comparison i386/dsputil_mmx.c @ 1530:3b31998fe22f libavcodec

disable encoders where appropriate (patch courtesy of BERO <bero -at- geocities.co.jp>)

1529:cb523a2ca00f

#undef PAVGB

/***********************************/
/* standard MMX */

static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
{
    asm volatile(
        "movl $-128, %%eax	\n\t"
        "pxor %%mm7, %%mm7	\n\t"

        : "+r" (s1), "+r" (s2)
        : "r" (block+64), "r" (stride)
        : "%eax"
    );
}

void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
{
    const DCTELEM *p;
    uint8_t *pix;

                : : "r" (((int)blocks)+128*6)
                : "%eax"
        );
}

static int pix_sum16_mmx(uint8_t * pix, int line_size){
    const int h=16;
    int sum;
    int index= -line_size*h;

                : "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"

    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

static int pix_norm1_mmx(uint8_t *pix, int line_size) {
    int tmp;
  asm volatile (
      "movl $16,%%ecx\n"
      "pxor %%mm0,%%mm0\n"

}


WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
WARPER88_1616(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

                c->idct    = ff_mmx_idct;
            }
            c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
        }
        
        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->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->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;
        
        c->hadamard8_diff[0]= hadamard8_diff16_mmx;
        c->hadamard8_diff[1]= hadamard8_diff_mmx;
        
	c->pix_norm1 = pix_norm1_mmx;
	c->sse[0] = sse16_mmx;
        
        if (mm_flags & MM_MMXEXT) {
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
            c->put_pixels_tab[0][2] = put_pixels16_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->hadamard8_diff[0]= hadamard8_diff16_mmx2;
            c->hadamard8_diff[1]= hadamard8_diff_mmx2;

            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
                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->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
        }
    }
        
    dsputil_init_pix_mmx(c, avctx);
#if 0
    // for speed testing
    get_pixels = just_return;
    put_pixels_clamped = just_return;
    add_pixels_clamped = just_return;

1530:3b31998fe22f

#undef PAVGB

/***********************************/
/* standard MMX */

#ifdef CONFIG_ENCODERS
static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
{
    asm volatile(
        "movl $-128, %%eax	\n\t"
        "pxor %%mm7, %%mm7	\n\t"

        : "+r" (s1), "+r" (s2)
        : "r" (block+64), "r" (stride)
        : "%eax"
    );
}
#endif //CONFIG_ENCODERS

void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
{
    const DCTELEM *p;
    uint8_t *pix;

                : : "r" (((int)blocks)+128*6)
                : "%eax"
        );
}

#ifdef CONFIG_ENCODERS
static int pix_sum16_mmx(uint8_t * pix, int line_size){
    const int h=16;
    int sum;
    int index= -line_size*h;

                : "r" (pix - index), "r" (line_size)
        );

        return sum;
}
#endif //CONFIG_ENCODERS

static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
    int i=0;
    asm volatile(
        "1:				\n\t"

    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

#ifdef CONFIG_ENCODERS
static int pix_norm1_mmx(uint8_t *pix, int line_size) {
    int tmp;
  asm volatile (
      "movl $16,%%ecx\n"
      "pxor %%mm0,%%mm0\n"

}


WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
WARPER88_1616(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)
#endif //CONFIG_ENCODERS

#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

                c->idct    = ff_mmx_idct;
            }
            c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
        }
        
#ifdef CONFIG_ENCODERS
        c->get_pixels = get_pixels_mmx;
        c->diff_pixels = diff_pixels_mmx;
#endif //CONFIG_ENCODERS
        c->put_pixels_clamped = put_pixels_clamped_mmx;
        c->add_pixels_clamped = add_pixels_clamped_mmx;
        c->clear_blocks = clear_blocks_mmx;
#ifdef CONFIG_ENCODERS
        c->pix_sum = pix_sum16_mmx;
#endif //CONFIG_ENCODERS

        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->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;
#ifdef CONFIG_ENCODERS
        c->diff_bytes= diff_bytes_mmx;
        
        c->hadamard8_diff[0]= hadamard8_diff16_mmx;
        c->hadamard8_diff[1]= hadamard8_diff_mmx;
        
	c->pix_norm1 = pix_norm1_mmx;
	c->sse[0] = sse16_mmx;
#endif //CONFIG_ENCODERS
        
        if (mm_flags & MM_MMXEXT) {
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
            c->put_pixels_tab[0][2] = put_pixels16_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;

#ifdef CONFIG_ENCODERS
            c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
            c->hadamard8_diff[1]= hadamard8_diff_mmx2;
#endif //CONFIG_ENCODERS

            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
                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->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
        }
    }
        
#ifdef CONFIG_ENCODERS
    dsputil_init_pix_mmx(c, avctx);
#endif //CONFIG_ENCODERS
#if 0
    // for speed testing
    get_pixels = just_return;
    put_pixels_clamped = just_return;
    add_pixels_clamped = just_return;