http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49061
Summary: wrong code with ARM NEON intrinsics
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: hu...@free.fr
The code below generates wrong results when compiled with optimizations
enabled:
#include <stdio.h>
#include <string.h>
#include <arm_neon.h>
static void vvalidate(float *f) {
float32x4_t a0 = { 0, 1, 2, 3};
float32x4_t a1 = { 4, 5, 6, 7};
float32x4_t a2 = { 8, 9, 10, 11};
float32x4x2_t t0 = vzipq_f32(a1, a2);
t0 = vzipq_f32(a0, a2);
memcpy(f, &t0.val[0], 8*sizeof(float));
}
int main() {
float f[8];
vvalidate(f);
printf("interleaved([0 1 2 3], [8 9 10 11])=%g %g %g %g %g %g %g %g\n", f[0],
f[1], f[2], f[3], f[4], f[5], f[6], f[7]);
if (f[4] != 2) printf("BUUUUGGG -- should be 0 8 1 9 2 10 3 11\n");
return 0;
}
# gcc-4.6.0 -mfloat-abi=softfp -mfpu=neon -O0 ./neon_bug.c && ./a.out
interleaved([0 1 2 3], [8 9 10 11])=0 8 1 9 2 10 3 11
# gcc-4.6.0 -mfloat-abi=softfp -mfpu=neon -O1 ./neon_bug.c && ./a.out
interleaved([0 1 2 3], [8 9 10 11])=0 8 1 9 6 10 7 11
BUUUUGGG -- should be 0 8 1 9 2 10 3 11
The bug also happens with the gcc-4.5 (Ubuntu/Linaro 4.5.2-8ubuntu4) and
gcc-4.4 (Ubuntu/Linaro 4.4.5-15ubuntu1) that ship with ubuntu 11.04.
