This patch fixes a bug in which testcases using thread_limit larger than
the number of physical threads would crash with a memory fault. This was
exacerbated in testcases with a lot of register pressure because the
autoscaling reduces the number of physical threads to compensate for the
increased resource usage.
Committed to devel/omp/gcc-11.
@ Thomas, this should probably be folded into another patch when
upstreaming OG11 to mainline.
Andrew
libgomp amdgcn: Fix issues with dynamic OpenMP thread scaling
libgomp/ChangeLog:
* config/gcn/bar.h (gomp_barrier_init): Limit thread count to the
actual physical number.
* config/gcn/team.c (gomp_team_start): Don't attempt to set up
threads that do not exist.
diff --git a/libgomp/config/gcn/bar.h b/libgomp/config/gcn/bar.h
index bbd3141837f..63e803bd72b 100644
--- a/libgomp/config/gcn/bar.h
+++ b/libgomp/config/gcn/bar.h
@@ -55,6 +55,9 @@ typedef unsigned int gomp_barrier_state_t;
static inline void gomp_barrier_init (gomp_barrier_t *bar, unsigned count)
{
+ unsigned actual_thread_count = __builtin_gcn_dim_size (1);
+ if (count > actual_thread_count)
+ count = actual_thread_count;
bar->total = count;
bar->awaited = count;
bar->awaited_final = count;
diff --git a/libgomp/config/gcn/team.c b/libgomp/config/gcn/team.c
index 627210ea407..6aa74744315 100644
--- a/libgomp/config/gcn/team.c
+++ b/libgomp/config/gcn/team.c
@@ -187,6 +187,10 @@ gomp_team_start (void (*fn) (void *), void *data, unsigned
nthreads,
if (nthreads == 1)
return;
+ unsigned actual_thread_count = __builtin_gcn_dim_size (1);
+ if (nthreads > actual_thread_count)
+ nthreads = actual_thread_count;
+
/* Release existing idle threads. */
for (unsigned i = 1; i < nthreads; ++i)
{
