[Openmp-commits] [openmp] [OpenMP] Fix distributed barrier hang for OMP_WAIT_POLICY=passive (PR #83058)

[Jonathan Peyton via Openmp-commits]

https://github.com/jpeyton52 created https://github.com/llvm/llvm-project/pull/83058

The resume thread logic inside __kmp_free_team() is faulty. Only checking b_go for sleep status doesn't wake up distributed barrier. Change to generic check for th_sleep_loc and calling __kmp_null_resume_wrapper().

Fixes: #80664

>From d10e1b505a4cbd7632d2bc83de9a00a2cc83b9c2 Mon Sep 17 00:00:00 2001
From: Jonathan Peyton <jonathan.l.peyton at intel.com>
Date: Mon, 26 Feb 2024 14:23:46 -0600
Subject: [PATCH] [OpenMP] Fix distributed barrier hang for
 OMP_WAIT_POLICY=passive

The resume thread logic inside __kmp_free_team() is faulty. Only
checking b_go for sleep status doesn't wake up distributed barrier.
Change to generic check for th_sleep_loc and calling
__kmp_null_resume_wrapper().

Fixes: #80664
---
 openmp/runtime/src/kmp_runtime.cpp            |  5 ++-
 .../runtime/test/barrier/llvm-issue-80664.c   | 34 +++++++++++++++++++
 2 files changed, 36 insertions(+), 3 deletions(-)
 create mode 100644 openmp/runtime/test/barrier/llvm-issue-80664.c

diff --git a/openmp/runtime/src/kmp_runtime.cpp b/openmp/runtime/src/kmp_runtime.cpp
index fc5e8405a415ed..7edb0b440acc7f 100644
--- a/openmp/runtime/src/kmp_runtime.cpp
+++ b/openmp/runtime/src/kmp_runtime.cpp
@@ -5708,9 +5708,8 @@ void __kmp_free_team(kmp_root_t *root,
           }
 #endif
           // first check if thread is sleeping
-          kmp_flag_64<> fl(&th->th.th_bar[bs_forkjoin_barrier].bb.b_go, th);
-          if (fl.is_sleeping())
-            fl.resume(__kmp_gtid_from_thread(th));
+          if (th->th.th_sleep_loc)
+            __kmp_null_resume_wrapper(th);
           KMP_CPU_PAUSE();
         }
       }
diff --git a/openmp/runtime/test/barrier/llvm-issue-80664.c b/openmp/runtime/test/barrier/llvm-issue-80664.c
new file mode 100644
index 00000000000000..c56052eafbea9d
--- /dev/null
+++ b/openmp/runtime/test/barrier/llvm-issue-80664.c
@@ -0,0 +1,34 @@
+// RUN: %libomp-compile
+// RUN: env OMP_WAIT_POLICY=passive KMP_FORKJOIN_BARRIER_PATTERN='linear,linear' %libomp-run
+// RUN: env OMP_WAIT_POLICY=passive KMP_FORKJOIN_BARRIER_PATTERN='tree,tree' %libomp-run
+// RUN: env OMP_WAIT_POLICY=passive KMP_FORKJOIN_BARRIER_PATTERN='hyper,hyper' %libomp-run
+// RUN: env OMP_WAIT_POLICY=passive KMP_FORKJOIN_BARRIER_PATTERN='dist,dist' %libomp-run
+//
+// LLVM ISSUE 80664: https://github.com/llvm/llvm-project/issues/80664
+//
+// Distributed barrier + OMP_WAIT_POLICY=passive hangs in library termination
+// Reason: the resume logic in __kmp_free_team() was faulty and, when checking
+// for sleep status, didn't look at correct location for distributed barrier.
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int a = 0;
+
+void test_omp_barrier()
+{
+  #pragma omp parallel
+  {
+    #pragma omp task
+    {
+      #pragma omp atomic
+      a++;
+    }
+  }
+}
+
+int main() {
+  test_omp_barrier();
+  printf("a = %d\n", a);
+  return EXIT_SUCCESS;
+}