https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122643
Bug ID: 122643
Summary: Code generated with -O1 -fexpensive-optimizations
produces NaN on aarch64
Product: gcc
Version: 15.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: rnhmjoj at eurofusion dot eu
Target Milestone: ---
The following program compiled with -O1 -fexpensive-optimizations on aarch64
outputs: "nan, 0.000000". Instead, when compiled without optimizations or on
other architectures (with or without optimizations) the output is the expected
one: "0.000000 0.000000".
#include <stdio.h>
#include <math.h>
void main() {
int n = 41;
double x;
for (int i = 1; i <= n; i++) {
x = sqrt(1 - (i-1) * (1.0/(n-1)));
}
printf("%f %f\n", x, sqrt(1 - (n-1) * (1.0/(n-1))));
}
Note: the problem was actually discovered in a Fortran program compiled with
gfortran, so it's not specific to the C front end.
