dsputil_mmx.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.7平台上编译为嵌入式图形界面操作系统。

     	 "movq %%mm0, (%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "addl %%eax, %1		\n\t"
         "addl %%eax, %2       		\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq %%mm0, (%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "addl %%eax, %1		\n\t"
	 "addl %%eax, %2       		\n\t"
	 "subl $4, %0			\n\t"
	 "jnz 1b			\n\t"
	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"(line_size)
	 : "%eax", "memory"
	);
}

static void put_pixels16_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
{
    __asm __volatile(
	 "lea (%3, %3), %%eax		\n\t"
	 ".balign 8			\n\t"
	 "1:				\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq 8(%1), %%mm4		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq 8(%1, %3), %%mm5		\n\t"
     	 "movq %%mm0, (%2)		\n\t"
     	 "movq %%mm4, 8(%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "movq %%mm5, 8(%2, %3)		\n\t"
	 "addl %%eax, %1		\n\t"
         "addl %%eax, %2       		\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq 8(%1), %%mm4		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq 8(%1, %3), %%mm5		\n\t"
	 "movq %%mm0, (%2)		\n\t"
	 "movq %%mm4, 8(%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "movq %%mm5, 8(%2, %3)		\n\t"
	 "addl %%eax, %1		\n\t"
	 "addl %%eax, %2       		\n\t"
	 "subl $4, %0			\n\t"
	 "jnz 1b			\n\t"
	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"(line_size)
	 : "%eax", "memory"
	);
}

static void clear_blocks_mmx(DCTELEM *blocks)
{
    __asm __volatile(
                "pxor %%mm7, %%mm7		\n\t"
                "movl $-128*6, %%eax		\n\t"
                "1:				\n\t"
                "movq %%mm7, (%0, %%eax)	\n\t"
                "movq %%mm7, 8(%0, %%eax)	\n\t"
                "movq %%mm7, 16(%0, %%eax)	\n\t"
                "movq %%mm7, 24(%0, %%eax)	\n\t"
                "addl $32, %%eax		\n\t"
                " js 1b				\n\t"
                : : "r" (((int)blocks)+128*6)
                : "%eax"
        );
}

static int pix_sum16_mmx(UINT8 * pix, int line_size){
    const int h=16;
    int sum;
    int index= -line_size*h;

    __asm __volatile(
                "pxor %%mm7, %%mm7		\n\t"
                "pxor %%mm6, %%mm6		\n\t"
                "1:				\n\t"
                "movq (%2, %1), %%mm0		\n\t"
                "movq (%2, %1), %%mm1		\n\t"
                "movq 8(%2, %1), %%mm2		\n\t"
                "movq 8(%2, %1), %%mm3		\n\t"
                "punpcklbw %%mm7, %%mm0		\n\t"
                "punpckhbw %%mm7, %%mm1		\n\t"
                "punpcklbw %%mm7, %%mm2		\n\t"
                "punpckhbw %%mm7, %%mm3		\n\t"
                "paddw %%mm0, %%mm1		\n\t"
                "paddw %%mm2, %%mm3		\n\t"
                "paddw %%mm1, %%mm3		\n\t"
                "paddw %%mm3, %%mm6		\n\t"
                "addl %3, %1			\n\t"
                " js 1b				\n\t"
                "movq %%mm6, %%mm5		\n\t"
                "psrlq $32, %%mm6		\n\t"
                "paddw %%mm5, %%mm6		\n\t"
                "movq %%mm6, %%mm5		\n\t"
                "psrlq $16, %%mm6		\n\t"
                "paddw %%mm5, %%mm6		\n\t"
                "movd %%mm6, %0			\n\t"
                "andl $0xFFFF, %0		\n\t"
                : "=&r" (sum), "+r" (index)
                : "r" (pix - index), "r" (line_size)
        );

        return sum;
}

static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
    int i=0;
    asm volatile(
        "1:				\n\t"
        "movq  (%1, %0), %%mm0		\n\t"
        "movq  (%2, %0), %%mm1		\n\t"
        "paddb %%mm0, %%mm1		\n\t"
        "movq %%mm1, (%2, %0)		\n\t"
        "movq 8(%1, %0), %%mm0		\n\t"
        "movq 8(%2, %0), %%mm1		\n\t"
        "paddb %%mm0, %%mm1		\n\t"
        "movq %%mm1, 8(%2, %0)		\n\t"
        "addl $16, %0			\n\t"
        "cmpl %3, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src), "r"(dst), "r"(w-15)
    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
    int i=0;
    asm volatile(
        "1:				\n\t"
        "movq  (%2, %0), %%mm0		\n\t"
        "movq  (%1, %0), %%mm1		\n\t"
        "psubb %%mm0, %%mm1		\n\t"
        "movq %%mm1, (%3, %0)		\n\t"
        "movq 8(%2, %0), %%mm0		\n\t"
        "movq 8(%1, %0), %%mm1		\n\t"
        "psubb %%mm0, %%mm1		\n\t"
        "movq %%mm1, 8(%3, %0)		\n\t"
        "addl $16, %0			\n\t"
        "cmpl %4, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src1), "r"(src2), "r"(dst), "r"(w-15)
    );
    for(; i<w; i++)
        dst[i+0] = src1[i+0]-src2[i+0];
}


#if 0
static void just_return() { return; }
#endif

void dsputil_init_mmx(DSPContext* c, unsigned mask)
{
    mm_flags = mm_support();
#if 0
    fprintf(stderr, "libavcodec: CPU flags:");
    if (mm_flags & MM_MMX)
        fprintf(stderr, " mmx");
    if (mm_flags & MM_MMXEXT)
        fprintf(stderr, " mmxext");
    if (mm_flags & MM_3DNOW)
        fprintf(stderr, " 3dnow");
    if (mm_flags & MM_SSE)
        fprintf(stderr, " sse");
    if (mm_flags & MM_SSE2)
        fprintf(stderr, " sse2");
    fprintf(stderr, "\n");
#endif

    if (mm_flags & MM_MMX) {
        c->get_pixels = get_pixels_mmx;
        c->diff_pixels = diff_pixels_mmx;
        c->put_pixels_clamped = put_pixels_clamped_mmx;
        c->add_pixels_clamped = add_pixels_clamped_mmx;
        c->clear_blocks = clear_blocks_mmx;
        c->pix_sum = pix_sum16_mmx;

        c->pix_abs16x16     = pix_abs16x16_mmx;
        c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;
        c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;
        c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
        c->pix_abs8x8     = pix_abs8x8_mmx;
        c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx;
        c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx;
        c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx;

        c->put_pixels_tab[0][0] = put_pixels16_mmx;
        c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;
        c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;
        c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

        c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;
        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
        c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

        c->avg_pixels_tab[0][0] = avg_pixels16_mmx;
        c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;
        c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;
        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

        c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;
        c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;
        c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;
        c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

        c->put_pixels_tab[1][0] = put_pixels8_mmx;
        c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;
        c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;
        c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

        c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
        c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

        c->avg_pixels_tab[1][0] = avg_pixels8_mmx;
        c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;
        c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
        c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

        c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
        c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
        c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
        c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
        
        c->add_bytes= add_bytes_mmx;
        c->diff_bytes= diff_bytes_mmx;

        if (mm_flags & MM_MMXEXT) {
            c->pix_abs16x16     = pix_abs16x16_mmx2;
            c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;
            c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx2;
            c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2;

            c->pix_abs8x8     = pix_abs8x8_mmx2;
            c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx2;
            c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;
            c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;

            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
        } else if (mm_flags & MM_3DNOW) {
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
        }
    }

#if 0
    // for speed testing
    get_pixels = just_return;
    put_pixels_clamped = just_return;
    add_pixels_clamped = just_return;

    pix_abs16x16 = just_return;
    pix_abs16x16_x2 = just_return;
    pix_abs16x16_y2 = just_return;
    pix_abs16x16_xy2 = just_return;

    put_pixels_tab[0] = just_return;
    put_pixels_tab[1] = just_return;
    put_pixels_tab[2] = just_return;
    put_pixels_tab[3] = just_return;

    put_no_rnd_pixels_tab[0] = just_return;
    put_no_rnd_pixels_tab[1] = just_return;
    put_no_rnd_pixels_tab[2] = just_return;
    put_no_rnd_pixels_tab[3] = just_return;

    avg_pixels_tab[0] = just_return;
    avg_pixels_tab[1] = just_return;
    avg_pixels_tab[2] = just_return;
    avg_pixels_tab[3] = just_return;

    avg_no_rnd_pixels_tab[0] = just_return;
    avg_no_rnd_pixels_tab[1] = just_return;
    avg_no_rnd_pixels_tab[2] = just_return;
    avg_no_rnd_pixels_tab[3] = just_return;

    //av_fdct = just_return;
    //ff_idct = just_return;
#endif
}

/* remove any non bit exact operation (testing purpose). NOTE that
   this function should be kept as small as possible because it is
   always difficult to test automatically non bit exact cases. */
void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask)
{
    if (mm_flags & MM_MMX) {
        /* MMX2 & 3DNOW */
        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
        c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

        if (mm_flags & MM_MMXEXT) {
            c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;
            c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;
            c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
            c->pix_abs8x8_x2 = pix_abs8x8_x2_mmx;
            c->pix_abs8x8_y2 = pix_abs8x8_y2_mmx;
            c->pix_abs8x8_xy2= pix_abs8x8_xy2_mmx;
        }
    }
}