https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128149
Bug ID: 128149
Summary: wrong code with -fopenmp at -O1 and above: a
100-iteration loop with schedule(simd:dynamic,
LONG_MAX) does not terminate
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: kyuwoncho18 at gmail dot com
Target Milestone: ---
Target: x86_64-pc-linux-gnu
#include <limits.h>
#include <stdio.h>
int main(void)
{
int count = 0;
#pragma omp parallel for simd num_threads(2) schedule(simd:dynamic, LONG_MAX)
reduction(+:count)
for (long i = 0; i < 100; ++i)
count++;
printf("%d\n", count);
return count != 100;
}
godbolt: https://godbolt.org/z/bTY1nY5rn
On x86-64, -O0 -fopenmp prints 100 and exits normally, while -O1, -O2, -O3 and
-Os with -fopenmp produce no output and hit Compiler Explorer's 20-second
execution limit. The loop has 100 iterations and a positive, constant chunk
size, so the reduction must produce 100 and the program must terminate. With
-O1 -fopenmp, either removing the simd schedule modifier or changing the chunk
size to LONG_MAX - 15 makes it print 100 and exit normally.
