commit c2462eaedc665b5fea5d6ef3576dc5f29f456fce
author Xing Xue <xingxue@outlook.com> Fri Apr 19 09:08:31 2024 -0400
committer Copybara-Service <copybara-worker@google.com> Fri Apr 19 06:15:55 2024 -0700
tree 591edeacbe016940fade89021b7554aa75d9c171
parent a6cbdd378934200c188a5d9663058ac3041d16b2

[OpenMP] Use half of available logical processors for collapse tests (#88319)

The new collapse test cases define `MAX_THREADS` to be 256 and use all
available threads/logical processors on the system. This triples the
testing time on an AIX machine that has 128 logical processors. This
patch changes to use half of available logical processors to avoid over
subscribing because there are other libomp tests running at the same
time, including 2 other such collapse tests.

GitOrigin-RevId: 0a8cd1ed1f4f35905df318015b0dbcb69d81d7c2

runtime/test/worksharing/for/collapse_test.inc

diff --git a/runtime/test/worksharing/for/collapse_test.inc b/runtime/test/worksharing/for/collapse_test.inc
index de0e7e4..3075bd0 100644
--- a/runtime/test/worksharing/for/collapse_test.inc
+++ b/runtime/test/worksharing/for/collapse_test.inc
@@ -18,9 +18,9 @@
 #define MAX_THREADS 256
 
 #if defined VERBOSE
-#define PRINTF printf
+#define PRINTF(...) printf(__VA_ARGS__)
 #else
-#define PRINTF
+#define PRINTF(...)
 #endif
 
 LOOP_TYPE0 iLB, iUB;
@@ -106,12 +106,21 @@
   unsigned scalarCount = 0;
   unsigned uselessThreadsOpenMP = 0;
   unsigned usefulThreadsOpenMP = 0;
-  unsigned chunkSizesOpenmp[MAX_THREADS] = {0};
 
-  unsigned num_threads = omp_get_max_threads();
+  // Use half of the available threads/logical processors.
+  unsigned num_threads = omp_get_max_threads() / 2;
+
+  // Make sure num_threads is not 0 after the division in case
+  // omp_get_max_threads() returns 1.
+  if (num_threads == 0)
+    num_threads = 1;
+
   if (num_threads > MAX_THREADS)
     num_threads = MAX_THREADS;
-  omp_set_num_threads(num_threads);
+
+  unsigned long *chunkSizesOpenmp =
+      (unsigned long *)malloc(sizeof(unsigned long) * num_threads);
+  memset(chunkSizesOpenmp, 0, sizeof(unsigned long) * num_threads);
 
   // count iterations and allocate space
   LOOP { ++trueCount; }
@@ -129,10 +138,10 @@
   // perform and record OpenMP iterations and thread use
 #pragma omp parallel num_threads(num_threads)
   {
+    unsigned gtid = omp_get_thread_num();
 #pragma omp for collapse(3) private(i, j, k)
     LOOP {
       unsigned count;
-      unsigned gtid = omp_get_thread_num();
 #pragma omp atomic update
       ++chunkSizesOpenmp[gtid];
 #pragma omp atomic capture
@@ -197,5 +206,6 @@
   // clean up space
   FreeSpace(openmpSpace);
   FreeSpace(scalarSpace);
+  free(chunkSizesOpenmp);
   return pass;
 }