Merging r318848:
------------------------------------------------------------------------
r318848 | hahnfeld | 2017-11-22 09:15:20 -0800 (Wed, 22 Nov 2017) | 7 lines
Fix for OMP doacross implementation on Power
Power has a weak consistency model so we need memory barriers to
make writes (both from runtime and from user code) available for
all threads.
Differential Revision: https://reviews.llvm.org/D40175
------------------------------------------------------------------------
git-svn-id: https://llvm.org/svn/llvm-project/openmp/branches/release_50@319057 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/runtime/src/kmp_csupport.cpp b/runtime/src/kmp_csupport.cpp
index 1f61870..d6e4ff5 100644
--- a/runtime/src/kmp_csupport.cpp
+++ b/runtime/src/kmp_csupport.cpp
@@ -3108,7 +3108,9 @@
if (flags == NULL) {
// we are the first thread, allocate the array of flags
size_t size = trace_count / 8 + 8; // in bytes, use single bit per iteration
- sh_buf->doacross_flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
+ flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
+ KMP_MB();
+ sh_buf->doacross_flags = flags;
} else if (flags == (kmp_uint32 *)1) {
#if KMP_32_BIT_ARCH
// initialization is still in progress, need to wait
@@ -3117,6 +3119,9 @@
while (*(volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
#endif
KMP_YIELD(TRUE);
+ KMP_MB();
+ } else {
+ KMP_MB();
}
KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1); // check ptr value
pr_buf->th_doacross_flags =
@@ -3212,6 +3217,7 @@
while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
KMP_YIELD(TRUE);
}
+ KMP_MB();
KA_TRACE(20,
("__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
gtid, (iter_number << 5) + shft));
@@ -3264,6 +3270,7 @@
shft = iter_number % 32; // use 32-bit granularity
iter_number >>= 5; // divided by 32
flag = 1 << shft;
+ KMP_MB();
if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
KA_TRACE(20, ("__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
diff --git a/runtime/test/worksharing/for/kmp_doacross_check.c b/runtime/test/worksharing/for/kmp_doacross_check.c
index c0ee201..a088965 100644
--- a/runtime/test/worksharing/for/kmp_doacross_check.c
+++ b/runtime/test/worksharing/for/kmp_doacross_check.c
@@ -24,7 +24,7 @@
dims.lo = 1;
dims.up = N-1;
dims.st = 1;
- #pragma omp parallel
+ #pragma omp parallel num_threads(4)
{
int i, gtid;
long long vec;
