comparison i386/dsputil_mmx.c @ 342:8635a7036395 libavcodec

* fixes problem with -funroll-loops and buggy gcc compiler

341:bf26081c373c

    /* read the pixels */
    p = block;
    pix = pixels;
    MOVQ_ZERO(mm7);
    i = 4;
    while (i) {
	__asm __volatile(
		"movq	%2, %%mm0\n\t"
		"movq	8%2, %%mm1\n\t"
		"movq	16%2, %%mm2\n\t"
		"movq	24%2, %%mm3\n\t"
		"movq	%0, %%mm4\n\t"
		"movq	%1, %%mm6\n\t"
		"movq	%%mm4, %%mm5\n\t"
		"punpcklbw %%mm7, %%mm4\n\t"
		"punpckhbw %%mm7, %%mm5\n\t"

		"packuswb %%mm1, %%mm0\n\t"
		"packuswb %%mm3, %%mm2\n\t"
		"movq	%%mm0, %0\n\t"
		"movq	%%mm2, %1\n\t"
		:"+m"(*pix), "+m"(*(pix+line_size))
		:"m"(*p)
		:"memory");
        pix += line_size*2;
        p += 16;
        i--;
    };
}

static void put_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
{
    int hh;

342:8635a7036395

    /* read the pixels */
    p = block;
    pix = pixels;
    MOVQ_ZERO(mm7);
    i = 4;
    do {
	__asm __volatile(
		"movq	(%2), %%mm0\n\t"
		"movq	8(%2), %%mm1\n\t"
		"movq	16(%2), %%mm2\n\t"
		"movq	24(%2), %%mm3\n\t"
		"movq	%0, %%mm4\n\t"
		"movq	%1, %%mm6\n\t"
		"movq	%%mm4, %%mm5\n\t"
		"punpcklbw %%mm7, %%mm4\n\t"
		"punpckhbw %%mm7, %%mm5\n\t"

		"packuswb %%mm1, %%mm0\n\t"
		"packuswb %%mm3, %%mm2\n\t"
		"movq	%%mm0, %0\n\t"
		"movq	%%mm2, %1\n\t"
		:"+m"(*pix), "+m"(*(pix+line_size))
		:"r"(p)
		:"memory");
        pix += line_size*2;
        p += 16;
    } while (--i);
}

static void put_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
{
    int hh;