https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120937
Bug ID: 120937
Summary: floating-point arithmetic can result zero with
incorrect sign
Product: gcc
Version: 13.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: madams846 at hotmail dot com
Target Milestone: ---
The result of a*b+c can be zero with incorrect negative sign for some values
of a, b, c.
The program
#include <stdio.h>
int main() {
double a= 5.174463056948843312e-23;
double b= 7.778216080286462531e-19;
double c= -4.024809449233739591e-41;
printf(" a*b+c %f \n", a*b+c );
return 0 ;
}
compiled with “gcc program.c” prints out “a*b+c -0.000000”. Another set of
values that returns negative zero is a= -2.542400512695312500e+02, b=
-8.732892029672259986e-42, c= -2.220250918039383963e-39. The result is
negative zero also for these values as types long double and __float128 BUT not
for type float. The same program with a= 2.0 b=2.0 c= -4.0 prints out
“a*b+c 0.000000”.
According to the Wikipedia page on signed zero the IEEE 754 – 2008 rules have
x – x = x + (-x) = +0 for any finite x except when rounding toward negative.
This bug seems to be new. It arose when testing a larger program with random
values. This bug did not appear on my system last week but now appears
consistently in testing. I run sudo apt update and upgrade on Linux about once
a week.
My system:
Ubuntu 24.04.2 LTS (GNU/Linux 6.6.87.1-microsoft-standard-WSL2 x86_64)
WSL 5.10.102.1
Dell PC 64-bit 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz
Windows 11 Home 24H2 OS build 26100.4351
gcc version 13.3.0-6ubuntu2~24.04
