[Bug tree-optimization/127969] New: wrong code with -fno-signed-zeros: _Decimal64 (x == C) ? x : C folded to C, changing the quantum

[kyuwoncho18 at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127969

            Bug ID: 127969
           Summary: wrong code with -fno-signed-zeros: _Decimal64 (x == C)
                    ? x : C folded to C, changing the quantum
           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 <string.h>

_Decimal64 f(_Decimal64 x) {
  if (x == 1.0dd)
    return x;
  return 1.0dd;
}

int main(void) {
  _Decimal64 x = 1.00dd, r = f(x);
  return memcmp(&x, &r, sizeof x) != 0;
}

godbolt: https://godbolt.org/z/vx3dzxvzh

With -O0, or with -O2 without -fno-signed-zeros, it returns 0, but with GCC -O1
and above with -fno-signed-zeros it returns 1: f(1.00dd) returns the constant
1.0dd (equal value, different quantum exponent) instead of x. -fno-ssa-phiopt
avoids it. Written as `return x == 1.0dd ? x : 1.0dd;`, f returns 1.0dd already
at -O0 -fno-signed-zeros, and -fno-ssa-phiopt then makes no difference.

Disclaimer: This test case was generated by a fuzzer, and the report was
written with LLM assistance.