diff --git a/runtime/test/worksharing/for/omp_for_bigbounds.c b/runtime/test/worksharing/for/omp_for_bigbounds.c
index adc00ff..c6f1693 100644
--- a/runtime/test/worksharing/for/omp_for_bigbounds.c
+++ b/runtime/test/worksharing/for/omp_for_bigbounds.c
@@ -7,6 +7,9 @@
 
 // XFAIL: gcc, clang-3, clang-4, clang-5, clang-6, clang-7, clang-8, clang-9, clang-10
 
+// icc 21 seems to have an issue with the loop boundaries and runs very long
+// UNSUPPORTED: icc-21
+
 /*
  * Test that large bounds are handled properly and calculations of
  * loop iterations don't accidentally overflow
@@ -17,8 +20,8 @@
 #include <limits.h>
 #include "omp_testsuite.h"
 
-#define INCR 50000000
-#define MY_MAX 2000000000
+#define INCR      50000000
+#define MY_MAX  2000000000
 #define MY_MIN -2000000000
 #ifndef MY_SCHEDULE
 # define MY_SCHEDULE static
@@ -33,14 +36,12 @@
   #pragma omp parallel
   {
     int i;
-    #pragma omp for schedule(MY_SCHEDULE)
+    #pragma omp for schedule(MY_SCHEDULE) reduction(+:a)
     for (i = INT_MIN; i < MY_MAX; i+=INCR) {
-        #pragma omp atomic
         a++;
     }
-    #pragma omp for schedule(MY_SCHEDULE)
+    #pragma omp for schedule(MY_SCHEDULE) reduction(+:b)
     for (i = INT_MAX; i >= MY_MIN; i-=INCR) {
-        #pragma omp atomic
         b++;
     }
   }