yading@10: /*
yading@10:  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/intmath.h"
yading@10: #include "libavutil/log.h"
yading@10: #include "libavutil/opt.h"
yading@10: #include "avcodec.h"
yading@10: #include "dsputil.h"
yading@10: #include "snow_dwt.h"
yading@10: #include "internal.h"
yading@10: #include "snow.h"
yading@10: #include "snowdata.h"
yading@10: 
yading@10: #include "rangecoder.h"
yading@10: #include "mathops.h"
yading@10: #include "h263.h"
yading@10: 
yading@10: 
yading@10: void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
yading@10:                               int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
yading@10:     int y, x;
yading@10:     IDWTELEM * dst;
yading@10:     for(y=0; y<b_h; y++){
yading@10:         //FIXME ugly misuse of obmc_stride
yading@10:         const uint8_t *obmc1= obmc + y*obmc_stride;
yading@10:         const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
yading@10:         const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
yading@10:         const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
yading@10:         dst = slice_buffer_get_line(sb, src_y + y);
yading@10:         for(x=0; x<b_w; x++){
yading@10:             int v=   obmc1[x] * block[3][x + y*src_stride]
yading@10:                     +obmc2[x] * block[2][x + y*src_stride]
yading@10:                     +obmc3[x] * block[1][x + y*src_stride]
yading@10:                     +obmc4[x] * block[0][x + y*src_stride];
yading@10: 
yading@10:             v <<= 8 - LOG2_OBMC_MAX;
yading@10:             if(FRAC_BITS != 8){
yading@10:                 v >>= 8 - FRAC_BITS;
yading@10:             }
yading@10:             if(add){
yading@10:                 v += dst[x + src_x];
yading@10:                 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
yading@10:                 if(v&(~255)) v= ~(v>>31);
yading@10:                 dst8[x + y*src_stride] = v;
yading@10:             }else{
yading@10:                 dst[x + src_x] -= v;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_snow_reset_contexts(SnowContext *s){ //FIXME better initial contexts
yading@10:     int plane_index, level, orientation;
yading@10: 
yading@10:     for(plane_index=0; plane_index<3; plane_index++){
yading@10:         for(level=0; level<MAX_DECOMPOSITIONS; level++){
yading@10:             for(orientation=level ? 1:0; orientation<4; orientation++){
yading@10:                 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
yading@10:             }
yading@10:         }
yading@10:     }
yading@10:     memset(s->header_state, MID_STATE, sizeof(s->header_state));
yading@10:     memset(s->block_state, MID_STATE, sizeof(s->block_state));
yading@10: }
yading@10: 
yading@10: int ff_snow_alloc_blocks(SnowContext *s){
yading@10:     int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
yading@10:     int h= -((-s->avctx->height)>>LOG2_MB_SIZE);
yading@10: 
yading@10:     s->b_width = w;
yading@10:     s->b_height= h;
yading@10: 
yading@10:     av_free(s->block);
yading@10:     s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void init_qexp(void){
yading@10:     int i;
yading@10:     double v=128;
yading@10: 
yading@10:     for(i=0; i<QROOT; i++){
yading@10:         ff_qexp[i]= lrintf(v);
yading@10:         v *= pow(2, 1.0 / QROOT);
yading@10:     }
yading@10: }
yading@10: static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, int stride, int b_w, int b_h, int dx, int dy){
yading@10:     static const uint8_t weight[64]={
yading@10:     8,7,6,5,4,3,2,1,
yading@10:     7,7,0,0,0,0,0,1,
yading@10:     6,0,6,0,0,0,2,0,
yading@10:     5,0,0,5,0,3,0,0,
yading@10:     4,0,0,0,4,0,0,0,
yading@10:     3,0,0,5,0,3,0,0,
yading@10:     2,0,6,0,0,0,2,0,
yading@10:     1,7,0,0,0,0,0,1,
yading@10:     };
yading@10: 
yading@10:     static const uint8_t brane[256]={
yading@10:     0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
yading@10:     0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
yading@10:     0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
yading@10:     0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
yading@10:     0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
yading@10:     0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
yading@10:     0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
yading@10:     0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
yading@10:     0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
yading@10:     0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
yading@10:     0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
yading@10:     0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
yading@10:     0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
yading@10:     0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
yading@10:     0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
yading@10:     0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
yading@10:     };
yading@10: 
yading@10:     static const uint8_t needs[16]={
yading@10:     0,1,0,0,
yading@10:     2,4,2,0,
yading@10:     0,1,0,0,
yading@10:     15
yading@10:     };
yading@10: 
yading@10:     int x, y, b, r, l;
yading@10:     int16_t tmpIt   [64*(32+HTAPS_MAX)];
yading@10:     uint8_t tmp2t[3][64*(32+HTAPS_MAX)];
yading@10:     int16_t *tmpI= tmpIt;
yading@10:     uint8_t *tmp2= tmp2t[0];
yading@10:     const uint8_t *hpel[11];
yading@10:     av_assert2(dx<16 && dy<16);
yading@10:     r= brane[dx + 16*dy]&15;
yading@10:     l= brane[dx + 16*dy]>>4;
yading@10: 
yading@10:     b= needs[l] | needs[r];
yading@10:     if(p && !p->diag_mc)
yading@10:         b= 15;
yading@10: 
yading@10:     if(b&5){
yading@10:         for(y=0; y < b_h+HTAPS_MAX-1; y++){
yading@10:             for(x=0; x < b_w; x++){
yading@10:                 int a_1=src[x + HTAPS_MAX/2-4];
yading@10:                 int a0= src[x + HTAPS_MAX/2-3];
yading@10:                 int a1= src[x + HTAPS_MAX/2-2];
yading@10:                 int a2= src[x + HTAPS_MAX/2-1];
yading@10:                 int a3= src[x + HTAPS_MAX/2+0];
yading@10:                 int a4= src[x + HTAPS_MAX/2+1];
yading@10:                 int a5= src[x + HTAPS_MAX/2+2];
yading@10:                 int a6= src[x + HTAPS_MAX/2+3];
yading@10:                 int am=0;
yading@10:                 if(!p || p->fast_mc){
yading@10:                     am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
yading@10:                     tmpI[x]= am;
yading@10:                     am= (am+16)>>5;
yading@10:                 }else{
yading@10:                     am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
yading@10:                     tmpI[x]= am;
yading@10:                     am= (am+32)>>6;
yading@10:                 }
yading@10: 
yading@10:                 if(am&(~255)) am= ~(am>>31);
yading@10:                 tmp2[x]= am;
yading@10:             }
yading@10:             tmpI+= 64;
yading@10:             tmp2+= 64;
yading@10:             src += stride;
yading@10:         }
yading@10:         src -= stride*y;
yading@10:     }
yading@10:     src += HTAPS_MAX/2 - 1;
yading@10:     tmp2= tmp2t[1];
yading@10: 
yading@10:     if(b&2){
yading@10:         for(y=0; y < b_h; y++){
yading@10:             for(x=0; x < b_w+1; x++){
yading@10:                 int a_1=src[x + (HTAPS_MAX/2-4)*stride];
yading@10:                 int a0= src[x + (HTAPS_MAX/2-3)*stride];
yading@10:                 int a1= src[x + (HTAPS_MAX/2-2)*stride];
yading@10:                 int a2= src[x + (HTAPS_MAX/2-1)*stride];
yading@10:                 int a3= src[x + (HTAPS_MAX/2+0)*stride];
yading@10:                 int a4= src[x + (HTAPS_MAX/2+1)*stride];
yading@10:                 int a5= src[x + (HTAPS_MAX/2+2)*stride];
yading@10:                 int a6= src[x + (HTAPS_MAX/2+3)*stride];
yading@10:                 int am=0;
yading@10:                 if(!p || p->fast_mc)
yading@10:                     am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
yading@10:                 else
yading@10:                     am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;
yading@10: 
yading@10:                 if(am&(~255)) am= ~(am>>31);
yading@10:                 tmp2[x]= am;
yading@10:             }
yading@10:             src += stride;
yading@10:             tmp2+= 64;
yading@10:         }
yading@10:         src -= stride*y;
yading@10:     }
yading@10:     src += stride*(HTAPS_MAX/2 - 1);
yading@10:     tmp2= tmp2t[2];
yading@10:     tmpI= tmpIt;
yading@10:     if(b&4){
yading@10:         for(y=0; y < b_h; y++){
yading@10:             for(x=0; x < b_w; x++){
yading@10:                 int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
yading@10:                 int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
yading@10:                 int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
yading@10:                 int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
yading@10:                 int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
yading@10:                 int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
yading@10:                 int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
yading@10:                 int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
yading@10:                 int am=0;
yading@10:                 if(!p || p->fast_mc)
yading@10:                     am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
yading@10:                 else
yading@10:                     am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
yading@10:                 if(am&(~255)) am= ~(am>>31);
yading@10:                 tmp2[x]= am;
yading@10:             }
yading@10:             tmpI+= 64;
yading@10:             tmp2+= 64;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     hpel[ 0]= src;
yading@10:     hpel[ 1]= tmp2t[0] + 64*(HTAPS_MAX/2-1);
yading@10:     hpel[ 2]= src + 1;
yading@10: 
yading@10:     hpel[ 4]= tmp2t[1];
yading@10:     hpel[ 5]= tmp2t[2];
yading@10:     hpel[ 6]= tmp2t[1] + 1;
yading@10: 
yading@10:     hpel[ 8]= src + stride;
yading@10:     hpel[ 9]= hpel[1] + 64;
yading@10:     hpel[10]= hpel[8] + 1;
yading@10: 
yading@10: #define MC_STRIDE(x) (needs[x] ? 64 : stride)
yading@10: 
yading@10:     if(b==15){
yading@10:         int dxy = dx / 8 + dy / 8 * 4;
yading@10:         const uint8_t *src1 = hpel[dxy    ];
yading@10:         const uint8_t *src2 = hpel[dxy + 1];
yading@10:         const uint8_t *src3 = hpel[dxy + 4];
yading@10:         const uint8_t *src4 = hpel[dxy + 5];
yading@10:         int stride1 = MC_STRIDE(dxy);
yading@10:         int stride2 = MC_STRIDE(dxy + 1);
yading@10:         int stride3 = MC_STRIDE(dxy + 4);
yading@10:         int stride4 = MC_STRIDE(dxy + 5);
yading@10:         dx&=7;
yading@10:         dy&=7;
yading@10:         for(y=0; y < b_h; y++){
yading@10:             for(x=0; x < b_w; x++){
yading@10:                 dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
yading@10:                          (8-dx)*   dy *src3[x] + dx*   dy *src4[x]+32)>>6;
yading@10:             }
yading@10:             src1+=stride1;
yading@10:             src2+=stride2;
yading@10:             src3+=stride3;
yading@10:             src4+=stride4;
yading@10:             dst +=stride;
yading@10:         }
yading@10:     }else{
yading@10:         const uint8_t *src1= hpel[l];
yading@10:         const uint8_t *src2= hpel[r];
yading@10:         int stride1 = MC_STRIDE(l);
yading@10:         int stride2 = MC_STRIDE(r);
yading@10:         int a= weight[((dx&7) + (8*(dy&7)))];
yading@10:         int b= 8-a;
yading@10:         for(y=0; y < b_h; y++){
yading@10:             for(x=0; x < b_w; x++){
yading@10:                 dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
yading@10:             }
yading@10:             src1+=stride1;
yading@10:             src2+=stride2;
yading@10:             dst +=stride;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
yading@10:     if(block->type & BLOCK_INTRA){
yading@10:         int x, y;
yading@10:         const unsigned color  = block->color[plane_index];
yading@10:         const unsigned color4 = color*0x01010101;
yading@10:         if(b_w==32){
yading@10:             for(y=0; y < b_h; y++){
yading@10:                 *(uint32_t*)&dst[0 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[4 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[8 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[12+ y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[16+ y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[20+ y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[24+ y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[28+ y*stride]= color4;
yading@10:             }
yading@10:         }else if(b_w==16){
yading@10:             for(y=0; y < b_h; y++){
yading@10:                 *(uint32_t*)&dst[0 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[4 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[8 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[12+ y*stride]= color4;
yading@10:             }
yading@10:         }else if(b_w==8){
yading@10:             for(y=0; y < b_h; y++){
yading@10:                 *(uint32_t*)&dst[0 + y*stride]= color4;
yading@10:                 *(uint32_t*)&dst[4 + y*stride]= color4;
yading@10:             }
yading@10:         }else if(b_w==4){
yading@10:             for(y=0; y < b_h; y++){
yading@10:                 *(uint32_t*)&dst[0 + y*stride]= color4;
yading@10:             }
yading@10:         }else{
yading@10:             for(y=0; y < b_h; y++){
yading@10:                 for(x=0; x < b_w; x++){
yading@10:                     dst[x + y*stride]= color;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }else{
yading@10:         uint8_t *src= s->last_picture[block->ref].data[plane_index];
yading@10:         const int scale= plane_index ?  (2*s->mv_scale)>>s->chroma_h_shift : 2*s->mv_scale;
yading@10:         int mx= block->mx*scale;
yading@10:         int my= block->my*scale;
yading@10:         const int dx= mx&15;
yading@10:         const int dy= my&15;
yading@10:         const int tab_index= 3 - (b_w>>2) + (b_w>>4);
yading@10:         sx += (mx>>4) - (HTAPS_MAX/2-1);
yading@10:         sy += (my>>4) - (HTAPS_MAX/2-1);
yading@10:         src += sx + sy*stride;
yading@10:         if(   (unsigned)sx >= FFMAX(w - b_w - (HTAPS_MAX-2), 0)
yading@10:            || (unsigned)sy >= FFMAX(h - b_h - (HTAPS_MAX-2), 0)){
yading@10:             s->vdsp.emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h);
yading@10:             src= tmp + MB_SIZE;
yading@10:         }
yading@10: 
yading@10:         av_assert2(s->chroma_h_shift == s->chroma_v_shift); // only one mv_scale
yading@10: 
yading@10:         av_assert2(b_w>1 && b_h>1);
yading@10:         av_assert2((tab_index>=0 && tab_index<4) || b_w==32);
yading@10:         if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
yading@10:             mc_block(&s->plane[plane_index], dst, src, stride, b_w, b_h, dx, dy);
yading@10:         else if(b_w==32){
yading@10:             int y;
yading@10:             for(y=0; y<b_h; y+=16){
yading@10:                 s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
yading@10:                 s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
yading@10:             }
yading@10:         }else if(b_w==b_h)
yading@10:             s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
yading@10:         else if(b_w==2*b_h){
yading@10:             s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst    ,src + 3       + 3*stride,stride);
yading@10:             s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
yading@10:         }else{
yading@10:             av_assert2(2*b_w==b_h);
yading@10:             s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst           ,src + 3 + 3*stride           ,stride);
yading@10:             s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: #define mca(dx,dy,b_w)\
yading@10: static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, ptrdiff_t stride, int h){\
yading@10:     av_assert2(h==b_w);\
yading@10:     mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\
yading@10: }
yading@10: 
yading@10: mca( 0, 0,16)
yading@10: mca( 8, 0,16)
yading@10: mca( 0, 8,16)
yading@10: mca( 8, 8,16)
yading@10: mca( 0, 0,8)
yading@10: mca( 8, 0,8)
yading@10: mca( 0, 8,8)
yading@10: mca( 8, 8,8)
yading@10: 
yading@10: av_cold int ff_snow_common_init(AVCodecContext *avctx){
yading@10:     SnowContext *s = avctx->priv_data;
yading@10:     int width, height;
yading@10:     int i, j;
yading@10: 
yading@10:     s->avctx= avctx;
yading@10:     s->max_ref_frames=1; //just make sure it's not an invalid value in case of no initial keyframe
yading@10: 
yading@10:     ff_dsputil_init(&s->dsp, avctx);
yading@10:     ff_hpeldsp_init(&s->hdsp, avctx->flags);
yading@10:     ff_videodsp_init(&s->vdsp, 8);
yading@10:     ff_dwt_init(&s->dwt);
yading@10:     ff_h264qpel_init(&s->h264qpel, 8);
yading@10: 
yading@10: #define mcf(dx,dy)\
yading@10:     s->dsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
yading@10:     s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
yading@10:         s->h264qpel.put_h264_qpel_pixels_tab[0][dy+dx/4];\
yading@10:     s->dsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
yading@10:     s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
yading@10:         s->h264qpel.put_h264_qpel_pixels_tab[1][dy+dx/4];
yading@10: 
yading@10:     mcf( 0, 0)
yading@10:     mcf( 4, 0)
yading@10:     mcf( 8, 0)
yading@10:     mcf(12, 0)
yading@10:     mcf( 0, 4)
yading@10:     mcf( 4, 4)
yading@10:     mcf( 8, 4)
yading@10:     mcf(12, 4)
yading@10:     mcf( 0, 8)
yading@10:     mcf( 4, 8)
yading@10:     mcf( 8, 8)
yading@10:     mcf(12, 8)
yading@10:     mcf( 0,12)
yading@10:     mcf( 4,12)
yading@10:     mcf( 8,12)
yading@10:     mcf(12,12)
yading@10: 
yading@10: #define mcfh(dx,dy)\
yading@10:     s->hdsp.put_pixels_tab       [0][dy/4+dx/8]=\
yading@10:     s->hdsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
yading@10:         mc_block_hpel ## dx ## dy ## 16;\
yading@10:     s->hdsp.put_pixels_tab       [1][dy/4+dx/8]=\
yading@10:     s->hdsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
yading@10:         mc_block_hpel ## dx ## dy ## 8;
yading@10: 
yading@10:     mcfh(0, 0)
yading@10:     mcfh(8, 0)
yading@10:     mcfh(0, 8)
yading@10:     mcfh(8, 8)
yading@10: 
yading@10:     init_qexp();
yading@10: 
yading@10: //    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
yading@10: 
yading@10:     width= s->avctx->width;
yading@10:     height= s->avctx->height;
yading@10: 
yading@10:     FF_ALLOCZ_OR_GOTO(avctx, s->spatial_idwt_buffer, width * height * sizeof(IDWTELEM), fail);
yading@10:     FF_ALLOCZ_OR_GOTO(avctx, s->spatial_dwt_buffer,  width * height * sizeof(DWTELEM),  fail); //FIXME this does not belong here
yading@10:     FF_ALLOCZ_OR_GOTO(avctx, s->temp_dwt_buffer,     width * sizeof(DWTELEM),  fail);
yading@10:     FF_ALLOCZ_OR_GOTO(avctx, s->temp_idwt_buffer,    width * sizeof(IDWTELEM), fail);
yading@10:     FF_ALLOC_OR_GOTO(avctx,  s->run_buffer,          ((width + 1) >> 1) * ((height + 1) >> 1) * sizeof(*s->run_buffer), fail);
yading@10: 
yading@10:     for(i=0; i<MAX_REF_FRAMES; i++) {
yading@10:         for(j=0; j<MAX_REF_FRAMES; j++)
yading@10:             ff_scale_mv_ref[i][j] = 256*(i+1)/(j+1);
yading@10:         avcodec_get_frame_defaults(&s->last_picture[i]);
yading@10:     }
yading@10: 
yading@10:     avcodec_get_frame_defaults(&s->mconly_picture);
yading@10:     avcodec_get_frame_defaults(&s->current_picture);
yading@10: 
yading@10:     return 0;
yading@10: fail:
yading@10:     return AVERROR(ENOMEM);
yading@10: }
yading@10: 
yading@10: int ff_snow_common_init_after_header(AVCodecContext *avctx) {
yading@10:     SnowContext *s = avctx->priv_data;
yading@10:     int plane_index, level, orientation;
yading@10:     int ret, emu_buf_size;
yading@10: 
yading@10:     if(!s->scratchbuf) {
yading@10:         if ((ret = ff_get_buffer(s->avctx, &s->mconly_picture,
yading@10:                                  AV_GET_BUFFER_FLAG_REF)) < 0)
yading@10:             return ret;
yading@10:         FF_ALLOCZ_OR_GOTO(avctx, s->scratchbuf, FFMAX(s->mconly_picture.linesize[0], 2*avctx->width+256)*7*MB_SIZE, fail);
yading@10:         emu_buf_size = FFMAX(s->mconly_picture.linesize[0], 2*avctx->width+256) * (2 * MB_SIZE + HTAPS_MAX - 1);
yading@10:         FF_ALLOC_OR_GOTO(avctx, s->emu_edge_buffer, emu_buf_size, fail);
yading@10:     }
yading@10: 
yading@10:     if(s->mconly_picture.format != avctx->pix_fmt) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "pixel format changed\n");
yading@10:         return AVERROR_INVALIDDATA;
yading@10:     }
yading@10: 
yading@10:     for(plane_index=0; plane_index<3; plane_index++){
yading@10:         int w= s->avctx->width;
yading@10:         int h= s->avctx->height;
yading@10: 
yading@10:         if(plane_index){
yading@10:             w>>= s->chroma_h_shift;
yading@10:             h>>= s->chroma_v_shift;
yading@10:         }
yading@10:         s->plane[plane_index].width = w;
yading@10:         s->plane[plane_index].height= h;
yading@10: 
yading@10:         for(level=s->spatial_decomposition_count-1; level>=0; level--){
yading@10:             for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                 SubBand *b= &s->plane[plane_index].band[level][orientation];
yading@10: 
yading@10:                 b->buf= s->spatial_dwt_buffer;
yading@10:                 b->level= level;
yading@10:                 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
yading@10:                 b->width = (w + !(orientation&1))>>1;
yading@10:                 b->height= (h + !(orientation>1))>>1;
yading@10: 
yading@10:                 b->stride_line = 1 << (s->spatial_decomposition_count - level);
yading@10:                 b->buf_x_offset = 0;
yading@10:                 b->buf_y_offset = 0;
yading@10: 
yading@10:                 if(orientation&1){
yading@10:                     b->buf += (w+1)>>1;
yading@10:                     b->buf_x_offset = (w+1)>>1;
yading@10:                 }
yading@10:                 if(orientation>1){
yading@10:                     b->buf += b->stride>>1;
yading@10:                     b->buf_y_offset = b->stride_line >> 1;
yading@10:                 }
yading@10:                 b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);
yading@10: 
yading@10:                 if(level)
yading@10:                     b->parent= &s->plane[plane_index].band[level-1][orientation];
yading@10:                 //FIXME avoid this realloc
yading@10:                 av_freep(&b->x_coeff);
yading@10:                 b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
yading@10:             }
yading@10:             w= (w+1)>>1;
yading@10:             h= (h+1)>>1;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: fail:
yading@10:     return AVERROR(ENOMEM);
yading@10: }
yading@10: 
yading@10: #define USE_HALFPEL_PLANE 0
yading@10: 
yading@10: static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
yading@10:     int p,x,y;
yading@10: 
yading@10:     for(p=0; p<3; p++){
yading@10:         int is_chroma= !!p;
yading@10:         int w= is_chroma ? s->avctx->width >>s->chroma_h_shift : s->avctx->width;
yading@10:         int h= is_chroma ? s->avctx->height>>s->chroma_v_shift : s->avctx->height;
yading@10:         int ls= frame->linesize[p];
yading@10:         uint8_t *src= frame->data[p];
yading@10: 
yading@10:         halfpel[1][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
yading@10:         halfpel[2][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
yading@10:         halfpel[3][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
yading@10: 
yading@10:         halfpel[0][p]= src;
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int i= y*ls + x;
yading@10: 
yading@10:                 halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
yading@10:             }
yading@10:         }
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int i= y*ls + x;
yading@10: 
yading@10:                 halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
yading@10:             }
yading@10:         }
yading@10:         src= halfpel[1][p];
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int i= y*ls + x;
yading@10: 
yading@10:                 halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
yading@10:             }
yading@10:         }
yading@10: 
yading@10: //FIXME border!
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_snow_release_buffer(AVCodecContext *avctx)
yading@10: {
yading@10:     SnowContext *s = avctx->priv_data;
yading@10:     int i;
yading@10: 
yading@10:     if(s->last_picture[s->max_ref_frames-1].data[0]){
yading@10:         av_frame_unref(&s->last_picture[s->max_ref_frames-1]);
yading@10:         for(i=0; i<9; i++)
yading@10:             if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
yading@10:                 av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3]));
yading@10:     }
yading@10: }
yading@10: 
yading@10: int ff_snow_frame_start(SnowContext *s){
yading@10:    AVFrame tmp;
yading@10:    int i, ret;
yading@10:    int w= s->avctx->width; //FIXME round up to x16 ?
yading@10:    int h= s->avctx->height;
yading@10: 
yading@10:     if (s->current_picture.data[0] && !(s->avctx->flags&CODEC_FLAG_EMU_EDGE)) {
yading@10:         s->dsp.draw_edges(s->current_picture.data[0],
yading@10:                           s->current_picture.linesize[0], w   , h   ,
yading@10:                           EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
yading@10:         s->dsp.draw_edges(s->current_picture.data[1],
yading@10:                           s->current_picture.linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
yading@10:                           EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
yading@10:         s->dsp.draw_edges(s->current_picture.data[2],
yading@10:                           s->current_picture.linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
yading@10:                           EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
yading@10:     }
yading@10: 
yading@10:     ff_snow_release_buffer(s->avctx);
yading@10: 
yading@10:     av_frame_move_ref(&tmp, &s->last_picture[s->max_ref_frames-1]);
yading@10:     for(i=s->max_ref_frames-1; i>0; i--)
yading@10:         av_frame_move_ref(&s->last_picture[i+1], &s->last_picture[i]);
yading@10:     memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
yading@10:     if(USE_HALFPEL_PLANE && s->current_picture.data[0])
yading@10:         halfpel_interpol(s, s->halfpel_plane[0], &s->current_picture);
yading@10:     av_frame_move_ref(&s->last_picture[0], &s->current_picture);
yading@10:     av_frame_move_ref(&s->current_picture, &tmp);
yading@10: 
yading@10:     if(s->keyframe){
yading@10:         s->ref_frames= 0;
yading@10:     }else{
yading@10:         int i;
yading@10:         for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++)
yading@10:             if(i && s->last_picture[i-1].key_frame)
yading@10:                 break;
yading@10:         s->ref_frames= i;
yading@10:         if(s->ref_frames==0){
yading@10:             av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
yading@10:             return -1;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     if ((ret = ff_get_buffer(s->avctx, &s->current_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
yading@10:         return ret;
yading@10: 
yading@10:     s->current_picture.key_frame= s->keyframe;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: av_cold void ff_snow_common_end(SnowContext *s)
yading@10: {
yading@10:     int plane_index, level, orientation, i;
yading@10: 
yading@10:     av_freep(&s->spatial_dwt_buffer);
yading@10:     av_freep(&s->temp_dwt_buffer);
yading@10:     av_freep(&s->spatial_idwt_buffer);
yading@10:     av_freep(&s->temp_idwt_buffer);
yading@10:     av_freep(&s->run_buffer);
yading@10: 
yading@10:     s->m.me.temp= NULL;
yading@10:     av_freep(&s->m.me.scratchpad);
yading@10:     av_freep(&s->m.me.map);
yading@10:     av_freep(&s->m.me.score_map);
yading@10:     av_freep(&s->m.obmc_scratchpad);
yading@10: 
yading@10:     av_freep(&s->block);
yading@10:     av_freep(&s->scratchbuf);
yading@10:     av_freep(&s->emu_edge_buffer);
yading@10: 
yading@10:     for(i=0; i<MAX_REF_FRAMES; i++){
yading@10:         av_freep(&s->ref_mvs[i]);
yading@10:         av_freep(&s->ref_scores[i]);
yading@10:         if(s->last_picture[i].data[0]) {
yading@10:             av_assert0(s->last_picture[i].data[0] != s->current_picture.data[0]);
yading@10:             av_frame_unref(&s->last_picture[i]);
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     for(plane_index=0; plane_index<3; plane_index++){
yading@10:         for(level=s->spatial_decomposition_count-1; level>=0; level--){
yading@10:             for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                 SubBand *b= &s->plane[plane_index].band[level][orientation];
yading@10: 
yading@10:                 av_freep(&b->x_coeff);
yading@10:             }
yading@10:         }
yading@10:     }
yading@10:     av_frame_unref(&s->mconly_picture);
yading@10:     av_frame_unref(&s->current_picture);
yading@10: }