https://github.com/OpenMP/Examples/blob/main/synchronization/sources/acquire_release.3.c
+
+#include <stdio.h>
+#include <omp.h>
+
+typedef void ident_t;
+extern void __kmpc_flush_explicit(ident_t *, int order, int scope);
+
+int test_memorder(int write_order, int read_order) {
+  int x = 0, y = 0;
+  int num_fails = 0;
+#pragma omp parallel num_threads(2)
+  {
+    int thrd = omp_get_thread_num();
+    if (thrd == 0) {
+      x = 10;
+      __kmpc_flush_explicit(NULL, write_order, 0);
+#pragma omp atomic write // or with relaxed clause
+      y = 1;
+    } else {
+      int tmp = 0;
+      while (tmp == 0) {
+#pragma omp atomic read // or with relaxed clause
+        tmp = y;
+      }
+      __kmpc_flush_explicit(NULL, read_order, 0);
+      // printf("x = %d\n", x);  // always "x = 10"
+      if (x != 10)
+        num_fails++;
+    }
+  }
+  return num_fails;
+}
+
+int main() {
+  // Clang-based compiler has predefined macro __ATOMIC_<memory_order>.
+  int write_order[3] = {__ATOMIC_SEQ_CST, __ATOMIC_ACQ_REL, __ATOMIC_RELEASE};
+  int read_order[3] = {__ATOMIC_SEQ_CST, __ATOMIC_ACQ_REL, __ATOMIC_ACQUIRE};
+
+  // Repeat 1000 times
+  for (int n = 0; n < 1000; n++) {
+    for (int i = 0; i < 3; i++) {
+      for (int j = 0; j < 3; j++) {
+        if (test_memorder(write_order[i], read_order[j])) {
+          printf("failed\n");
+          exit(EXIT_FAILURE);
+        }
+      }
+    }
+  }
+
+  printf("passed\n");
+  return EXIT_SUCCESS;
+}