https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127649
Bug ID: 127649
Summary: wrong code at -O{2,3} loses FE_INVALID on a quiet NaN
vectorization
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: kyuwoncho18 at gmail dot com
Target Milestone: ---
#include <stdio.h>
#include <fenv.h>
#define N 64
static float a[N], b[N];
static int r[N];
static void ltgt (float *x, float *y, int *z, int n)
{
for (int i = 0; i < n; i++)
z[i] = (x[i] < y[i]) | (x[i] > y[i]);
}
int main (void)
{
for (int i = 0; i < N; i++)
{
a[i] = __builtin_nanf (""); /* quiet NaN */
b[i] = 1.0f;
}
feclearexcept (FE_ALL_EXCEPT);
ltgt (a, b, r, N);
printf ("FE_INVALID = %d\n", fetestexcept (FE_INVALID));
return 0;
}
The code with -O{0,1} sets FE_INVALID after the ltgt call because a is NaN, but
the flag is not set with -O{2,3}.
https://gcc.godbolt.org/z/8q8on8a1e
