[Bug tree-optimization/128267] New: wrong code at -O2 with -march=x86-64-v4: a signed char shift induction is 64 again instead of 0 after 256 iterations

[kyuwoncho18 at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128267

            Bug ID: 128267
           Summary: wrong code at -O2 with -march=x86-64-v4: a signed char
                    shift induction is 64 again instead of 0 after 256
                    iterations
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kyuwoncho18 at gmail dot com
  Target Milestone: ---
            Target: x86_64-pc-linux-gnu

#include <stdio.h>

__attribute__((noipa)) void foo(signed char *a)
{
  signed char b = 64;
  for (unsigned i = 0; i < 257; ++i) {
    a[i] = b;
    b >>= 1;
  }
}

int main(void)
{
  signed char a[257];
  foo(a);
  printf("%d\n", a[256]);
  return 0;
}

godbolt: https://godbolt.org/z/nPWKGWY8h

b is 0 from the eighth iteration on, so a[256] is 0 and the program must print
0, as it does at -O0, at -O2 -march=x86-64-v3 and with -O2 -march=x86-64-v4
-fno-tree-loop-vectorize. On an AVX-512 machine, -O2 -march=x86-64-v4 and -O3
-march=x86-64-v4 print 64.