https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85804
Bug ID: 85804
Summary: [8/9 Regression][AArch64] Mis-compilation of loop with
strided array access and xor reduction
Product: gcc
Version: 8.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: sudi at gcc dot gnu.org
Target Milestone: ---
The following test case:
#include <stdio.h>
long d[32] = {0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
int main() {
int b = 0;
for (int c = 0; c <= 5; c++)
b ^= d[c * 5 + 1];
printf("checksum = %x\n", b);
}
when compiled with:
aarch64-none-linux-gnu-gcc -O3 f.c
prints "checksum = 1".
All the elements being xor'd (1,6,11,16,21,26) are 0s and thus the result
should also be 0.
The assembly for main looks like:
main:
.LFB11:
.cfi_startproc
adrp x1, .LANCHOR0
add x1, x1, :lo12:.LANCHOR0
stp x29, x30, [sp, -16]!
.cfi_def_cfa_offset 16
.cfi_offset 29, -16
.cfi_offset 30, -8
adrp x0, .LC0
add x0, x0, :lo12:.LC0
mov x29, sp
ldr q1, [x1, 8]
ldr q2, [x1, 24]
ldr q0, [x1, 40]
xtn v2.2s, v2.2d
xtn v1.2s, v1.2d
xtn v0.2s, v0.2d
eor v1.8b, v1.8b, v2.8b
eor v0.8b, v0.8b, v1.8b
ushr d1, d0, 32
eor v0.8b, v0.8b, v1.8b
umov w1, v0.s[0]
bl printf
mov w0, 0
ldp x29, x30, [sp], 16
.cfi_restore 30
.cfi_restore 29
.cfi_def_cfa_offset 0
ret
.cfi_endproc
This goes away with -fno-tree-loop-vectorize.
