https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127367
Bug ID: 127367
Summary: [17 Regression] RISC-V: wrong code for unsigned
_BitInt(96) %
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: shimizu2486 at gmail dot com
Target Milestone: ---
Target: riscv64-unknown-linux-gnu
Live Reproducer: https://godbolt.org/z/n9dKerYYE
An unsigned _BitInt(96) modulo gives the wrong value from -O2 up.
$ cat t1.c
#include <stdio.h>
_BitInt(5) g;
_BitInt(65) prod;
int main(void) {
_BitInt(6) t = ~g;
_BitInt(96) m = t % (unsigned _BitInt(96))362;
_BitInt(5) r = __builtin_mul_overflow(t, m, &prod) ?: prod;
printf("m=%lld r=%d prod=%lld\n", (long long)m, (int)r, (long long)prod);
}
g is zero, so t is -1, (unsigned _BitInt(96))t is 2^96-1, and m must be
(2^96-1) % 362 == 297.
$ riscv64-unknown-linux-gnu-gcc -O0 -static t1.c -o t1 && qemu-riscv64 ./t1
m=297 r=-9 prod=-297 (correct)
$ riscv64-unknown-linux-gnu-gcc -O2 -static t1.c -o t1 && qemu-riscv64 ./t1
m=125 r=3 prod=-125 (wrong)
125 is (2^128-1) % 362, so I assume the 96-bit value is compiled into a 128-bit
one.
16.1.0 correct
trunk (83c53762) wrong
-O0 and -O1 are correct; -O2, -O3 and -Os are wrong.
