https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102224
Martin Liška <marxin at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[9/10/11/12 regession] |[9/10/11/12 regession] |wrong code for `x * |wrong code for `x * |copysign(1.0, x)` |copysign(1.0, x)` since | |r9-5298-g33142cf9cf82aa1f CC| |hjl at gcc dot gnu.org, | |marxin at gcc dot gnu.org --- Comment #9 from Martin Liška <marxin at gcc dot gnu.org> --- (In reply to Gabriel Ravier from comment #5) > Actually it seems to me like this is a GCC 9 regression, ever since this > pattern exists: GCC 9, 10 and 11 emit the exact same faulty code. I can confirm that: $ cat pr102224.c float __attribute__((noipa)) f(float x, float y) { return x * __builtin_copysignf(1.0f, x); } int main() { float a = 1.23f; volatile float b = f(a, a); __builtin_printf ("a: %.2f, b: %.2f\n", a, b); if (b != a) __builtin_abort (); return 0; } $ gcc pr102224.c -g -O3 && ./a.out a: 1.23, b: 0.00 Aborted (core dumped) Started with r9-5298-g33142cf9cf82aa1f.