From 890fafc2f7e4407dc8da824b2caee457f0a21e95 Mon Sep 17 00:00:00 2001
From: Wenkai Du <43822138+wenkaidu@users.noreply.github.com>
Date: Wed, 3 Apr 2024 11:29:21 -0700
Subject: [PATCH] rccl_prim_test: increase max number of workgroups and test
 iterations (#1132)

[ROCm/rccl commit: e8c76fd8068a6fe1c03579bd8fb00cc10092cb21]
---
 projects/rccl/tools/rccl-prim-test/rccl_prim_test.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/projects/rccl/tools/rccl-prim-test/rccl_prim_test.cpp b/projects/rccl/tools/rccl-prim-test/rccl_prim_test.cpp
index 05ea1912fc..7f3826199c 100644
--- a/projects/rccl/tools/rccl-prim-test/rccl_prim_test.cpp
+++ b/projects/rccl/tools/rccl-prim-test/rccl_prim_test.cpp
@@ -34,7 +34,7 @@ THE SOFTWARE.
 #include "copy_kernel.h"
 
 #define MAX_GPU 16
-#define MAX_WORKGROUPS 32
+#define MAX_WORKGROUPS 56
 #define THREADS 256
 #define NGPUS 2
 
@@ -377,7 +377,7 @@ int main(int argc,char* argv[])
     workgroups = atol(wg);
   printf("Benchmarking using %d workgroups\n", workgroups);
 
-  int iters = 10;
+  int iters = 1000;
   char *it = getCmdOption(argv, argv + argc, "-i");
   if (it)
     iters = atol(it);
@@ -582,8 +582,8 @@ int main(int argc,char* argv[])
   for (int op = begin_op; op < end_op; op ++) {
     const char *OpsName[] = {"Copy", "Local Copy", "Double Copy", "doublecopylocal", "Reduce", "ReduceCopy", "Read"};
     printf("\n[Testing %s]: \n", OpsName[op]);
-    // 4 warm up cycles
-    for (int j = 0; j < 4; j ++) {
+    // 20 warm up cycles
+    for (int j = 0; j < 20; j ++) {
       for (int i = 0; i < nGpu; i ++) {
 #if 0
         args[i*3] = &transfer_data[i];