view ppc/snow_altivec.c @ 3995:b00c06477dff libavcodec

write cabac low and range variables as early as possible to prevent stalls from reading them before they where written, the P4 is said to disslike that alot, on P3 its 2% faster (START/STOP_TIMER over decode_residual)

/*
 * Altivec optimized snow DSP utils
 * Copyright (c) 2006 Luca Barbato <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 *
 *
 */

#include "../dsputil.h"

#include "gcc_fixes.h"
#include "dsputil_altivec.h"
#include "../snow.h"

#undef NDEBUG
#include <assert.h>



//FIXME remove this replication
#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))

static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
{
    int offset;
    DWTELEM * buffer;

//  av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);

    assert(buf->data_stack_top >= 0);
//  assert(!buf->line[line]);
    if (buf->line[line])
        return buf->line[line];

    offset = buf->line_width * line;
    buffer = buf->data_stack[buf->data_stack_top];
    buf->data_stack_top--;
    buf->line[line] = buffer;

//  av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1);

    return buffer;
}


//altivec code

void ff_snow_horizontal_compose97i_altivec(DWTELEM *b, int width)
{
    const int w2= (width+1)>>1;
    DECLARE_ALIGNED_16(DWTELEM, temp[(width>>1)]);
    const int w_l= (width>>1);
    const int w_r= w2 - 1;
    int i;
    vector signed int t1, t2, x, y, tmp1, tmp2;
    vector signed int *vbuf, *vtmp;
    vector unsigned char align;



    { // Lift 0
        DWTELEM * const ref = b + w2 - 1;
        DWTELEM b_0 = b[0];
        vbuf = (vector signed int *)b;

        tmp1 = vec_ld (0, ref);
        align = vec_lvsl (0, ref);
        tmp2 = vec_ld (15, ref);
        t1= vec_perm(tmp1, tmp2, align);

        i = 0;

        for (i=0; i<w_l-15; i+=16) {
#if 0
        b[i+0] = b[i+0] - ((3 * (ref[i+0] + ref[i+1]) + 4) >> 3);
        b[i+1] = b[i+1] - ((3 * (ref[i+1] + ref[i+2]) + 4) >> 3);
        b[i+2] = b[i+2] - ((3 * (ref[i+2] + ref[i+3]) + 4) >> 3);
        b[i+3] = b[i+3] - ((3 * (ref[i+3] + ref[i+4]) + 4) >> 3);
#else

        tmp1 = vec_ld (0, ref+4+i);
        tmp2 = vec_ld (15, ref+4+i);

        t2 = vec_perm(tmp1, tmp2, align);

        y = vec_add(t1,vec_sld(t1,t2,4));
        y = vec_add(vec_add(y,y),y);

        tmp1 = vec_ld (0, ref+8+i);

        y = vec_add(y, vec_splat_s32(4));
        y = vec_sra(y, vec_splat_u32(3));

        tmp2 = vec_ld (15, ref+8+i);

        *vbuf = vec_sub(*vbuf, y);

        t1=t2;

        vbuf++;

        t2 = vec_perm(tmp1, tmp2, align);

        y = vec_add(t1,vec_sld(t1,t2,4));
        y = vec_add(vec_add(y,y),y);

        tmp1 = vec_ld (0, ref+12+i);

        y = vec_add(y, vec_splat_s32(4));
        y = vec_sra(y, vec_splat_u32(3));

        tmp2 = vec_ld (15, ref+12+i);

        *vbuf = vec_sub(*vbuf, y);

        t1=t2;

        vbuf++;

        t2 = vec_perm(tmp1, tmp2, align);

        y = vec_add(t1,vec_sld(t1,t2,4));
        y = vec_add(vec_add(y,y),y);

        tmp1 = vec_ld (0, ref+16+i);

        y = vec_add(y, vec_splat_s32(4));
        y = vec_sra(y, vec_splat_u32(3));

        tmp2 = vec_ld (15, ref+16+i);

        *vbuf = vec_sub(*vbuf, y);

        t1=t2;

        t2 = vec_perm(tmp1, tmp2, align);

        y = vec_add(t1,vec_sld(t1,t2,4));
        y = vec_add(vec_add(y,y),y);

        vbuf++;

        y = vec_add(y, vec_splat_s32(4));
        y = vec_sra(y, vec_splat_u32(3));
        *vbuf = vec_sub(*vbuf, y);

        t1=t2;

        vbuf++;

#endif
        }

        snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
        b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
    }

    { // Lift 1
        DWTELEM * const dst = b+w2;

        i = 0;
        for(; (((long)&dst[i]) & 0xF) && i<w_r; i++){
            dst[i] = dst[i] - (b[i] + b[i + 1]);
        }

        align = vec_lvsl(0, b+i);
        tmp1 = vec_ld(0, b+i);
        vbuf = (vector signed int*) (dst + i);
        tmp2 = vec_ld(15, b+i);

        t1 = vec_perm(tmp1, tmp2, align);

        for (; i<w_r-3; i+=4) {

#if 0
            dst[i]   = dst[i]   - (b[i]   + b[i + 1]);
            dst[i+1] = dst[i+1] - (b[i+1] + b[i + 2]);
            dst[i+2] = dst[i+2] - (b[i+2] + b[i + 3]);
            dst[i+3] = dst[i+3] - (b[i+3] + b[i + 4]);
#else

        tmp1 = vec_ld(0, b+4+i);
        tmp2 = vec_ld(15, b+4+i);

        t2 = vec_perm(tmp1, tmp2, align);

        y = vec_add(t1, vec_sld(t1,t2,4));
        *vbuf = vec_sub (*vbuf, y);

        vbuf++;

        t1 = t2;

#endif

        }

        snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
    }

    { // Lift 2
        DWTELEM * const ref = b+w2 - 1;
        DWTELEM b_0 = b[0];
        vbuf= (vector signed int *) b;

        tmp1 = vec_ld (0, ref);
        align = vec_lvsl (0, ref);
        tmp2 = vec_ld (15, ref);
        t1= vec_perm(tmp1, tmp2, align);

        i = 0;
        for (; i<w_l-15; i+=16) {
#if 0
            b[i]   = b[i]   - (((8 -(ref[i]   + ref[i+1])) - (b[i]  <<2)) >> 4);
            b[i+1] = b[i+1] - (((8 -(ref[i+1] + ref[i+2])) - (b[i+1]<<2)) >> 4);
            b[i+2] = b[i+2] - (((8 -(ref[i+2] + ref[i+3])) - (b[i+2]<<2)) >> 4);
            b[i+3] = b[i+3] - (((8 -(ref[i+3] + ref[i+4])) - (b[i+3]<<2)) >> 4);
#else
            tmp1 = vec_ld (0, ref+4+i);
            tmp2 = vec_ld (15, ref+4+i);

            t2 = vec_perm(tmp1, tmp2, align);

            y = vec_add(t1,vec_sld(t1,t2,4));
            y = vec_sub(vec_splat_s32(8),y);

            tmp1 = vec_ld (0, ref+8+i);

            x = vec_sl(*vbuf,vec_splat_u32(2));
            y = vec_sra(vec_sub(y,x),vec_splat_u32(4));

            tmp2 = vec_ld (15, ref+8+i);

            *vbuf = vec_sub( *vbuf, y);

            t1 = t2;

            vbuf++;

            t2 = vec_perm(tmp1, tmp2, align);

            y = vec_add(t1,vec_sld(t1,t2,4));
            y = vec_sub(vec_splat_s32(8),y);

            tmp1 = vec_ld (0, ref+12+i);

            x = vec_sl(*vbuf,vec_splat_u32(2));
            y = vec_sra(vec_sub(y,x),vec_splat_u32(4));

            tmp2 = vec_ld (15, ref+12+i);

            *vbuf = vec_sub( *vbuf, y);

            t1 = t2;

            vbuf++;

            t2 = vec_perm(tmp1, tmp2, align);

            y = vec_add(t1,vec_sld(t1,t2,4));
            y = vec_sub(vec_splat_s32(8),y);

            tmp1 = vec_ld (0, ref+16+i);

            x = vec_sl(*vbuf,vec_splat_u32(2));
            y = vec_sra(vec_sub(y,x),vec_splat_u32(4));

            tmp2 = vec_ld (15, ref+16+i);

            *vbuf = vec_sub( *vbuf, y);

            t1 = t2;

            vbuf++;

            t2 = vec_perm(tmp1, tmp2, align);

            y = vec_add(t1,vec_sld(t1,t2,4));
            y = vec_sub(vec_splat_s32(8),y);

            t1 = t2;

            x = vec_sl(*vbuf,vec_splat_u32(2));
            y = vec_sra(vec_sub(y,x),vec_splat_u32(4));
            *vbuf = vec_sub( *vbuf, y);

            vbuf++;

#endif
        }

        snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
        b[0] = b_0 - (((-2 * ref[1] + W_BO) - 4 * b_0) >> W_BS);
    }

    { // Lift 3
        DWTELEM * const src = b+w2;

        vbuf = (vector signed int *)b;
        vtmp = (vector signed int *)temp;

        i = 0;
        align = vec_lvsl(0, src);

        for (; i<w_r-3; i+=4) {
#if 0
            temp[i] = src[i] - ((-3*(b[i] + b[i+1]))>>1);
            temp[i+1] = src[i+1] - ((-3*(b[i+1] + b[i+2]))>>1);
            temp[i+2] = src[i+2] - ((-3*(b[i+2] + b[i+3]))>>1);
            temp[i+3] = src[i+3] - ((-3*(b[i+3] + b[i+4]))>>1);
#else
            tmp1 = vec_ld(0,src+i);
            t1 = vec_add(vbuf[0],vec_sld(vbuf[0],vbuf[1],4));
            tmp2 = vec_ld(15,src+i);
            t1 = vec_sub(vec_splat_s32(0),t1); //bad!
            t1 = vec_add(t1,vec_add(t1,t1));
            t2 = vec_perm(tmp1 ,tmp2 ,align);
            t1 = vec_sra(t1,vec_splat_u32(1));
            vbuf++;
            *vtmp = vec_sub(t2,t1);
            vtmp++;

#endif

        }

        snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -3, 0, 1);
    }

    {
    //Interleave
        int a;
        vector signed int *t = (vector signed int *)temp,
                          *v = (vector signed int *)b;

        snow_interleave_line_header(&i, width, b, temp);

        for (; (i & 0xE) != 0xE; i-=2){
            b[i+1] = temp[i>>1];
            b[i] = b[i>>1];
        }
        for (i-=14; i>=0; i-=16){
           a=i/4;

           v[a+3]=vec_mergel(v[(a>>1)+1],t[(a>>1)+1]);
           v[a+2]=vec_mergeh(v[(a>>1)+1],t[(a>>1)+1]);
           v[a+1]=vec_mergel(v[a>>1],t[a>>1]);
           v[a]=vec_mergeh(v[a>>1],t[a>>1]);

        }

    }
}

void ff_snow_vertical_compose97i_altivec(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width)
{
    int i, w4 = width/4;
    vector signed int *v0, *v1,*v2,*v3,*v4,*v5;
    vector signed int t1, t2;

    v0=(vector signed int *)b0;
    v1=(vector signed int *)b1;
    v2=(vector signed int *)b2;
    v3=(vector signed int *)b3;
    v4=(vector signed int *)b4;
    v5=(vector signed int *)b5;

    for (i=0; i< w4;i++)
    {

    #if 0
        b4[i] -= (3*(b3[i] + b5[i])+4)>>3;
        b3[i] -= ((b2[i] + b4[i]));
        b2[i] += ((b1[i] + b3[i])+4*b2[i]+8)>>4;
        b1[i] += (3*(b0[i] + b2[i]))>>1;
    #else
        t1 = vec_add(v3[i], v5[i]);
        t2 = vec_add(t1, vec_add(t1,t1));
        t1 = vec_add(t2, vec_splat_s32(4));
        v4[i] = vec_sub(v4[i], vec_sra(t1,vec_splat_u32(3)));

        v3[i] = vec_sub(v3[i], vec_add(v2[i], v4[i]));

        t1 = vec_add(vec_splat_s32(8), vec_add(v1[i], v3[i]));
        t2 = vec_sl(v2[i], vec_splat_u32(2));
        v2[i] = vec_add(v2[i], vec_sra(vec_add(t1,t2),vec_splat_u32(4)));
        t1 = vec_add(v0[i], v2[i]);
        t2 = vec_add(t1, vec_add(t1,t1));
        v1[i] = vec_add(v1[i], vec_sra(t2,vec_splat_u32(1)));

    #endif
    }

    for(i*=4; i < width; i++)
    {
        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
        b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
        b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
}

#define LOAD_BLOCKS \
            tmp1 = vec_ld(0, &block[3][y*src_stride]);\
            align = vec_lvsl(0, &block[3][y*src_stride]);\
            tmp2 = vec_ld(15, &block[3][y*src_stride]);\
\
            b3 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, &block[2][y*src_stride]);\
            align = vec_lvsl(0, &block[2][y*src_stride]);\
            tmp2 = vec_ld(15, &block[2][y*src_stride]);\
\
            b2 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, &block[1][y*src_stride]);\
            align = vec_lvsl(0, &block[1][y*src_stride]);\
            tmp2 = vec_ld(15, &block[1][y*src_stride]);\
\
            b1 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, &block[0][y*src_stride]);\
            align = vec_lvsl(0, &block[0][y*src_stride]);\
            tmp2 = vec_ld(15, &block[0][y*src_stride]);\
\
            b0 = vec_perm(tmp1,tmp2,align);

#define LOAD_OBMCS \
            tmp1 = vec_ld(0, obmc1);\
            align = vec_lvsl(0, obmc1);\
            tmp2 = vec_ld(15, obmc1);\
\
            ob1 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, obmc2);\
            align = vec_lvsl(0, obmc2);\
            tmp2 = vec_ld(15, obmc2);\
\
            ob2 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, obmc3);\
            align = vec_lvsl(0, obmc3);\
            tmp2 = vec_ld(15, obmc3);\
\
            ob3 = vec_perm(tmp1,tmp2,align);\
\
            tmp1 = vec_ld(0, obmc4);\
            align = vec_lvsl(0, obmc4);\
            tmp2 = vec_ld(15, obmc4);\
\
            ob4 = vec_perm(tmp1,tmp2,align);

/* interleave logic
 * h1 <- [ a,b,a,b, a,b,a,b, a,b,a,b, a,b,a,b ]
 * h2 <- [ c,d,c,d, c,d,c,d, c,d,c,d, c,d,c,d ]
 * h  <- [ a,b,c,d, a,b,c,d, a,b,c,d, a,b,c,d ]
 */

#define STEPS_0_1\
            h1 = (vector unsigned short)\
                 vec_mergeh(ob1, ob2);\
\
            h2 = (vector unsigned short)\
                 vec_mergeh(ob3, ob4);\
\
            ih = (vector unsigned char)\
                 vec_mergeh(h1,h2);\
\
            l1 = (vector unsigned short) vec_mergeh(b3, b2);\
\
            ih1 = (vector unsigned char) vec_mergel(h1, h2);\
\
            l2 = (vector unsigned short) vec_mergeh(b1, b0);\
\
            il = (vector unsigned char) vec_mergeh(l1, l2);\
\
            v[0] = (vector signed int) vec_msum(ih, il, vec_splat_u32(0));\
\
            il1 = (vector unsigned char) vec_mergel(l1, l2);\
\
            v[1] = (vector signed int) vec_msum(ih1, il1, vec_splat_u32(0));

#define FINAL_STEP_SCALAR\
        for(x=0; x<b_w; x++)\
            if(add){\
                vbuf[x] += dst[x + src_x];\
                vbuf[x] = (vbuf[x] + (1<<(FRAC_BITS-1))) >> FRAC_BITS;\
                if(vbuf[x]&(~255)) vbuf[x]= ~(vbuf[x]>>31);\
                dst8[x + y*src_stride] = vbuf[x];\
            }else{\
                dst[x + src_x] -= vbuf[x];\
            }

static void inner_add_yblock_bw_8_obmc_16_altivec(uint8_t *obmc,
                                             const int obmc_stride,
                                             uint8_t * * block, int b_w,
                                             int b_h, int src_x, int src_y,
                                             int src_stride, slice_buffer * sb,
                                             int add, uint8_t * dst8)
{
    int y, x;
    DWTELEM * dst;
    vector unsigned short h1, h2, l1, l2;
    vector unsigned char ih, il, ih1, il1, tmp1, tmp2, align;
    vector unsigned char b0,b1,b2,b3;
    vector unsigned char ob1,ob2,ob3,ob4;

    DECLARE_ALIGNED_16(int, vbuf[16]);
    vector signed int *v = (vector signed int *)vbuf, *d;

    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride

        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);

        dst = slice_buffer_get_line(sb, src_y + y);
        d = (vector signed int *)(dst + src_x);

//FIXME i could avoid some loads!

        // load blocks
        LOAD_BLOCKS

        // load obmcs
        LOAD_OBMCS

        // steps 0 1
        STEPS_0_1

        FINAL_STEP_SCALAR

       }

}

#define STEPS_2_3\
            h1 = (vector unsigned short) vec_mergel(ob1, ob2);\
\
            h2 = (vector unsigned short) vec_mergel(ob3, ob4);\
\
            ih = (vector unsigned char) vec_mergeh(h1,h2);\
\
            l1 = (vector unsigned short) vec_mergel(b3, b2);\
\
            l2 = (vector unsigned short) vec_mergel(b1, b0);\
\
            ih1 = (vector unsigned char) vec_mergel(h1,h2);\
\
            il = (vector unsigned char) vec_mergeh(l1,l2);\
\
            v[2] = (vector signed int) vec_msum(ih, il, vec_splat_u32(0));\
\
            il1 = (vector unsigned char) vec_mergel(l1,l2);\
\
            v[3] = (vector signed int) vec_msum(ih1, il1, vec_splat_u32(0));


static void inner_add_yblock_bw_16_obmc_32_altivec(uint8_t *obmc,
                                             const int obmc_stride,
                                             uint8_t * * block, int b_w,
                                             int b_h, int src_x, int src_y,
                                             int src_stride, slice_buffer * sb,
                                             int add, uint8_t * dst8)
{
    int y, x;
    DWTELEM * dst;
    vector unsigned short h1, h2, l1, l2;
    vector unsigned char ih, il, ih1, il1, tmp1, tmp2, align;
    vector unsigned char b0,b1,b2,b3;
    vector unsigned char ob1,ob2,ob3,ob4;
    DECLARE_ALIGNED_16(int, vbuf[b_w]);
    vector signed int *v = (vector signed int *)vbuf, *d;

    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride

        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);

        dst = slice_buffer_get_line(sb, src_y + y);
        d = (vector signed int *)(dst + src_x);

        // load blocks
        LOAD_BLOCKS

        // load obmcs
        LOAD_OBMCS

        // steps 0 1 2 3
        STEPS_0_1

        STEPS_2_3

        FINAL_STEP_SCALAR

    }
}

#define FINAL_STEP_VEC \
\
    if(add)\
        {\
            for(x=0; x<b_w/4; x++)\
            {\
                v[x] = vec_add(v[x], d[x]);\
                v[x] = vec_sra(vec_add(v[x],\
                                       vec_sl( vec_splat_s32(1),\
                                               vec_splat_u32(7))),\
                               vec_splat_u32(8));\
\
                mask = (vector bool int) vec_sl((vector signed int)\
                        vec_cmpeq(v[x],v[x]),vec_splat_u32(8));\
                mask = (vector bool int) vec_and(v[x],vec_nor(mask,mask));\
\
                mask = (vector bool int)\
                        vec_cmpeq((vector signed int)mask,\
                                  (vector signed int)vec_splat_u32(0));\
\
                vs = vec_sra(v[x],vec_splat_u32(8));\
                vs = vec_sra(v[x],vec_splat_u32(8));\
                vs = vec_sra(v[x],vec_splat_u32(15));\
\
                vs = vec_nor(vs,vs);\
\
                v[x]= vec_sel(v[x],vs,mask);\
            }\
\
            for(x=0; x<b_w; x++)\
                dst8[x + y*src_stride] = vbuf[x];\
\
        }\
         else\
            for(x=0; x<b_w/4; x++)\
                d[x] = vec_sub(d[x], v[x]);

static void inner_add_yblock_a_bw_8_obmc_16_altivec(uint8_t *obmc,
                                             const int obmc_stride,
                                             uint8_t * * block, int b_w,
                                             int b_h, int src_x, int src_y,
                                             int src_stride, slice_buffer * sb,
                                             int add, uint8_t * dst8)
{
    int y, x;
    DWTELEM * dst;
    vector bool int mask;
    vector signed int vs;
    vector unsigned short h1, h2, l1, l2;
    vector unsigned char ih, il, ih1, il1, tmp1, tmp2, align;
    vector unsigned char b0,b1,b2,b3;
    vector unsigned char ob1,ob2,ob3,ob4;

    DECLARE_ALIGNED_16(int, vbuf[16]);
    vector signed int *v = (vector signed int *)vbuf, *d;

    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride

        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);

        dst = slice_buffer_get_line(sb, src_y + y);
        d = (vector signed int *)(dst + src_x);

//FIXME i could avoid some loads!

        // load blocks
        LOAD_BLOCKS

        // load obmcs
        LOAD_OBMCS

        // steps 0 1
        STEPS_0_1

        FINAL_STEP_VEC

       }

}

static void inner_add_yblock_a_bw_16_obmc_32_altivec(uint8_t *obmc,
                                             const int obmc_stride,
                                             uint8_t * * block, int b_w,
                                             int b_h, int src_x, int src_y,
                                             int src_stride, slice_buffer * sb,
                                             int add, uint8_t * dst8)
{
    int y, x;
    DWTELEM * dst;
    vector bool int mask;
    vector signed int vs;
    vector unsigned short h1, h2, l1, l2;
    vector unsigned char ih, il, ih1, il1, tmp1, tmp2, align;
    vector unsigned char b0,b1,b2,b3;
    vector unsigned char ob1,ob2,ob3,ob4;
    DECLARE_ALIGNED_16(int, vbuf[b_w]);
    vector signed int *v = (vector signed int *)vbuf, *d;

    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride

        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);

        dst = slice_buffer_get_line(sb, src_y + y);
        d = (vector signed int *)(dst + src_x);

        // load blocks
        LOAD_BLOCKS

        // load obmcs
        LOAD_OBMCS

        // steps 0 1 2 3
        STEPS_0_1

        STEPS_2_3

        FINAL_STEP_VEC

    }
}


void ff_snow_inner_add_yblock_altivec(uint8_t *obmc, const int obmc_stride,
                                      uint8_t * * block, int b_w, int b_h,
                                      int src_x, int src_y, int src_stride,
                                      slice_buffer * sb, int add,
                                      uint8_t * dst8)
{
    if (src_x&15) {
        if (b_w == 16)
            inner_add_yblock_bw_16_obmc_32_altivec(obmc, obmc_stride, block,
                                                   b_w, b_h, src_x, src_y,
                                                   src_stride, sb, add, dst8);
        else if (b_w == 8)
            inner_add_yblock_bw_8_obmc_16_altivec(obmc, obmc_stride, block,
                                                  b_w, b_h, src_x, src_y,
                                                  src_stride, sb, add, dst8);
        else
            ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,
                                     src_y, src_stride, sb, add, dst8);
    } else {
        if (b_w == 16)
            inner_add_yblock_a_bw_16_obmc_32_altivec(obmc, obmc_stride, block,
                                                     b_w, b_h, src_x, src_y,
                                                     src_stride, sb, add, dst8);
        else if (b_w == 8)
            inner_add_yblock_a_bw_8_obmc_16_altivec(obmc, obmc_stride, block,
                                                    b_w, b_h, src_x, src_y,
                                                    src_stride, sb, add, dst8);
        else
            ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,
                                     src_y, src_stride, sb, add, dst8);
    }
}


void snow_init_altivec(DSPContext* c, AVCodecContext *avctx)
{
        c->horizontal_compose97i = ff_snow_horizontal_compose97i_altivec;
        c->vertical_compose97i = ff_snow_vertical_compose97i_altivec;
        c->inner_add_yblock = ff_snow_inner_add_yblock_altivec;
}