https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128159
Bug ID: 128159
Summary: wrong code: conversion of a _BitInt(256) constant to
_Decimal128 rounds an exact halfway value down instead
of to even
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: yuki.tanaka at defaultcase dot de
Target Milestone: ---
typedef _BitInt(256) B;
#define E10 ((B) 10000000000)
__attribute__((noipa)) _Decimal128 a(void)
{
return (E10 * E10 * E10 * 10000 + 15) * E10 * E10 * E10 * E10;
}
int main(void)
{
if (a () != 1.000000000000000000000000000000002E74DL)
__builtin_abort ();
return 0;
}
$ gcc -O2 dfp_bitint.c -o dfp_bitint
$ ./dfp_bitint; echo $?
Aborted ./dfp_bitint
134
The value (10^34 + 15) * 10^40 has 35 significant digits and is exactly halfway
between two _Decimal128 values. Round-half-even gives
1.000000000000000000000000000000002E74DL. GCC folds the conversion in a to
1.000000000000000000000000000000001E74DL, so main aborts.
Compiler Explorer:
https://godbolt.org/z/9n11oro47
