https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120691
Bug ID: 120691
Summary: _Decimal128 arithmetic error under FE_UPWARD
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 program
#include <stdio.h>
#include <fenv.h>
_Decimal128 x2 = 10000 ;
int main() {
fesetround( FE_UPWARD );
_Decimal128 x1 = 9825 ;
printf(" x1/x2 %DDf ", x1 / x2 );
return 0 ;
}
compiled with gcc program.c -lm -ldfp
produces output x1/x2 9.825000 on my system. Same result for any
optimization level.
The same program produces correct output " x1/x2 0.982500 " without
fesetround or with type double constants 10000.0 9825.0 or with other
decimal types _Decimal64 _Decimal32. The same program with other x1 values
9823 or 9826 also produces correct output.
This bug arose in a larger program that converts floating points to
multi-integers for computations then back from integers to floating points, and
that larger program is supposed to work for any float type and any rounding
mode.
(The sample program above declares x2 as a global so that printf will work on
my system, reported.)
My system: gcc version 13.3.0-6ubuntu2~24.04, ldfp version 1.0.16-1ubuntu2,
libc6 version 2.39-0ubuntu8.4, installed on Ubuntu 24.04.2 LTS (GNU/Linux
6.6.87.1-microsoft-standard-WSL2 x86_64) running in WSL 5.10.102.1 on Windows
11 Home 24H2 OS build 26100.4351, on a Dell PC 64-bit 11th Gen Intel(R)
Core(TM) i5-1135G7 @ 2.40GHz .
Thanks.
