Running error of SSE2 code in VS2013

Question

I have the following SIMD code trying to run in vs2013. It can be well compiled but cannot run. Anyone knows why?

#include <cstdio>
#include <xmmintrin.h>

int main() 
{
    const size_t num = 7;
    float a[num] = { 1, 2, 3, 4, 5, 6, 7 };
    float b[num] = { 1, -1, -2, 1, -3, -2, 5 };
    float c[num];
    __m128 A, B, C;
    A = _mm_load_ps(&a[0]); // <== crash here.
    B = _mm_load_ps(&b[0]);
    C = _mm_add_ps(A, B);
    _mm_store_ps(&c[0], C);

    return 0;
}

Either align your data or use `_mm_loadu_ps`/`_mm_storeu_ps`. — Paul R, May 12 '16 at 07:29

score 1 · Answer 1 · answered May 11 '16 at 19:10

1

The address being loaded from or stored to using these intrinsics needs to be 16 byte aligned (divisible by 16). See https://msdn.microsoft.com/en-us/library/zzd50xxt(v=vs.90).aspx

You should declare the variables a,b and c like this:

__declspec(align(16)) float a[num] = { 1, 2, 3, 4, 5, 6, 7 };

answered May 11 '16 at 19:10

Sami Sallinen

3,203
12
16

What if using g++? Is __declspec(align(16)) still necessary? – C. Wang May 11 '16 at 19:28
1

On gcc it is __attribute __((aligned(16))) - see http://stackoverflow.com/questions/841433/gcc-attribute-alignedx-explanation – Sami Sallinen May 11 '16 at 19:34

Running error of SSE2 code in VS2013

1 Answers1