What is causing this segmentation fault while using SSE instructions?

Question

This problem is driving me a bit crazy. The code seems to be segmentation faulting for no good reason:

#define MULT_FLOAT4(X, Y) ({ \
    asm volatile ( \
        "movups (%0), %%xmm0\n\t" \
        "mulps (%1), %%xmm0\n\t" \
        "movups %%xmm0, (%1)" \
        :: "r" (X), "r" (Y)); })

int main(void)
{
    int before;
    float a[4] = { 10, 20, 30, 40 };
    float b[4] = { 0.1, 0.1, 0.1, 0.1 };

    /* Segmentation faults if I uncomment this line below, otherwise works fine. Why?? */
    int after;

    MULT_FLOAT4(a, b);

    return 0;
}

Note that so long as I have defined both the 'before' and 'after' variables it segmentation faults. If I just have 'before' or just have 'after' then it works fine.

I'm on Ubuntu Hardy (8.04), GCC version 4.2.4 (Ubuntu 4.2.4-1ubuntu4). Linux kernel: 2.6.24-16-generic.

Answer 1

Check the address of a and b. I suspect you'll find that they must be aligned to a 16 byte boundary in order to avoid a segfault. Adding __ attribute __((aligned(16))) after their declarations should do the trick.

That's two underscores on each side of attribute and connected to it, BTW.