[Bug target/123978] New: [16 Regression] leptonica miscompilation on aarch64 since r16-1398

[jakub at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123978

            Bug ID: 123978
           Summary: [16 Regression] leptonica miscompilation on aarch64
                    since r16-1398
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org
  Target Milestone: ---

struct A { unsigned b, c, *d; };

[[gnu::noipa]] int
foo (struct A *a)
{
  __builtin_memset (a->d, 0, ((long long) sizeof (unsigned)) * a->b * a->c);
  return 0;
}

int
main ()
{
  struct A a;
  unsigned b[1024];
  __builtin_memset (b, 0x55, sizeof (b));
  a.b = 31;
  a.c = 31;
  a.d = b;
  foo (&a);
  for (int i = 0; i < 961; ++i)
    if (b[i] != 0)
      __builtin_abort ();
}

is miscompiled with -O2 starting with my
r16-1398-g8154fc95f097a146f9c80edcaafb2baff73065b5 ranger use during expansion
patch on aarch64-linux.
The memset call is somehow optimized away for some reason.
optimized dump has:
  _1 = a_10(D)->d;
  _2 = a_10(D)->b;
  _4 = a_10(D)->c;
  # RANGE [irange] long long int [0, +INF] MASK 0x7fffffffffffffff VALUE 0x0
  _6 = _2 w* _4;
  # RANGE [irange] long long int [0, 0][4, 9223372036854775804] MASK
0x7ffffffffffffffc VALUE 0x0
  _7 = _6 * 4;
  # RANGE [irange] long unsigned int [0, 0][4, 9223372036854775804] MASK
0x7ffffffffffffffc VALUE 0x0
  _8 = (long unsigned int) _7;
  __builtin_memset (_1, 0, _8);
  return 0;