comparison snow.c @ 5652:941e5deeb2a4 libavcodec

rename HTAPS -> HTAPS_MAX (later is correct after the previous change)

5651:ab023c9f03d0

};

#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)
#define ENCODER_EXTRA_BITS 4
#define HTAPS 8

typedef struct x_and_coeff{
    int16_t x;
    uint16_t coeff;
} x_and_coeff;

    int width;
    int height;
    SubBand band[MAX_DECOMPOSITIONS][4];

    int htaps;
    int8_t hcoeff[HTAPS/2];
    int diag_mc;
    int fast_mc;

    int last_htaps;
    int8_t last_hcoeff[HTAPS/2];
    int last_diag_mc;
}Plane;

typedef struct SnowContext{
//    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX)

    0,1,0,0,
    15
    };

    int x, y, b, r, l;
    int16_t tmpIt   [64*(32+HTAPS)];
    uint8_t tmp2t[3][stride*(32+HTAPS)];
    int16_t *tmpI= tmpIt;
    uint8_t *tmp2= tmp2t[0];
    uint8_t *hpel[11];
START_TIMER
    assert(dx<16 && dy<16);

    b= needs[l] | needs[r];
    if(p && !p->diag_mc)
        b= 15;

    if(b&5){
        for(y=0; y < b_h+HTAPS-1; y++){
            for(x=0; x < b_w; x++){
                int a_1=src[x + HTAPS/2-4];
                int a0= src[x + HTAPS/2-3];
                int a1= src[x + HTAPS/2-2];
                int a2= src[x + HTAPS/2-1];
                int a3= src[x + HTAPS/2+0];
                int a4= src[x + HTAPS/2+1];
                int a5= src[x + HTAPS/2+2];
                int a6= src[x + HTAPS/2+3];
                int am=0;
                if(!p || p->fast_mc){
                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
                    tmpI[x]= am;
                    am= (am+16)>>5;

            tmp2+= stride;
            src += stride;
        }
        src -= stride*y;
    }
    src += HTAPS/2 - 1;
    tmp2= tmp2t[1];

    if(b&2){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w+1; x++){
                int a_1=src[x + (HTAPS/2-4)*stride];
                int a0= src[x + (HTAPS/2-3)*stride];
                int a1= src[x + (HTAPS/2-2)*stride];
                int a2= src[x + (HTAPS/2-1)*stride];
                int a3= src[x + (HTAPS/2+0)*stride];
                int a4= src[x + (HTAPS/2+1)*stride];
                int a5= src[x + (HTAPS/2+2)*stride];
                int a6= src[x + (HTAPS/2+3)*stride];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;

            src += stride;
            tmp2+= stride;
        }
        src -= stride*y;
    }
    src += stride*(HTAPS/2 - 1);
    tmp2= tmp2t[2];
    tmpI= tmpIt;
    if(b&4){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                int a_1=tmpI[x + (HTAPS/2-4)*64];
                int a0= tmpI[x + (HTAPS/2-3)*64];
                int a1= tmpI[x + (HTAPS/2-2)*64];
                int a2= tmpI[x + (HTAPS/2-1)*64];
                int a3= tmpI[x + (HTAPS/2+0)*64];
                int a4= tmpI[x + (HTAPS/2+1)*64];
                int a5= tmpI[x + (HTAPS/2+2)*64];
                int a6= tmpI[x + (HTAPS/2+3)*64];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;

            tmp2+= stride;
        }
    }

    hpel[ 0]= src;
    hpel[ 1]= tmp2t[0] + stride*(HTAPS/2-1);
    hpel[ 2]= src + 1;

    hpel[ 4]= tmp2t[1];
    hpel[ 5]= tmp2t[2];
    hpel[ 6]= tmp2t[1] + 1;

STOP_TIMER("mc_block")
}

#define mca(dx,dy,b_w)\
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
    uint8_t tmp[stride*(b_w+HTAPS-1)];\
    assert(h==b_w);\
    mc_block(NULL, dst, src-(HTAPS/2-1)-(HTAPS/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
}

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)

        int mx= block->mx*scale;
        int my= block->my*scale;
        const int dx= mx&15;
        const int dy= my&15;
        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
        sx += (mx>>4) - (HTAPS/2-1);
        sy += (my>>4) - (HTAPS/2-1);
        src += sx + sy*stride;
        if(   (unsigned)sx >= w - b_w - (HTAPS-2)
           || (unsigned)sy >= h - b_h - (HTAPS-2)){
            ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS-1, b_h+HTAPS-1, sx, sy, w, h);
            src= tmp + MB_SIZE;
        }
//        assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
//        assert(!(b_w&(b_w-1)));
        assert(b_w>1 && b_h>1);

    const int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst= s->current_picture.data[plane_index];
    uint8_t *src= s->  input_picture.data[plane_index];
    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
    uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment
    uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS-1)];
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    const int w= p->width;
    const int h= p->height;
    int distortion;

            for(plane_index=0; plane_index<2; plane_index++){
                int htaps, i, sum=0, absum=0;
                Plane *p= &s->plane[plane_index];
                p->diag_mc= get_rac(&s->c, s->header_state);
                htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
                if((unsigned)htaps > HTAPS || htaps==0)
                    return -1;
                p->htaps= htaps;
                for(i= htaps/2; i; i--){
                    p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
                    sum += p->hcoeff[i];

5652:941e5deeb2a4

};

#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)
#define ENCODER_EXTRA_BITS 4
#define HTAPS_MAX 8

typedef struct x_and_coeff{
    int16_t x;
    uint16_t coeff;
} x_and_coeff;

    int width;
    int height;
    SubBand band[MAX_DECOMPOSITIONS][4];

    int htaps;
    int8_t hcoeff[HTAPS_MAX/2];
    int diag_mc;
    int fast_mc;

    int last_htaps;
    int8_t last_hcoeff[HTAPS_MAX/2];
    int last_diag_mc;
}Plane;

typedef struct SnowContext{
//    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX)

    0,1,0,0,
    15
    };

    int x, y, b, r, l;
    int16_t tmpIt   [64*(32+HTAPS_MAX)];
    uint8_t tmp2t[3][stride*(32+HTAPS_MAX)];
    int16_t *tmpI= tmpIt;
    uint8_t *tmp2= tmp2t[0];
    uint8_t *hpel[11];
START_TIMER
    assert(dx<16 && dy<16);

    b= needs[l] | needs[r];
    if(p && !p->diag_mc)
        b= 15;

    if(b&5){
        for(y=0; y < b_h+HTAPS_MAX-1; y++){
            for(x=0; x < b_w; x++){
                int a_1=src[x + HTAPS_MAX/2-4];
                int a0= src[x + HTAPS_MAX/2-3];
                int a1= src[x + HTAPS_MAX/2-2];
                int a2= src[x + HTAPS_MAX/2-1];
                int a3= src[x + HTAPS_MAX/2+0];
                int a4= src[x + HTAPS_MAX/2+1];
                int a5= src[x + HTAPS_MAX/2+2];
                int a6= src[x + HTAPS_MAX/2+3];
                int am=0;
                if(!p || p->fast_mc){
                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
                    tmpI[x]= am;
                    am= (am+16)>>5;

            tmp2+= stride;
            src += stride;
        }
        src -= stride*y;
    }
    src += HTAPS_MAX/2 - 1;
    tmp2= tmp2t[1];

    if(b&2){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w+1; x++){
                int a_1=src[x + (HTAPS_MAX/2-4)*stride];
                int a0= src[x + (HTAPS_MAX/2-3)*stride];
                int a1= src[x + (HTAPS_MAX/2-2)*stride];
                int a2= src[x + (HTAPS_MAX/2-1)*stride];
                int a3= src[x + (HTAPS_MAX/2+0)*stride];
                int a4= src[x + (HTAPS_MAX/2+1)*stride];
                int a5= src[x + (HTAPS_MAX/2+2)*stride];
                int a6= src[x + (HTAPS_MAX/2+3)*stride];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;

            src += stride;
            tmp2+= stride;
        }
        src -= stride*y;
    }
    src += stride*(HTAPS_MAX/2 - 1);
    tmp2= tmp2t[2];
    tmpI= tmpIt;
    if(b&4){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
                int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
                int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
                int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
                int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
                int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
                int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
                int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;

            tmp2+= stride;
        }
    }

    hpel[ 0]= src;
    hpel[ 1]= tmp2t[0] + stride*(HTAPS_MAX/2-1);
    hpel[ 2]= src + 1;

    hpel[ 4]= tmp2t[1];
    hpel[ 5]= tmp2t[2];
    hpel[ 6]= tmp2t[1] + 1;

STOP_TIMER("mc_block")
}

#define mca(dx,dy,b_w)\
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
    uint8_t tmp[stride*(b_w+HTAPS_MAX-1)];\
    assert(h==b_w);\
    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
}

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)

        int mx= block->mx*scale;
        int my= block->my*scale;
        const int dx= mx&15;
        const int dy= my&15;
        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
        sx += (mx>>4) - (HTAPS_MAX/2-1);
        sy += (my>>4) - (HTAPS_MAX/2-1);
        src += sx + sy*stride;
        if(   (unsigned)sx >= w - b_w - (HTAPS_MAX-2)
           || (unsigned)sy >= h - b_h - (HTAPS_MAX-2)){
            ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h);
            src= tmp + MB_SIZE;
        }
//        assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
//        assert(!(b_w&(b_w-1)));
        assert(b_w>1 && b_h>1);

    const int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst= s->current_picture.data[plane_index];
    uint8_t *src= s->  input_picture.data[plane_index];
    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
    uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment
    uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS_MAX-1)];
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    const int w= p->width;
    const int h= p->height;
    int distortion;

            for(plane_index=0; plane_index<2; plane_index++){
                int htaps, i, sum=0, absum=0;
                Plane *p= &s->plane[plane_index];
                p->diag_mc= get_rac(&s->c, s->header_state);
                htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
                if((unsigned)htaps > HTAPS_MAX || htaps==0)
                    return -1;
                p->htaps= htaps;
                for(i= htaps/2; i; i--){
                    p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
                    sum += p->hcoeff[i];