/* gcc -W -Wall -fPIC -c test2.c -o test2.o */

#include <stdint.h>

static void f(uint8_t *q, uint8_t *p, int y, int x)
{
    asm volatile(
        "movd  %0, %%mm0		\n\t"
        "movd  %1, %%mm1		\n\t"
        "movd  %2, %%mm2		\n\t"
        "movd  %3, %%mm3		\n\t"

        :  "m" (*(uint32_t*)(p + 0*x)),
           "m" (*(uint32_t*)(p + 1*x)),
           "m" (*(uint32_t*)(p + 2*x)),
           "m" (*(uint32_t*)(p + 3*x))
    );

    asm volatile(
        "# Lots of stuff going on here	\n\t"
        "movd  %%mm0, %0		\n\t"
        "movd  %%mm1, %1		\n\t"
        "movd  %%mm2, %2		\n\t"
        "movd  %%mm3, %3		\n\t"

        : "=m" (*(uint32_t*)(q + 0*y)),
          "=m" (*(uint32_t*)(q + 1*y)),
          "=m" (*(uint32_t*)(q + 2*y)),
          "=m" (*(uint32_t*)(q + 3*y))
    );
}