[Bug c/126976] New: forwprop introduce the result divergence

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126976

            Bug ID: 126976
           Summary: forwprop introduce the result divergence
           Product: gcc
           Version: 16.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kyuwoncho18 at gmail dot com
  Target Milestone: ---

Created attachment 65380
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65380&action=edit
same code as I include on description

Consider the code:
#include <stdio.h>

typedef signed char   v8qi  __attribute__((vector_size(8)));
typedef unsigned char v8uqi __attribute__((vector_size(8)));
typedef short         v8hi  __attribute__((vector_size(16)));

__attribute__((noipa))
v8hi f(v8qi A, v8qi B) {
  v8uqi Au = (v8uqi)A;
  v8hi r = { Au[0], Au[1], Au[2], Au[3], B[4], B[5], B[6], B[7] };
  return r;
}

int main(void) {
  v8qi A = { -1, -2, 3, 4, 5, 6, 7, 8 };
  v8qi B = { 10, 20, 30, 40, -50, -60, -70, -128 };
  v8hi r = f(A, B);
  short out[8];
  __builtin_memcpy(out, &r, sizeof out);
  for (int i = 0; i < 8; i++) printf("%d ", out[i]);
  printf("\n");
  return 0;
}

And with O0, it prints "255 254 3 4 -50 -60 -70 -128", but with O2, it prints
"255 254 3 4 206 196 186 128".

Godbolt: https://godbolt.org/z/jj1ba1584
Only affected after GCC 16.1; on 15.3, it can not be reproduced.