yading@10: /*
yading@10:  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
yading@10:  * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
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: #ifndef AVCODEC_SNOW_H
yading@10: #define AVCODEC_SNOW_H
yading@10: 
yading@10: #include "dsputil.h"
yading@10: #include "hpeldsp.h"
yading@10: #include "snow_dwt.h"
yading@10: 
yading@10: #include "rangecoder.h"
yading@10: #include "mathops.h"
yading@10: #include "mpegvideo.h"
yading@10: #include "h264qpel.h"
yading@10: 
yading@10: #define MID_STATE 128
yading@10: 
yading@10: #define MAX_PLANES 4
yading@10: #define QSHIFT 5
yading@10: #define QROOT (1<<QSHIFT)
yading@10: #define LOSSLESS_QLOG -128
yading@10: #define FRAC_BITS 4
yading@10: #define MAX_REF_FRAMES 8
yading@10: 
yading@10: #define LOG2_OBMC_MAX 8
yading@10: #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
yading@10: typedef struct BlockNode{
yading@10:     int16_t mx;
yading@10:     int16_t my;
yading@10:     uint8_t ref;
yading@10:     uint8_t color[3];
yading@10:     uint8_t type;
yading@10: //#define TYPE_SPLIT    1
yading@10: #define BLOCK_INTRA   1
yading@10: #define BLOCK_OPT     2
yading@10: //#define TYPE_NOCOLOR  4
yading@10:     uint8_t level; //FIXME merge into type?
yading@10: }BlockNode;
yading@10: 
yading@10: static const BlockNode null_block= { //FIXME add border maybe
yading@10:     .color= {128,128,128},
yading@10:     .mx= 0,
yading@10:     .my= 0,
yading@10:     .ref= 0,
yading@10:     .type= 0,
yading@10:     .level= 0,
yading@10: };
yading@10: 
yading@10: #define LOG2_MB_SIZE 4
yading@10: #define MB_SIZE (1<<LOG2_MB_SIZE)
yading@10: #define ENCODER_EXTRA_BITS 4
yading@10: #define HTAPS_MAX 8
yading@10: 
yading@10: typedef struct x_and_coeff{
yading@10:     int16_t x;
yading@10:     uint16_t coeff;
yading@10: } x_and_coeff;
yading@10: 
yading@10: typedef struct SubBand{
yading@10:     int level;
yading@10:     int stride;
yading@10:     int width;
yading@10:     int height;
yading@10:     int qlog;        ///< log(qscale)/log[2^(1/6)]
yading@10:     DWTELEM *buf;
yading@10:     IDWTELEM *ibuf;
yading@10:     int buf_x_offset;
yading@10:     int buf_y_offset;
yading@10:     int stride_line; ///< Stride measured in lines, not pixels.
yading@10:     x_and_coeff * x_coeff;
yading@10:     struct SubBand *parent;
yading@10:     uint8_t state[/*7*2*/ 7 + 512][32];
yading@10: }SubBand;
yading@10: 
yading@10: typedef struct Plane{
yading@10:     int width;
yading@10:     int height;
yading@10:     SubBand band[MAX_DECOMPOSITIONS][4];
yading@10: 
yading@10:     int htaps;
yading@10:     int8_t hcoeff[HTAPS_MAX/2];
yading@10:     int diag_mc;
yading@10:     int fast_mc;
yading@10: 
yading@10:     int last_htaps;
yading@10:     int8_t last_hcoeff[HTAPS_MAX/2];
yading@10:     int last_diag_mc;
yading@10: }Plane;
yading@10: 
yading@10: typedef struct SnowContext{
yading@10:     AVClass *class;
yading@10:     AVCodecContext *avctx;
yading@10:     RangeCoder c;
yading@10:     DSPContext dsp;
yading@10:     HpelDSPContext hdsp;
yading@10:     VideoDSPContext vdsp;
yading@10:     H264QpelContext h264qpel;
yading@10:     SnowDWTContext dwt;
yading@10:     AVFrame new_picture;
yading@10:     AVFrame input_picture;              ///< new_picture with the internal linesizes
yading@10:     AVFrame current_picture;
yading@10:     AVFrame last_picture[MAX_REF_FRAMES];
yading@10:     uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
yading@10:     AVFrame mconly_picture;
yading@10: //     uint8_t q_context[16];
yading@10:     uint8_t header_state[32];
yading@10:     uint8_t block_state[128 + 32*128];
yading@10:     int keyframe;
yading@10:     int always_reset;
yading@10:     int version;
yading@10:     int spatial_decomposition_type;
yading@10:     int last_spatial_decomposition_type;
yading@10:     int temporal_decomposition_type;
yading@10:     int spatial_decomposition_count;
yading@10:     int last_spatial_decomposition_count;
yading@10:     int temporal_decomposition_count;
yading@10:     int max_ref_frames;
yading@10:     int ref_frames;
yading@10:     int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
yading@10:     uint32_t *ref_scores[MAX_REF_FRAMES];
yading@10:     DWTELEM *spatial_dwt_buffer;
yading@10:     DWTELEM *temp_dwt_buffer;
yading@10:     IDWTELEM *spatial_idwt_buffer;
yading@10:     IDWTELEM *temp_idwt_buffer;
yading@10:     int *run_buffer;
yading@10:     int colorspace_type;
yading@10:     int chroma_h_shift;
yading@10:     int chroma_v_shift;
yading@10:     int spatial_scalability;
yading@10:     int qlog;
yading@10:     int last_qlog;
yading@10:     int lambda;
yading@10:     int lambda2;
yading@10:     int pass1_rc;
yading@10:     int mv_scale;
yading@10:     int last_mv_scale;
yading@10:     int qbias;
yading@10:     int last_qbias;
yading@10: #define QBIAS_SHIFT 3
yading@10:     int b_width;
yading@10:     int b_height;
yading@10:     int block_max_depth;
yading@10:     int last_block_max_depth;
yading@10:     Plane plane[MAX_PLANES];
yading@10:     BlockNode *block;
yading@10: #define ME_CACHE_SIZE 1024
yading@10:     unsigned me_cache[ME_CACHE_SIZE];
yading@10:     unsigned me_cache_generation;
yading@10:     slice_buffer sb;
yading@10:     int memc_only;
yading@10:     int no_bitstream;
yading@10: 
yading@10:     MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
yading@10: 
yading@10:     uint8_t *scratchbuf;
yading@10:     uint8_t *emu_edge_buffer;
yading@10: }SnowContext;
yading@10: 
yading@10: /* Tables */
yading@10: extern const uint8_t * const ff_obmc_tab[4];
yading@10: extern uint8_t ff_qexp[QROOT];
yading@10: extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
yading@10: 
yading@10: /* C bits used by mmx/sse2/altivec */
yading@10: 
yading@10: static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
yading@10:     (*i) = (width) - 2;
yading@10: 
yading@10:     if (width & 1){
yading@10:         low[(*i)+1] = low[((*i)+1)>>1];
yading@10:         (*i)--;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
yading@10:     for (; (*i)>=0; (*i)-=2){
yading@10:         low[(*i)+1] = high[(*i)>>1];
yading@10:         low[*i] = low[(*i)>>1];
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
yading@10:     for(; i<w; i++){
yading@10:         dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
yading@10:     }
yading@10: 
yading@10:     if((width^lift_high)&1){
yading@10:         dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
yading@10:         for(; i<w; i++){
yading@10:             dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
yading@10:         }
yading@10: 
yading@10:         if(width&1){
yading@10:             dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
yading@10:         }
yading@10: }
yading@10: 
yading@10: /* common code */
yading@10: 
yading@10: int ff_snow_common_init(AVCodecContext *avctx);
yading@10: int ff_snow_common_init_after_header(AVCodecContext *avctx);
yading@10: void ff_snow_common_end(SnowContext *s);
yading@10: void ff_snow_release_buffer(AVCodecContext *avctx);
yading@10: void ff_snow_reset_contexts(SnowContext *s);
yading@10: int ff_snow_alloc_blocks(SnowContext *s);
yading@10: int ff_snow_frame_start(SnowContext *s);
yading@10: void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride,
yading@10:                      int sx, int sy, int b_w, int b_h, BlockNode *block,
yading@10:                      int plane_index, int w, int h);
yading@10: /* common inline functions */
yading@10: //XXX doublecheck all of them should stay inlined
yading@10: 
yading@10: static inline void snow_set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
yading@10:     const int w= s->b_width << s->block_max_depth;
yading@10:     const int rem_depth= s->block_max_depth - level;
yading@10:     const int index= (x + y*w) << rem_depth;
yading@10:     const int block_w= 1<<rem_depth;
yading@10:     BlockNode block;
yading@10:     int i,j;
yading@10: 
yading@10:     block.color[0]= l;
yading@10:     block.color[1]= cb;
yading@10:     block.color[2]= cr;
yading@10:     block.mx= mx;
yading@10:     block.my= my;
yading@10:     block.ref= ref;
yading@10:     block.type= type;
yading@10:     block.level= level;
yading@10: 
yading@10:     for(j=0; j<block_w; j++){
yading@10:         for(i=0; i<block_w; i++){
yading@10:             s->block[index + i + j*w]= block;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
yading@10:                            const BlockNode *left, const BlockNode *top, const BlockNode *tr){
yading@10:     if(s->ref_frames == 1){
yading@10:         *mx = mid_pred(left->mx, top->mx, tr->mx);
yading@10:         *my = mid_pred(left->my, top->my, tr->my);
yading@10:     }else{
yading@10:         const int *scale = ff_scale_mv_ref[ref];
yading@10:         *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
yading@10:                        (top ->mx * scale[top ->ref] + 128) >>8,
yading@10:                        (tr  ->mx * scale[tr  ->ref] + 128) >>8);
yading@10:         *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
yading@10:                        (top ->my * scale[top ->ref] + 128) >>8,
yading@10:                        (tr  ->my * scale[tr  ->ref] + 128) >>8);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline int same_block(BlockNode *a, BlockNode *b){
yading@10:     if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
yading@10:         return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
yading@10:     }else{
yading@10:         return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
yading@10:     }
yading@10: }
yading@10: 
yading@10: //FIXME name cleanup (b_w, block_w, b_width stuff)
yading@10: //XXX should we really inline it?
yading@10: static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
yading@10:     const int b_width = s->b_width  << s->block_max_depth;
yading@10:     const int b_height= s->b_height << s->block_max_depth;
yading@10:     const int b_stride= b_width;
yading@10:     BlockNode *lt= &s->block[b_x + b_y*b_stride];
yading@10:     BlockNode *rt= lt+1;
yading@10:     BlockNode *lb= lt+b_stride;
yading@10:     BlockNode *rb= lb+1;
yading@10:     uint8_t *block[4];
yading@10:     int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
yading@10:     uint8_t *tmp = s->scratchbuf;
yading@10:     uint8_t *ptmp;
yading@10:     int x,y;
yading@10: 
yading@10:     if(b_x<0){
yading@10:         lt= rt;
yading@10:         lb= rb;
yading@10:     }else if(b_x + 1 >= b_width){
yading@10:         rt= lt;
yading@10:         rb= lb;
yading@10:     }
yading@10:     if(b_y<0){
yading@10:         lt= lb;
yading@10:         rt= rb;
yading@10:     }else if(b_y + 1 >= b_height){
yading@10:         lb= lt;
yading@10:         rb= rt;
yading@10:     }
yading@10: 
yading@10:     if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
yading@10:         obmc -= src_x;
yading@10:         b_w += src_x;
yading@10:         if(!sliced && !offset_dst)
yading@10:             dst -= src_x;
yading@10:         src_x=0;
yading@10:     }else if(src_x + b_w > w){
yading@10:         b_w = w - src_x;
yading@10:     }
yading@10:     if(src_y<0){
yading@10:         obmc -= src_y*obmc_stride;
yading@10:         b_h += src_y;
yading@10:         if(!sliced && !offset_dst)
yading@10:             dst -= src_y*dst_stride;
yading@10:         src_y=0;
yading@10:     }else if(src_y + b_h> h){
yading@10:         b_h = h - src_y;
yading@10:     }
yading@10: 
yading@10:     if(b_w<=0 || b_h<=0) return;
yading@10: 
yading@10:     av_assert2(src_stride > 2*MB_SIZE + 5);
yading@10: 
yading@10:     if(!sliced && offset_dst)
yading@10:         dst += src_x + src_y*dst_stride;
yading@10:     dst8+= src_x + src_y*src_stride;
yading@10: //    src += src_x + src_y*src_stride;
yading@10: 
yading@10:     ptmp= tmp + 3*tmp_step;
yading@10:     block[0]= ptmp;
yading@10:     ptmp+=tmp_step;
yading@10:     ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
yading@10: 
yading@10:     if(same_block(lt, rt)){
yading@10:         block[1]= block[0];
yading@10:     }else{
yading@10:         block[1]= ptmp;
yading@10:         ptmp+=tmp_step;
yading@10:         ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
yading@10:     }
yading@10: 
yading@10:     if(same_block(lt, lb)){
yading@10:         block[2]= block[0];
yading@10:     }else if(same_block(rt, lb)){
yading@10:         block[2]= block[1];
yading@10:     }else{
yading@10:         block[2]= ptmp;
yading@10:         ptmp+=tmp_step;
yading@10:         ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
yading@10:     }
yading@10: 
yading@10:     if(same_block(lt, rb) ){
yading@10:         block[3]= block[0];
yading@10:     }else if(same_block(rt, rb)){
yading@10:         block[3]= block[1];
yading@10:     }else if(same_block(lb, rb)){
yading@10:         block[3]= block[2];
yading@10:     }else{
yading@10:         block[3]= ptmp;
yading@10:         ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
yading@10:     }
yading@10:     if(sliced){
yading@10:         s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
yading@10:     }else{
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:             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 + y*dst_stride];
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 + y*dst_stride] -= v;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
yading@10:     Plane *p= &s->plane[plane_index];
yading@10:     const int mb_w= s->b_width  << s->block_max_depth;
yading@10:     const int mb_h= s->b_height << s->block_max_depth;
yading@10:     int x, y, mb_x;
yading@10:     int block_size = MB_SIZE >> s->block_max_depth;
yading@10:     int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
yading@10:     int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
yading@10:     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
yading@10:     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
yading@10:     int ref_stride= s->current_picture.linesize[plane_index];
yading@10:     uint8_t *dst8= s->current_picture.data[plane_index];
yading@10:     int w= p->width;
yading@10:     int h= p->height;
yading@10:     av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares
yading@10:     if(s->keyframe || (s->avctx->debug&512)){
yading@10:         if(mb_y==mb_h)
yading@10:             return;
yading@10: 
yading@10:         if(add){
yading@10:             for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
yading@10:                 for(x=0; x<w; x++){
yading@10:                     int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
yading@10:                     v >>= FRAC_BITS;
yading@10:                     if(v&(~255)) v= ~(v>>31);
yading@10:                     dst8[x + y*ref_stride]= v;
yading@10:                 }
yading@10:             }
yading@10:         }else{
yading@10:             for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
yading@10:                 for(x=0; x<w; x++){
yading@10:                     buf[x + y*w]-= 128<<FRAC_BITS;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10: 
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     for(mb_x=0; mb_x<=mb_w; mb_x++){
yading@10:         add_yblock(s, 0, NULL, buf, dst8, obmc,
yading@10:                    block_w*mb_x - block_w/2,
yading@10:                    block_h*mb_y - block_h/2,
yading@10:                    block_w, block_h,
yading@10:                    w, h,
yading@10:                    w, ref_stride, obmc_stride,
yading@10:                    mb_x - 1, mb_y - 1,
yading@10:                    add, 1, plane_index);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
yading@10:     const int mb_h= s->b_height << s->block_max_depth;
yading@10:     int mb_y;
yading@10:     for(mb_y=0; mb_y<=mb_h; mb_y++)
yading@10:         predict_slice(s, buf, plane_index, add, mb_y);
yading@10: }
yading@10: 
yading@10: static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
yading@10:     const int w= s->b_width << s->block_max_depth;
yading@10:     const int rem_depth= s->block_max_depth - level;
yading@10:     const int index= (x + y*w) << rem_depth;
yading@10:     const int block_w= 1<<rem_depth;
yading@10:     const int block_h= 1<<rem_depth; //FIXME "w!=h"
yading@10:     BlockNode block;
yading@10:     int i,j;
yading@10: 
yading@10:     block.color[0]= l;
yading@10:     block.color[1]= cb;
yading@10:     block.color[2]= cr;
yading@10:     block.mx= mx;
yading@10:     block.my= my;
yading@10:     block.ref= ref;
yading@10:     block.type= type;
yading@10:     block.level= level;
yading@10: 
yading@10:     for(j=0; j<block_h; j++){
yading@10:         for(i=0; i<block_w; i++){
yading@10:             s->block[index + i + j*w]= block;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
yading@10:     SnowContext *s = c->avctx->priv_data;
yading@10:     const int offset[3]= {
yading@10:           y*c->  stride + x,
yading@10:         ((y*c->uvstride + x)>>s->chroma_h_shift),
yading@10:         ((y*c->uvstride + x)>>s->chroma_h_shift),
yading@10:     };
yading@10:     int i;
yading@10:     for(i=0; i<3; i++){
yading@10:         c->src[0][i]= src [i];
yading@10:         c->ref[0][i]= ref [i] + offset[i];
yading@10:     }
yading@10:     av_assert2(!ref_index);
yading@10: }
yading@10: 
yading@10: 
yading@10: /* bitstream functions */
yading@10: 
yading@10: extern const int8_t ff_quant3bA[256];
yading@10: 
yading@10: #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
yading@10: 
yading@10: static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
yading@10:     int i;
yading@10: 
yading@10:     if(v){
yading@10:         const int a= FFABS(v);
yading@10:         const int e= av_log2(a);
yading@10:         const int el= FFMIN(e, 10);
yading@10:         put_rac(c, state+0, 0);
yading@10: 
yading@10:         for(i=0; i<el; i++){
yading@10:             put_rac(c, state+1+i, 1);  //1..10
yading@10:         }
yading@10:         for(; i<e; i++){
yading@10:             put_rac(c, state+1+9, 1);  //1..10
yading@10:         }
yading@10:         put_rac(c, state+1+FFMIN(i,9), 0);
yading@10: 
yading@10:         for(i=e-1; i>=el; i--){
yading@10:             put_rac(c, state+22+9, (a>>i)&1); //22..31
yading@10:         }
yading@10:         for(; i>=0; i--){
yading@10:             put_rac(c, state+22+i, (a>>i)&1); //22..31
yading@10:         }
yading@10: 
yading@10:         if(is_signed)
yading@10:             put_rac(c, state+11 + el, v < 0); //11..21
yading@10:     }else{
yading@10:         put_rac(c, state+0, 1);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
yading@10:     if(get_rac(c, state+0))
yading@10:         return 0;
yading@10:     else{
yading@10:         int i, e, a;
yading@10:         e= 0;
yading@10:         while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
yading@10:             e++;
yading@10:         }
yading@10: 
yading@10:         a= 1;
yading@10:         for(i=e-1; i>=0; i--){
yading@10:             a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
yading@10:         }
yading@10: 
yading@10:         e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
yading@10:         return (a^e)-e;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
yading@10:     int i;
yading@10:     int r= log2>=0 ? 1<<log2 : 1;
yading@10: 
yading@10:     av_assert2(v>=0);
yading@10:     av_assert2(log2>=-4);
yading@10: 
yading@10:     while(v >= r){
yading@10:         put_rac(c, state+4+log2, 1);
yading@10:         v -= r;
yading@10:         log2++;
yading@10:         if(log2>0) r+=r;
yading@10:     }
yading@10:     put_rac(c, state+4+log2, 0);
yading@10: 
yading@10:     for(i=log2-1; i>=0; i--){
yading@10:         put_rac(c, state+31-i, (v>>i)&1);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
yading@10:     int i;
yading@10:     int r= log2>=0 ? 1<<log2 : 1;
yading@10:     int v=0;
yading@10: 
yading@10:     av_assert2(log2>=-4);
yading@10: 
yading@10:     while(log2<28 && get_rac(c, state+4+log2)){
yading@10:         v+= r;
yading@10:         log2++;
yading@10:         if(log2>0) r+=r;
yading@10:     }
yading@10: 
yading@10:     for(i=log2-1; i>=0; i--){
yading@10:         v+= get_rac(c, state+31-i)<<i;
yading@10:     }
yading@10: 
yading@10:     return v;
yading@10: }
yading@10: 
yading@10: static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     int x,y;
yading@10: 
yading@10:     int run, runs;
yading@10:     x_and_coeff *xc= b->x_coeff;
yading@10:     x_and_coeff *prev_xc= NULL;
yading@10:     x_and_coeff *prev2_xc= xc;
yading@10:     x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
yading@10:     x_and_coeff *prev_parent_xc= parent_xc;
yading@10: 
yading@10:     runs= get_symbol2(&s->c, b->state[30], 0);
yading@10:     if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
yading@10:     else           run= INT_MAX;
yading@10: 
yading@10:     for(y=0; y<h; y++){
yading@10:         int v=0;
yading@10:         int lt=0, t=0, rt=0;
yading@10: 
yading@10:         if(y && prev_xc->x == 0){
yading@10:             rt= prev_xc->coeff;
yading@10:         }
yading@10:         for(x=0; x<w; x++){
yading@10:             int p=0;
yading@10:             const int l= v;
yading@10: 
yading@10:             lt= t; t= rt;
yading@10: 
yading@10:             if(y){
yading@10:                 if(prev_xc->x <= x)
yading@10:                     prev_xc++;
yading@10:                 if(prev_xc->x == x + 1)
yading@10:                     rt= prev_xc->coeff;
yading@10:                 else
yading@10:                     rt=0;
yading@10:             }
yading@10:             if(parent_xc){
yading@10:                 if(x>>1 > parent_xc->x){
yading@10:                     parent_xc++;
yading@10:                 }
yading@10:                 if(x>>1 == parent_xc->x){
yading@10:                     p= parent_xc->coeff;
yading@10:                 }
yading@10:             }
yading@10:             if(/*ll|*/l|lt|t|rt|p){
yading@10:                 int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
yading@10: 
yading@10:                 v=get_rac(&s->c, &b->state[0][context]);
yading@10:                 if(v){
yading@10:                     v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
yading@10:                     v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]);
yading@10: 
yading@10:                     xc->x=x;
yading@10:                     (xc++)->coeff= v;
yading@10:                 }
yading@10:             }else{
yading@10:                 if(!run){
yading@10:                     if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
yading@10:                     else           run= INT_MAX;
yading@10:                     v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
yading@10:                     v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
yading@10: 
yading@10:                     xc->x=x;
yading@10:                     (xc++)->coeff= v;
yading@10:                 }else{
yading@10:                     int max_run;
yading@10:                     run--;
yading@10:                     v=0;
yading@10:                     av_assert2(run >= 0);
yading@10:                     if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
yading@10:                     else  max_run= FFMIN(run, w-x-1);
yading@10:                     if(parent_xc)
yading@10:                         max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
yading@10:                     av_assert2(max_run >= 0 && max_run <= run);
yading@10: 
yading@10:                     x+= max_run;
yading@10:                     run-= max_run;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         (xc++)->x= w+1; //end marker
yading@10:         prev_xc= prev2_xc;
yading@10:         prev2_xc= xc;
yading@10: 
yading@10:         if(parent_xc){
yading@10:             if(y&1){
yading@10:                 while(parent_xc->x != parent->width+1)
yading@10:                     parent_xc++;
yading@10:                 parent_xc++;
yading@10:                 prev_parent_xc= parent_xc;
yading@10:             }else{
yading@10:                 parent_xc= prev_parent_xc;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     (xc++)->x= w+1; //end marker
yading@10: }
yading@10: 
yading@10: #endif /* AVCODEC_SNOW_H */