X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fsnow.c;h=a289ecfbced2e39d58ecd4cba38a95b5bd7429c4;hb=ad1c8dd6734f0aa7a7a87b4669a166715c114b46;hp=c22a910a59f1ee6521bdb0b70312e93a12e72d35;hpb=aedc908601de7396751a9a4504e064782d9f6a0b;p=ffmpeg diff --git a/libavcodec/snow.c b/libavcodec/snow.c index c22a910a59f..a289ecfbced 100644 --- a/libavcodec/snow.c +++ b/libavcodec/snow.c @@ -25,480 +25,50 @@ #include "dsputil.h" #include "dwt.h" #include "snow.h" +#include "snowdata.h" #include "rangecoder.h" #include "mathops.h" - -#include "mpegvideo.h" #include "h263.h" #undef NDEBUG #include -static const int8_t quant3bA[256]={ - 0, 0, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, - 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, -}; - -static const uint8_t obmc32[1024]={ - 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, - 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, - 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, - 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, - 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, - 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, - 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, - 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, - 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, - 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, - 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, - 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, - 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, - 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, - 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, - 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, - 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, - 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, - 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, - 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, - 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, - 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, - 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, - 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, - 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, - 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, - 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, - 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, - 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, - 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, - 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, - 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, - //error:0.000020 -}; -static const uint8_t obmc16[256]={ - 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, - 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, - 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, - 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, - 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, - 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, - 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, - 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, - 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, - 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, - 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, - 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, - 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, - 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, - 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, - 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, -//error:0.000015 -}; - -//linear *64 -static const uint8_t obmc8[64]={ - 4, 12, 20, 28, 28, 20, 12, 4, - 12, 36, 60, 84, 84, 60, 36, 12, - 20, 60,100,140,140,100, 60, 20, - 28, 84,140,196,196,140, 84, 28, - 28, 84,140,196,196,140, 84, 28, - 20, 60,100,140,140,100, 60, 20, - 12, 36, 60, 84, 84, 60, 36, 12, - 4, 12, 20, 28, 28, 20, 12, 4, -//error:0.000000 -}; - -//linear *64 -static const uint8_t obmc4[16]={ - 16, 48, 48, 16, - 48,144,144, 48, - 48,144,144, 48, - 16, 48, 48, 16, -//error:0.000000 -}; - -static const uint8_t * const obmc_tab[4]={ - obmc32, obmc16, obmc8, obmc4 -}; - -static int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; - -typedef struct BlockNode{ - int16_t mx; - int16_t my; - uint8_t ref; - uint8_t color[3]; - uint8_t type; -//#define TYPE_SPLIT 1 -#define BLOCK_INTRA 1 -#define BLOCK_OPT 2 -//#define TYPE_NOCOLOR 4 - uint8_t level; //FIXME merge into type? -}BlockNode; - -static const BlockNode null_block= { //FIXME add border maybe - .color= {128,128,128}, - .mx= 0, - .my= 0, - .ref= 0, - .type= 0, - .level= 0, -}; - -#define LOG2_MB_SIZE 4 -#define MB_SIZE (1<=el; i--){ - put_rac(c, state+22+9, (a>>i)&1); //22..31 - } - for(; i>=0; i--){ - put_rac(c, state+22+i, (a>>i)&1); //22..31 - } - - if(is_signed) - put_rac(c, state+11 + el, v < 0); //11..21 - }else{ - put_rac(c, state+0, 1); - } -} - -static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ - if(get_rac(c, state+0)) - return 0; - else{ - int i, e, a; - e= 0; - while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 - e++; - } - - a= 1; - for(i=e-1; i>=0; i--){ - a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 - } - - e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21 - return (a^e)-e; - } -} - -static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ - int i; - int r= log2>=0 ? 1<=0); - assert(log2>=-4); - - while(v >= r){ - put_rac(c, state+4+log2, 1); - v -= r; - log2++; - if(log2>0) r+=r; - } - put_rac(c, state+4+log2, 0); - - for(i=log2-1; i>=0; i--){ - put_rac(c, state+31-i, (v>>i)&1); - } -} - -static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ - int i; - int r= log2>=0 ? 1<=-4); - - while(get_rac(c, state+4+log2)){ - v+= r; - log2++; - if(log2>0) r+=r; - } - - for(i=log2-1; i>=0; i--){ - v+= get_rac(c, state+31-i)<width; - const int h= b->height; - int x,y; - - int run, runs; - x_and_coeff *xc= b->x_coeff; - x_and_coeff *prev_xc= NULL; - x_and_coeff *prev2_xc= xc; - x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; - x_and_coeff *prev_parent_xc= parent_xc; - - runs= get_symbol2(&s->c, b->state[30], 0); - if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); - else run= INT_MAX; - - for(y=0; yx == 0){ - rt= prev_xc->coeff; - } - for(x=0; xx <= x) - prev_xc++; - if(prev_xc->x == x + 1) - rt= prev_xc->coeff; - else - rt=0; - } - if(parent_xc){ - if(x>>1 > parent_xc->x){ - parent_xc++; - } - if(x>>1 == parent_xc->x){ - p= parent_xc->coeff; - } - } - if(/*ll|*/l|lt|t|rt|p){ - int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); - v=get_rac(&s->c, &b->state[0][context]); - if(v){ - v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); - v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]); +void ff_snow_inner_add_yblock(const 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; + IDWTELEM * dst; + for(y=0; y>1); + const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); + const uint8_t *obmc4= obmc3+ (obmc_stride>>1); + dst = slice_buffer_get_line(sb, src_y + y); + for(x=0; xx=x; - (xc++)->coeff= v; - } - }else{ - if(!run){ - if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); - else run= INT_MAX; - v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); - v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); - - xc->x=x; - (xc++)->coeff= v; - }else{ - int max_run; - run--; - v=0; - - if(y) max_run= FFMIN(run, prev_xc->x - x - 2); - else max_run= FFMIN(run, w-x-1); - if(parent_xc) - max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); - x+= max_run; - run-= max_run; - } + v <<= 8 - LOG2_OBMC_MAX; + if(FRAC_BITS != 8){ + v >>= 8 - FRAC_BITS; } - } - (xc++)->x= w+1; //end marker - prev_xc= prev2_xc; - prev2_xc= xc; - - if(parent_xc){ - if(y&1){ - while(parent_xc->x != parent->width+1) - parent_xc++; - parent_xc++; - prev_parent_xc= parent_xc; + if(add){ + v += dst[x + src_x]; + v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; + if(v&(~255)) v= ~(v>>31); + dst8[x + y*src_stride] = v; }else{ - parent_xc= prev_parent_xc; + dst[x + src_x] -= v; } } } - - (xc++)->x= w+1; //end marker -} - -static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){ - const int w= b->width; - int y; - const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); - int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); - int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; - int new_index = 0; - - if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){ - qadd= 0; - qmul= 1<stride_line + b->buf_y_offset) + b->buf_x_offset; - memset(line, 0, b->width*sizeof(IDWTELEM)); - v = b->x_coeff[new_index].coeff; - x = b->x_coeff[new_index++].x; - while(x < w){ - register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT; - register int u= -(v&1); - line[x] = (t^u) - u; - - v = b->x_coeff[new_index].coeff; - x = b->x_coeff[new_index++].x; - } - } - - /* Save our variables for the next slice. */ - save_state[0] = new_index; - - return; } -static void reset_contexts(SnowContext *s){ //FIXME better initial contexts +void ff_snow_reset_contexts(SnowContext *s){ //FIXME better initial contexts int plane_index, level, orientation; for(plane_index=0; plane_index<3; plane_index++){ @@ -512,7 +82,7 @@ static void reset_contexts(SnowContext *s){ //FIXME better initial contexts memset(s->block_state, MID_STATE, sizeof(s->block_state)); } -static int alloc_blocks(SnowContext *s){ +int ff_snow_alloc_blocks(SnowContext *s){ int w= -((-s->avctx->width )>>LOG2_MB_SIZE); int h= -((-s->avctx->height)>>LOG2_MB_SIZE); @@ -524,129 +94,15 @@ static int alloc_blocks(SnowContext *s){ return 0; } -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){ - const int w= s->b_width << s->block_max_depth; - const int rem_depth= s->block_max_depth - level; - const int index= (x + y*w) << rem_depth; - const int block_w= 1<block[index + i + j*w]= block; - } - } -} - -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){ - const int offset[3]= { - y*c-> stride + x, - ((y*c->uvstride + x)>>1), - ((y*c->uvstride + x)>>1), - }; +static void init_qexp(void){ int i; - for(i=0; i<3; i++){ - c->src[0][i]= src [i]; - c->ref[0][i]= ref [i] + offset[i]; - } - assert(!ref_index); -} - -static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref, - const BlockNode *left, const BlockNode *top, const BlockNode *tr){ - if(s->ref_frames == 1){ - *mx = mid_pred(left->mx, top->mx, tr->mx); - *my = mid_pred(left->my, top->my, tr->my); - }else{ - const int *scale = scale_mv_ref[ref]; - *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8, - (top ->mx * scale[top ->ref] + 128) >>8, - (tr ->mx * scale[tr ->ref] + 128) >>8); - *my = mid_pred((left->my * scale[left->ref] + 128) >>8, - (top ->my * scale[top ->ref] + 128) >>8, - (tr ->my * scale[tr ->ref] + 128) >>8); - } -} - -static av_always_inline int same_block(BlockNode *a, BlockNode *b){ - if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ - return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); - }else{ - return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); - } -} - -static void decode_q_branch(SnowContext *s, int level, int x, int y){ - const int w= s->b_width << s->block_max_depth; - const int rem_depth= s->block_max_depth - level; - const int index= (x + y*w) << rem_depth; - int trx= (x+1)<block[index-1] : &null_block; - const BlockNode *top = y ? &s->block[index-w] : &null_block; - const BlockNode *tl = y && x ? &s->block[index-w-1] : left; - const BlockNode *tr = y && trxblock[index-w+(1<level + 2*top->level + tl->level + tr->level; - - if(s->keyframe){ - set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA); - return; - } - - if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){ - int type, mx, my; - int l = left->color[0]; - int cb= left->color[1]; - int cr= left->color[2]; - int ref = 0; - int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); - int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx)); - int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my)); - - type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0; - - if(type){ - pred_mv(s, &mx, &my, 0, left, top, tr); - l += get_symbol(&s->c, &s->block_state[32], 1); - cb+= get_symbol(&s->c, &s->block_state[64], 1); - cr+= get_symbol(&s->c, &s->block_state[96], 1); - }else{ - if(s->ref_frames > 1) - ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0); - pred_mv(s, &mx, &my, ref, left, top, tr); - mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1); - my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1); - } - set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type); - }else{ - decode_q_branch(s, level+1, 2*x+0, 2*y+0); - decode_q_branch(s, level+1, 2*x+1, 2*y+0); - decode_q_branch(s, level+1, 2*x+0, 2*y+1); - decode_q_branch(s, level+1, 2*x+1, 2*y+1); - } -} - -static void decode_blocks(SnowContext *s){ - int x, y; - int w= s->b_width; - int h= s->b_height; + double v=128; - for(y=0; ytype & BLOCK_INTRA){ int x, y; - const int color = block->color[plane_index]; - const int color4= color*0x01010101; + const unsigned color = block->color[plane_index]; + const unsigned color4 = color*0x01010101; if(b_w==32){ for(y=0; y < b_h; y++){ *(uint32_t*)&dst[0 + y*stride]= color4; @@ -926,503 +367,79 @@ static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, i } } -void ff_snow_inner_add_yblock(const 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; - IDWTELEM * dst; - for(y=0; y>1); - const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); - const uint8_t *obmc4= obmc3+ (obmc_stride>>1); - dst = slice_buffer_get_line(sb, src_y + y); - for(x=0; x>= 8 - FRAC_BITS; - } - if(add){ - v += dst[x + src_x]; - v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst8[x + y*src_stride] = v; - }else{ - dst[x + src_x] -= v; - } - } - } +#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){\ + assert(h==b_w);\ + mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\ } -//FIXME name cleanup (b_w, block_w, b_width stuff) -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){ - const int b_width = s->b_width << s->block_max_depth; - const int b_height= s->b_height << s->block_max_depth; - const int b_stride= b_width; - BlockNode *lt= &s->block[b_x + b_y*b_stride]; - BlockNode *rt= lt+1; - BlockNode *lb= lt+b_stride; - BlockNode *rb= lb+1; - uint8_t *block[4]; - int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; - uint8_t *tmp = s->scratchbuf; - uint8_t *ptmp; - int x,y; - - if(b_x<0){ - lt= rt; - lb= rb; - }else if(b_x + 1 >= b_width){ - rt= lt; - rb= lb; - } - if(b_y<0){ - lt= lb; - rt= rb; - }else if(b_y + 1 >= b_height){ - lb= lt; - rb= rt; - } +mca( 0, 0,16) +mca( 8, 0,16) +mca( 0, 8,16) +mca( 8, 8,16) +mca( 0, 0,8) +mca( 8, 0,8) +mca( 0, 8,8) +mca( 8, 8,8) - if(src_x<0){ //FIXME merge with prev & always round internal width up to *16 - obmc -= src_x; - b_w += src_x; - if(!sliced && !offset_dst) - dst -= src_x; - src_x=0; - }else if(src_x + b_w > w){ - b_w = w - src_x; - } - if(src_y<0){ - obmc -= src_y*obmc_stride; - b_h += src_y; - if(!sliced && !offset_dst) - dst -= src_y*dst_stride; - src_y=0; - }else if(src_y + b_h> h){ - b_h = h - src_y; - } +av_cold int ff_snow_common_init(AVCodecContext *avctx){ + SnowContext *s = avctx->priv_data; + int width, height; + int i, j; - if(b_w<=0 || b_h<=0) return; + s->avctx= avctx; + s->max_ref_frames=1; //just make sure its not an invalid value in case of no initial keyframe - assert(src_stride > 2*MB_SIZE + 5); + dsputil_init(&s->dsp, avctx); + ff_dwt_init(&s->dwt); - if(!sliced && offset_dst) - dst += src_x + src_y*dst_stride; - dst8+= src_x + src_y*src_stride; -// src += src_x + src_y*src_stride; +#define mcf(dx,dy)\ + s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\ + s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\ + s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\ + s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=\ + s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\ + s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4]; - ptmp= tmp + 3*tmp_step; - block[0]= ptmp; - ptmp+=tmp_step; - pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); + mcf( 0, 0) + mcf( 4, 0) + mcf( 8, 0) + mcf(12, 0) + mcf( 0, 4) + mcf( 4, 4) + mcf( 8, 4) + mcf(12, 4) + mcf( 0, 8) + mcf( 4, 8) + mcf( 8, 8) + mcf(12, 8) + mcf( 0,12) + mcf( 4,12) + mcf( 8,12) + mcf(12,12) - if(same_block(lt, rt)){ - block[1]= block[0]; - }else{ - block[1]= ptmp; - ptmp+=tmp_step; - pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); - } +#define mcfh(dx,dy)\ + s->dsp.put_pixels_tab [0][dy/4+dx/8]=\ + s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\ + mc_block_hpel ## dx ## dy ## 16;\ + s->dsp.put_pixels_tab [1][dy/4+dx/8]=\ + s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\ + mc_block_hpel ## dx ## dy ## 8; - if(same_block(lt, lb)){ - block[2]= block[0]; - }else if(same_block(rt, lb)){ - block[2]= block[1]; - }else{ - block[2]= ptmp; - ptmp+=tmp_step; - pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); - } + mcfh(0, 0) + mcfh(8, 0) + mcfh(0, 8) + mcfh(8, 8) - if(same_block(lt, rb) ){ - block[3]= block[0]; - }else if(same_block(rt, rb)){ - block[3]= block[1]; - }else if(same_block(lb, rb)){ - block[3]= block[2]; - }else{ - block[3]= ptmp; - pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); - } - if(sliced){ - s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); - }else{ - for(y=0; y>1); - const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); - const uint8_t *obmc4= obmc3+ (obmc_stride>>1); - for(x=0; x>= 8 - FRAC_BITS; - } - if(add){ - v += dst[x + y*dst_stride]; - v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst8[x + y*src_stride] = v; - }else{ - dst[x + y*dst_stride] -= v; - } - } - } - } -} + init_qexp(); -static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){ - Plane *p= &s->plane[plane_index]; - const int mb_w= s->b_width << s->block_max_depth; - const int mb_h= s->b_height << s->block_max_depth; - int x, y, mb_x; - int block_size = MB_SIZE >> s->block_max_depth; - int block_w = plane_index ? block_size/2 : block_size; - const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; - int obmc_stride= plane_index ? block_size : 2*block_size; - int ref_stride= s->current_picture.linesize[plane_index]; - uint8_t *dst8= s->current_picture.data[plane_index]; - int w= p->width; - int h= p->height; - - if(s->keyframe || (s->avctx->debug&512)){ - if(mb_y==mb_h) - return; - - if(add){ - for(y=block_w*mb_y; yline[y]; - for(x=0; x>= FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst8[x + y*ref_stride]= v; - } - } - }else{ - for(y=block_w*mb_y; yline[y]; - for(x=0; xchroma_h_shift, s->chroma_v_shift); - return; - } + width= s->avctx->width; + height= s->avctx->height; - for(mb_x=0; mb_x<=mb_w; mb_x++){ - add_yblock(s, 1, sb, old_buffer, dst8, obmc, - block_w*mb_x - block_w/2, - block_w*mb_y - block_w/2, - block_w, block_w, - w, h, - w, ref_stride, obmc_stride, - mb_x - 1, mb_y - 1, - add, 0, plane_index); - } -} - -static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){ - Plane *p= &s->plane[plane_index]; - const int mb_w= s->b_width << s->block_max_depth; - const int mb_h= s->b_height << s->block_max_depth; - int x, y, mb_x; - int block_size = MB_SIZE >> s->block_max_depth; - int block_w = plane_index ? block_size/2 : block_size; - const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; - const int obmc_stride= plane_index ? block_size : 2*block_size; - int ref_stride= s->current_picture.linesize[plane_index]; - uint8_t *dst8= s->current_picture.data[plane_index]; - int w= p->width; - int h= p->height; - - if(s->keyframe || (s->avctx->debug&512)){ - if(mb_y==mb_h) - return; - - if(add){ - for(y=block_w*mb_y; y>= FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst8[x + y*ref_stride]= v; - } - } - }else{ - for(y=block_w*mb_y; yb_height << s->block_max_depth; - int mb_y; - for(mb_y=0; mb_y<=mb_h; mb_y++) - predict_slice(s, buf, plane_index, add, mb_y); -} - -static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){ - const int w= b->width; - const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); - const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); - const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; - int x,y; - - if(s->qlog == LOSSLESS_QLOG) return; - - for(y=start_y; ystride_line) + b->buf_y_offset) + b->buf_x_offset; - for(x=0; x>(QEXPSHIFT)); //FIXME try different bias - }else if(i>0){ - line[x]= (( i*qmul + qadd)>>(QEXPSHIFT)); - } - } - } -} - -static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){ - const int w= b->width; - int x,y; - - IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning - IDWTELEM * prev; - - if (start_y != 0) - line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; - - for(y=start_y; ystride_line) + b->buf_y_offset) + b->buf_x_offset; - for(x=0; xspatial_decomposition_count; level++){ - for(orientation=level ? 1:0; orientation<4; orientation++){ - int q; - if (plane_index==2) q= s->plane[1].band[level][orientation].qlog; - else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog; - else q= get_symbol(&s->c, s->header_state, 1); - s->plane[plane_index].band[level][orientation].qlog= q; - } - } - } -} - -#define GET_S(dst, check) \ - tmp= get_symbol(&s->c, s->header_state, 0);\ - if(!(check)){\ - av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\ - return -1;\ - }\ - dst= tmp; - -static int decode_header(SnowContext *s){ - int plane_index, tmp; - uint8_t kstate[32]; - - memset(kstate, MID_STATE, sizeof(kstate)); - - s->keyframe= get_rac(&s->c, kstate); - if(s->keyframe || s->always_reset){ - reset_contexts(s); - s->spatial_decomposition_type= - s->qlog= - s->qbias= - s->mv_scale= - s->block_max_depth= 0; - } - if(s->keyframe){ - GET_S(s->version, tmp <= 0U) - s->always_reset= get_rac(&s->c, s->header_state); - s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0); - s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0); - GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS) - s->colorspace_type= get_symbol(&s->c, s->header_state, 0); - s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0); - s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0); - s->spatial_scalability= get_rac(&s->c, s->header_state); -// s->rate_scalability= get_rac(&s->c, s->header_state); - GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES) - s->max_ref_frames++; - - decode_qlogs(s); - } - - if(!s->keyframe){ - if(get_rac(&s->c, s->header_state)){ - for(plane_index=0; plane_index<2; plane_index++){ - int htaps, i, sum=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]; - } - p->hcoeff[0]= 32-sum; - } - s->plane[2].diag_mc= s->plane[1].diag_mc; - s->plane[2].htaps = s->plane[1].htaps; - memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff)); - } - if(get_rac(&s->c, s->header_state)){ - GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS) - decode_qlogs(s); - } - } - - s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1); - if(s->spatial_decomposition_type > 1U){ - av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type); - return -1; - } - if(FFMIN(s->avctx-> width>>s->chroma_h_shift, - s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){ - av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count); - return -1; - } - - s->qlog += get_symbol(&s->c, s->header_state, 1); - s->mv_scale += get_symbol(&s->c, s->header_state, 1); - s->qbias += get_symbol(&s->c, s->header_state, 1); - s->block_max_depth+= get_symbol(&s->c, s->header_state, 1); - if(s->block_max_depth > 1 || s->block_max_depth < 0){ - av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth); - s->block_max_depth= 0; - return -1; - } - - return 0; -} - -static void init_qexp(void){ - int i; - double v=128; - - for(i=0; ipriv_data; - int width, height; - int i, j; - - s->avctx= avctx; - s->max_ref_frames=1; //just make sure its not an invalid value in case of no initial keyframe - - dsputil_init(&s->dsp, avctx); - ff_dwt_init(&s->dwt); - -#define mcf(dx,dy)\ - s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\ - s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\ - s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\ - s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=\ - s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\ - s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4]; - - mcf( 0, 0) - mcf( 4, 0) - mcf( 8, 0) - mcf(12, 0) - mcf( 0, 4) - mcf( 4, 4) - mcf( 8, 4) - mcf(12, 4) - mcf( 0, 8) - mcf( 4, 8) - mcf( 8, 8) - mcf(12, 8) - mcf( 0,12) - mcf( 4,12) - mcf( 8,12) - mcf(12,12) - -#define mcfh(dx,dy)\ - s->dsp.put_pixels_tab [0][dy/4+dx/8]=\ - s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\ - mc_block_hpel ## dx ## dy ## 16;\ - s->dsp.put_pixels_tab [1][dy/4+dx/8]=\ - s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\ - mc_block_hpel ## dx ## dy ## 8; - - mcfh(0, 0) - mcfh(8, 0) - mcfh(0, 8) - mcfh(8, 8) - - if(!qexp[0]) - init_qexp(); - -// dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift); - - width= s->avctx->width; - height= s->avctx->height; - - s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM)); - s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this does not belong here + s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM)); + s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this does not belong here for(i=0; ipriv_data; int plane_index, level, orientation; @@ -1487,130 +504,11 @@ static int common_init_after_header(AVCodecContext *avctx){ return 0; } -#define QUANTIZE2 0 - -#if QUANTIZE2==1 -#define Q2_STEP 8 - -static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){ - SubBand *b= &p->band[level][orientation]; - int x, y; - int xo=0; - int yo=0; - int step= 1 << (s->spatial_decomposition_count - level); - - if(orientation&1) - xo= step>>1; - if(orientation&2) - yo= step>>1; - - //FIXME bias for nonzero ? - //FIXME optimize - memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP)); - for(y=0; yheight; y++){ - for(x=0; xwidth; x++){ - int sx= (x-xo + step/2) / step / Q2_STEP; - int sy= (y-yo + step/2) / step / Q2_STEP; - int v= r0[x + y*p->width] - r1[x + y*p->width]; - assert(sx>=0 && sy>=0 && sx < score_stride); - v= ((v+8)>>4)<<4; - score[sx + sy*score_stride] += v*v; - assert(score[sx + sy*score_stride] >= 0); - } - } -} - -static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){ - int level, orientation; - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - IDWTELEM *dst= buffer + (b->ibuf - s->spatial_idwt_buffer); - - dequantize(s, b, dst, b->stride); - } - } -} - -static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){ - int level, orientation, ys, xs, x, y, pass; - IDWTELEM best_dequant[height * stride]; - IDWTELEM idwt2_buffer[height * stride]; - const int score_stride= (width + 10)/Q2_STEP; - int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size - int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size - int threshold= (s->m.lambda * s->m.lambda) >> 6; - - //FIXME pass the copy cleanly ? - -// memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM)); - ff_spatial_dwt(buffer, width, height, stride, type, s->spatial_decomposition_count); - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer); - DWTELEM *src= buffer + (b-> buf - s->spatial_dwt_buffer); - assert(src == b->buf); // code does not depend on this but it is true currently - - quantize(s, b, dst, src, b->stride, s->qbias); - } - } - for(pass=0; pass<1; pass++){ - if(s->qbias == 0) //keyframe - continue; - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer); - IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer); - - for(ys= 0; ysspatial_decomposition_count); - find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation); - memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); - for(y=ys; yheight; y+= Q2_STEP){ - for(x=xs; xwidth; x+= Q2_STEP){ - if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++; - if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--; - //FIXME try more than just -- - } - } - dequantize_all(s, p, idwt2_buffer, width, height); - ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count); - find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation); - for(y=ys; yheight; y+= Q2_STEP){ - for(x=xs; xwidth; x+= Q2_STEP){ - int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride; - if(score[score_idx] <= best_score[score_idx] + threshold){ - best_score[score_idx]= score[score_idx]; - if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++; - if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--; - //FIXME copy instead - } - } - } - } - } - } - } - } - memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end -} - -#endif /* QUANTIZE2==1 */ - #define USE_HALFPEL_PLANE 0 static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){ int p,x,y; - assert(!(s->avctx->flags & CODEC_FLAG_EMU_EDGE)); - for(p=0; p<3; p++){ int is_chroma= !!p; int w= s->avctx->width >>is_chroma; @@ -1618,9 +516,9 @@ static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *fr int ls= frame->linesize[p]; uint8_t *src= frame->data[p]; - halfpel[1][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); - halfpel[2][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); - halfpel[3][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); + halfpel[1][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls); + halfpel[2][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls); + halfpel[3][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls); halfpel[0][p]= src; for(y=0; ypriv_data; int i; @@ -1662,12 +561,12 @@ static void release_buffer(AVCodecContext *avctx){ } } -static int frame_start(SnowContext *s){ +int ff_snow_frame_start(SnowContext *s){ AVFrame tmp; int w= s->avctx->width; //FIXME round up to x16 ? int h= s->avctx->height; - if(s->current_picture.data[0]){ + if (s->current_picture.data[0] && !(s->avctx->flags&CODEC_FLAG_EMU_EDGE)) { s->dsp.draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH , EDGE_WIDTH , EDGE_TOP | EDGE_BOTTOM); @@ -1679,7 +578,7 @@ static int frame_start(SnowContext *s){ EDGE_WIDTH/2, EDGE_WIDTH/2, EDGE_TOP | EDGE_BOTTOM); } - release_buffer(s->avctx); + ff_snow_release_buffer(s->avctx); tmp= s->last_picture[s->max_ref_frames-1]; memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame)); @@ -1703,7 +602,7 @@ static int frame_start(SnowContext *s){ } } - s->current_picture.reference= 1; + s->current_picture.reference= 3; if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){ av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); return -1; @@ -1714,7 +613,8 @@ static int frame_start(SnowContext *s){ return 0; } -static av_cold void common_end(SnowContext *s){ +av_cold void ff_snow_common_end(SnowContext *s) +{ int plane_index, level, orientation, i; av_freep(&s->spatial_dwt_buffer); @@ -1751,2078 +651,3 @@ static av_cold void common_end(SnowContext *s){ s->avctx->release_buffer(s->avctx, &s->current_picture); } -static av_cold int decode_init(AVCodecContext *avctx) -{ - avctx->pix_fmt= PIX_FMT_YUV420P; - - common_init(avctx); - - return 0; -} - -static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt){ - const uint8_t *buf = avpkt->data; - int buf_size = avpkt->size; - SnowContext *s = avctx->priv_data; - RangeCoder * const c= &s->c; - int bytes_read; - AVFrame *picture = data; - int level, orientation, plane_index; - - ff_init_range_decoder(c, buf, buf_size); - ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); - - s->current_picture.pict_type= AV_PICTURE_TYPE_I; //FIXME I vs. P - if(decode_header(s)<0) - return -1; - common_init_after_header(avctx); - - // realloc slice buffer for the case that spatial_decomposition_count changed - ff_slice_buffer_destroy(&s->sb); - ff_slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer); - - for(plane_index=0; plane_index<3; plane_index++){ - Plane *p= &s->plane[plane_index]; - p->fast_mc= p->diag_mc && p->htaps==6 && p->hcoeff[0]==40 - && p->hcoeff[1]==-10 - && p->hcoeff[2]==2; - } - - alloc_blocks(s); - - if(frame_start(s) < 0) - return -1; - //keyframe flag duplication mess FIXME - if(avctx->debug&FF_DEBUG_PICT_INFO) - av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog); - - decode_blocks(s); - - for(plane_index=0; plane_index<3; plane_index++){ - Plane *p= &s->plane[plane_index]; - int w= p->width; - int h= p->height; - int x, y; - int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */ - - if(s->avctx->debug&2048){ - memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); - predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); - - for(y=0; ycurrent_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]; - s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v; - } - } - } - - { - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - unpack_coeffs(s, b, b->parent, orientation); - } - } - } - - { - const int mb_h= s->b_height << s->block_max_depth; - const int block_size = MB_SIZE >> s->block_max_depth; - const int block_w = plane_index ? block_size/2 : block_size; - int mb_y; - DWTCompose cs[MAX_DECOMPOSITIONS]; - int yd=0, yq=0; - int y; - int end_y; - - ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count); - for(mb_y=0; mb_y<=mb_h; mb_y++){ - - int slice_starty = block_w*mb_y; - int slice_h = block_w*(mb_y+1); - if (!(s->keyframe || s->avctx->debug&512)){ - slice_starty = FFMAX(0, slice_starty - (block_w >> 1)); - slice_h -= (block_w >> 1); - } - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - int start_y; - int end_y; - int our_mb_start = mb_y; - int our_mb_end = (mb_y + 1); - const int extra= 3; - start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0); - end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra); - if (!(s->keyframe || s->avctx->debug&512)){ - start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level))); - end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level))); - } - start_y = FFMIN(b->height, start_y); - end_y = FFMIN(b->height, end_y); - - if (start_y != end_y){ - if (orientation == 0){ - SubBand * correlate_band = &p->band[0][0]; - int correlate_end_y = FFMIN(b->height, end_y + 1); - int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0)); - decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]); - correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y); - dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y); - } - else - decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]); - } - } - } - - for(; yddwt, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); - } - - if(s->qlog == LOSSLESS_QLOG){ - for(; yqsb, yq); - for(x=0; xsb, s->spatial_idwt_buffer, plane_index, 1, mb_y); - - y = FFMIN(p->height, slice_starty); - end_y = FFMIN(p->height, slice_h); - while(y < end_y) - ff_slice_buffer_release(&s->sb, y++); - } - - ff_slice_buffer_flush(&s->sb); - } - - } - - emms_c(); - - release_buffer(avctx); - - if(!(s->avctx->debug&2048)) - *picture= s->current_picture; - else - *picture= s->mconly_picture; - - *data_size = sizeof(AVFrame); - - bytes_read= c->bytestream - c->bytestream_start; - if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME - - return bytes_read; -} - -static av_cold int decode_end(AVCodecContext *avctx) -{ - SnowContext *s = avctx->priv_data; - - ff_slice_buffer_destroy(&s->sb); - - common_end(s); - - return 0; -} - -AVCodec ff_snow_decoder = { - .name = "snow", - .type = AVMEDIA_TYPE_VIDEO, - .id = CODEC_ID_SNOW, - .priv_data_size = sizeof(SnowContext), - .init = decode_init, - .close = decode_end, - .decode = decode_frame, - .capabilities = CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/, - .long_name = NULL_IF_CONFIG_SMALL("Snow"), -}; - -#if CONFIG_SNOW_ENCODER -static av_cold int encode_init(AVCodecContext *avctx) -{ - SnowContext *s = avctx->priv_data; - int plane_index; - - if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){ - av_log(avctx, AV_LOG_ERROR, "This codec is under development, files encoded with it may not be decodable with future versions!!!\n" - "Use vstrict=-2 / -strict -2 to use it anyway.\n"); - return -1; - } - - if(avctx->prediction_method == DWT_97 - && (avctx->flags & CODEC_FLAG_QSCALE) - && avctx->global_quality == 0){ - av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n"); - return -1; - } - - s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type - - s->mv_scale = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4; - s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0; - - for(plane_index=0; plane_index<3; plane_index++){ - s->plane[plane_index].diag_mc= 1; - s->plane[plane_index].htaps= 6; - s->plane[plane_index].hcoeff[0]= 40; - s->plane[plane_index].hcoeff[1]= -10; - s->plane[plane_index].hcoeff[2]= 2; - s->plane[plane_index].fast_mc= 1; - } - - common_init(avctx); - alloc_blocks(s); - - s->version=0; - - s->m.avctx = avctx; - s->m.flags = avctx->flags; - s->m.bit_rate= avctx->bit_rate; - - s->m.me.temp = - s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); - s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); - s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); - s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t)); - h263_encode_init(&s->m); //mv_penalty - - s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1); - - if(avctx->flags&CODEC_FLAG_PASS1){ - if(!avctx->stats_out) - avctx->stats_out = av_mallocz(256); - } - if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){ - if(ff_rate_control_init(&s->m) < 0) - return -1; - } - s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2)); - - avctx->coded_frame= &s->current_picture; - switch(avctx->pix_fmt){ -// case PIX_FMT_YUV444P: -// case PIX_FMT_YUV422P: - case PIX_FMT_YUV420P: - case PIX_FMT_GRAY8: -// case PIX_FMT_YUV411P: -// case PIX_FMT_YUV410P: - s->colorspace_type= 0; - break; -/* case PIX_FMT_RGB32: - s->colorspace= 1; - break;*/ - default: - av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n"); - return -1; - } -// avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift); - s->chroma_h_shift= 1; - s->chroma_v_shift= 1; - - ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp); - ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); - - s->avctx->get_buffer(s->avctx, &s->input_picture); - - if(s->avctx->me_method == ME_ITER){ - int i; - int size= s->b_width * s->b_height << 2*s->block_max_depth; - for(i=0; imax_ref_frames; i++){ - s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2])); - s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t)); - } - } - - return 0; -} - -//near copy & paste from dsputil, FIXME -static int pix_sum(uint8_t * pix, int line_size, int w) -{ - int s, i, j; - - s = 0; - for (i = 0; i < w; i++) { - for (j = 0; j < w; j++) { - s += pix[0]; - pix ++; - } - pix += line_size - w; - } - return s; -} - -//near copy & paste from dsputil, FIXME -static int pix_norm1(uint8_t * pix, int line_size, int w) -{ - int s, i, j; - uint32_t *sq = ff_squareTbl + 256; - - s = 0; - for (i = 0; i < w; i++) { - for (j = 0; j < w; j ++) { - s += sq[pix[0]]; - pix ++; - } - pix += line_size - w; - } - return s; -} - -//FIXME copy&paste -#define P_LEFT P[1] -#define P_TOP P[2] -#define P_TOPRIGHT P[3] -#define P_MEDIAN P[4] -#define P_MV1 P[9] -#define FLAG_QPEL 1 //must be 1 - -static int encode_q_branch(SnowContext *s, int level, int x, int y){ - uint8_t p_buffer[1024]; - uint8_t i_buffer[1024]; - uint8_t p_state[sizeof(s->block_state)]; - uint8_t i_state[sizeof(s->block_state)]; - RangeCoder pc, ic; - uint8_t *pbbak= s->c.bytestream; - uint8_t *pbbak_start= s->c.bytestream_start; - int score, score2, iscore, i_len, p_len, block_s, sum, base_bits; - const int w= s->b_width << s->block_max_depth; - const int h= s->b_height << s->block_max_depth; - const int rem_depth= s->block_max_depth - level; - const int index= (x + y*w) << rem_depth; - const int block_w= 1<<(LOG2_MB_SIZE - level); - int trx= (x+1)<block[index-1] : &null_block; - const BlockNode *top = y ? &s->block[index-w] : &null_block; - const BlockNode *right = trxblock[index+1] : &null_block; - const BlockNode *bottom= tryblock[index+w] : &null_block; - const BlockNode *tl = y && x ? &s->block[index-w-1] : left; - const BlockNode *tr = y && trxblock[index-w+(1<color[0]; - int pcb= left->color[1]; - int pcr= left->color[2]; - int pmx, pmy; - int mx=0, my=0; - int l,cr,cb; - const int stride= s->current_picture.linesize[0]; - const int uvstride= s->current_picture.linesize[1]; - uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w, - s->input_picture.data[1] + (x + y*uvstride)*block_w/2, - s->input_picture.data[2] + (x + y*uvstride)*block_w/2}; - int P[10][2]; - int16_t last_mv[3][2]; - int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused - const int shift= 1+qpel; - MotionEstContext *c= &s->m.me; - int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); - int mx_context= av_log2(2*FFABS(left->mx - top->mx)); - int my_context= av_log2(2*FFABS(left->my - top->my)); - int s_context= 2*left->level + 2*top->level + tl->level + tr->level; - int ref, best_ref, ref_score, ref_mx, ref_my; - - assert(sizeof(s->block_state) >= 256); - if(s->keyframe){ - set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); - return 0; - } - -// clip predictors / edge ? - - P_LEFT[0]= left->mx; - P_LEFT[1]= left->my; - P_TOP [0]= top->mx; - P_TOP [1]= top->my; - P_TOPRIGHT[0]= tr->mx; - P_TOPRIGHT[1]= tr->my; - - last_mv[0][0]= s->block[index].mx; - last_mv[0][1]= s->block[index].my; - last_mv[1][0]= right->mx; - last_mv[1][1]= right->my; - last_mv[2][0]= bottom->mx; - last_mv[2][1]= bottom->my; - - s->m.mb_stride=2; - s->m.mb_x= - s->m.mb_y= 0; - c->skip= 0; - - assert(c-> stride == stride); - assert(c->uvstride == uvstride); - - c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp); - c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp); - c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp); - c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV; - - c->xmin = - x*block_w - 16+3; - c->ymin = - y*block_w - 16+3; - c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3; - c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3; - - if(P_LEFT[0] > (c->xmax<xmax< (c->ymax<ymax< (c->xmax<xmax< (c->ymax<ymax<xmin<xmin< (c->xmax<xmax< (c->ymax<ymax<pred_x= P_LEFT[0]; - c->pred_y= P_LEFT[1]; - } else { - c->pred_x = P_MEDIAN[0]; - c->pred_y = P_MEDIAN[1]; - } - - score= INT_MAX; - best_ref= 0; - for(ref=0; refref_frames; ref++){ - init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0); - - ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv, - (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w); - - assert(ref_mx >= c->xmin); - assert(ref_mx <= c->xmax); - assert(ref_my >= c->ymin); - assert(ref_my <= c->ymax); - - ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w); - ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0); - ref_score+= 2*av_log2(2*ref)*c->penalty_factor; - if(s->ref_mvs[ref]){ - s->ref_mvs[ref][index][0]= ref_mx; - s->ref_mvs[ref][index][1]= ref_my; - s->ref_scores[ref][index]= ref_score; - } - if(score > ref_score){ - score= ref_score; - best_ref= ref; - mx= ref_mx; - my= ref_my; - } - } - //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2 - - // subpel search - base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start); - pc= s->c; - pc.bytestream_start= - pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo - memcpy(p_state, s->block_state, sizeof(s->block_state)); - - if(level!=s->block_max_depth) - put_rac(&pc, &p_state[4 + s_context], 1); - put_rac(&pc, &p_state[1 + left->type + top->type], 0); - if(s->ref_frames > 1) - put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0); - pred_mv(s, &pmx, &pmy, best_ref, left, top, tr); - put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1); - put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1); - p_len= pc.bytestream - pc.bytestream_start; - score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT; - - block_s= block_w*block_w; - sum = pix_sum(current_data[0], stride, block_w); - l= (sum + block_s/2)/block_s; - iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s; - - block_s= block_w*block_w>>2; - sum = pix_sum(current_data[1], uvstride, block_w>>1); - cb= (sum + block_s/2)/block_s; -// iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s; - sum = pix_sum(current_data[2], uvstride, block_w>>1); - cr= (sum + block_s/2)/block_s; -// iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s; - - ic= s->c; - ic.bytestream_start= - ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo - memcpy(i_state, s->block_state, sizeof(s->block_state)); - if(level!=s->block_max_depth) - put_rac(&ic, &i_state[4 + s_context], 1); - put_rac(&ic, &i_state[1 + left->type + top->type], 1); - put_symbol(&ic, &i_state[32], l-pl , 1); - put_symbol(&ic, &i_state[64], cb-pcb, 1); - put_symbol(&ic, &i_state[96], cr-pcr, 1); - i_len= ic.bytestream - ic.bytestream_start; - iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT; - -// assert(score==256*256*256*64-1); - assert(iscore < 255*255*256 + s->lambda2*10); - assert(iscore >= 0); - assert(l>=0 && l<=255); - assert(pl>=0 && pl<=255); - - if(level==0){ - int varc= iscore >> 8; - int vard= score >> 8; - if (vard <= 64 || vard < varc) - c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); - else - c->scene_change_score+= s->m.qscale; - } - - if(level!=s->block_max_depth){ - put_rac(&s->c, &s->block_state[4 + s_context], 0); - score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0); - score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0); - score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1); - score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1); - score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead - - if(score2 < score && score2 < iscore) - return score2; - } - - if(iscore < score){ - pred_mv(s, &pmx, &pmy, 0, left, top, tr); - memcpy(pbbak, i_buffer, i_len); - s->c= ic; - s->c.bytestream_start= pbbak_start; - s->c.bytestream= pbbak + i_len; - set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA); - memcpy(s->block_state, i_state, sizeof(s->block_state)); - return iscore; - }else{ - memcpy(pbbak, p_buffer, p_len); - s->c= pc; - s->c.bytestream_start= pbbak_start; - s->c.bytestream= pbbak + p_len; - set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0); - memcpy(s->block_state, p_state, sizeof(s->block_state)); - return score; - } -} - -static void encode_q_branch2(SnowContext *s, int level, int x, int y){ - const int w= s->b_width << s->block_max_depth; - const int rem_depth= s->block_max_depth - level; - const int index= (x + y*w) << rem_depth; - int trx= (x+1)<block[index]; - const BlockNode *left = x ? &s->block[index-1] : &null_block; - const BlockNode *top = y ? &s->block[index-w] : &null_block; - const BlockNode *tl = y && x ? &s->block[index-w-1] : left; - const BlockNode *tr = y && trxblock[index-w+(1<color[0]; - int pcb= left->color[1]; - int pcr= left->color[2]; - int pmx, pmy; - int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); - int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref; - int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref; - int s_context= 2*left->level + 2*top->level + tl->level + tr->level; - - if(s->keyframe){ - set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); - return; - } - - if(level!=s->block_max_depth){ - if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){ - put_rac(&s->c, &s->block_state[4 + s_context], 1); - }else{ - put_rac(&s->c, &s->block_state[4 + s_context], 0); - encode_q_branch2(s, level+1, 2*x+0, 2*y+0); - encode_q_branch2(s, level+1, 2*x+1, 2*y+0); - encode_q_branch2(s, level+1, 2*x+0, 2*y+1); - encode_q_branch2(s, level+1, 2*x+1, 2*y+1); - return; - } - } - if(b->type & BLOCK_INTRA){ - pred_mv(s, &pmx, &pmy, 0, left, top, tr); - put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1); - put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1); - put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1); - put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1); - set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA); - }else{ - pred_mv(s, &pmx, &pmy, b->ref, left, top, tr); - put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0); - if(s->ref_frames > 1) - put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0); - put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1); - put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1); - set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0); - } -} - -static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){ - int i, x2, y2; - Plane *p= &s->plane[plane_index]; - const int block_size = MB_SIZE >> s->block_max_depth; - const int block_w = plane_index ? block_size/2 : block_size; - const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; - const int obmc_stride= plane_index ? block_size : 2*block_size; - const int ref_stride= s->current_picture.linesize[plane_index]; - uint8_t *src= s-> input_picture.data[plane_index]; - IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned - const int b_stride = s->b_width << s->block_max_depth; - const int w= p->width; - const int h= p->height; - int index= mb_x + mb_y*b_stride; - BlockNode *b= &s->block[index]; - BlockNode backup= *b; - int ab=0; - int aa=0; - - b->type|= BLOCK_INTRA; - b->color[plane_index]= 0; - memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM)); - - for(i=0; i<4; i++){ - int mb_x2= mb_x + (i &1) - 1; - int mb_y2= mb_y + (i>>1) - 1; - int x= block_w*mb_x2 + block_w/2; - int y= block_w*mb_y2 + block_w/2; - - add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc, - x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index); - - for(y2= FFMAX(y, 0); y2h) obmc_v += obmc[index - block_w*obmc_stride]; - if(x+block_w>w) obmc_v += obmc[index - block_w]; - //FIXME precalculate this or simplify it somehow else - - d = -dst[index] + (1<<(FRAC_BITS-1)); - dst[index] = d; - ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v; - aa += obmc_v * obmc_v; //FIXME precalculate this - } - } - } - *b= backup; - - return av_clip(((ab<b_width << s->block_max_depth; - const int b_height = s->b_height<< s->block_max_depth; - int index= x + y*b_stride; - const BlockNode *b = &s->block[index]; - const BlockNode *left = x ? &s->block[index-1] : &null_block; - const BlockNode *top = y ? &s->block[index-b_stride] : &null_block; - const BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left; - const BlockNode *tr = y && x+wblock[index-b_stride+w] : tl; - int dmx, dmy; -// int mx_context= av_log2(2*FFABS(left->mx - top->mx)); -// int my_context= av_log2(2*FFABS(left->my - top->my)); - - if(x<0 || x>=b_stride || y>=b_height) - return 0; -/* -1 0 0 -01X 1-2 1 -001XX 3-6 2-3 -0001XXX 7-14 4-7 -00001XXXX 15-30 8-15 -*/ -//FIXME try accurate rate -//FIXME intra and inter predictors if surrounding blocks are not the same type - if(b->type & BLOCK_INTRA){ - return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0])) - + av_log2(2*FFABS(left->color[1] - b->color[1])) - + av_log2(2*FFABS(left->color[2] - b->color[2]))); - }else{ - pred_mv(s, &dmx, &dmy, b->ref, left, top, tr); - dmx-= b->mx; - dmy-= b->my; - return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda - + av_log2(2*FFABS(dmy)) - + av_log2(2*b->ref)); - } -} - -static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){ - Plane *p= &s->plane[plane_index]; - const int block_size = MB_SIZE >> s->block_max_depth; - const int block_w = plane_index ? block_size/2 : block_size; - const int obmc_stride= plane_index ? block_size : 2*block_size; - 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 = s->scratchbuf; - 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; - int rate= 0; - const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); - int sx= block_w*mb_x - block_w/2; - int sy= block_w*mb_y - block_w/2; - int x0= FFMAX(0,-sx); - int y0= FFMAX(0,-sy); - int x1= FFMIN(block_w*2, w-sx); - int y1= FFMIN(block_w*2, h-sy); - int i,x,y; - - pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h); - - for(y=y0; y= LOG2_OBMC_MAX - int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX); -#else - int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS); -#endif - v = (v + pred1[x]) >> FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst1[x] = v; - } - } - - /* copy the regions where obmc[] = (uint8_t)256 */ - if(LOG2_OBMC_MAX == 8 - && (mb_x == 0 || mb_x == b_stride-1) - && (mb_y == 0 || mb_y == b_height-1)){ - if(mb_x == 0) - x1 = block_w; - else - x0 = block_w; - if(mb_y == 0) - y1 = block_w; - else - y0 = block_w; - for(y=y0; yavctx->me_cmp == FF_CMP_W97) - distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); - else if(s->avctx->me_cmp == FF_CMP_W53) - distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); - else{ - distortion = 0; - for(i=0; i<4; i++){ - int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; - distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16); - } - } - }else{ - assert(block_w==8); - distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2); - } - - if(plane_index==0){ - for(i=0; i<4; i++){ -/* ..RRr - * .RXx. - * rxx.. - */ - rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1); - } - if(mb_x == b_stride-2) - rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1); - } - return distortion + rate*penalty_factor; -} - -static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){ - int i, y2; - Plane *p= &s->plane[plane_index]; - const int block_size = MB_SIZE >> s->block_max_depth; - const int block_w = plane_index ? block_size/2 : block_size; - const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; - const int obmc_stride= plane_index ? block_size : 2*block_size; - 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]; - //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst - // const has only been removed from zero_dst to suppress a warning - static IDWTELEM zero_dst[4096]; //FIXME - const int b_stride = s->b_width << s->block_max_depth; - const int w= p->width; - const int h= p->height; - int distortion= 0; - int rate= 0; - const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); - - for(i=0; i<9; i++){ - int mb_x2= mb_x + (i%3) - 1; - int mb_y2= mb_y + (i/3) - 1; - int x= block_w*mb_x2 + block_w/2; - int y= block_w*mb_y2 + block_w/2; - - add_yblock(s, 0, NULL, zero_dst, dst, obmc, - x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index); - - //FIXME find a cleaner/simpler way to skip the outside stuff - for(y2= y; y2<0; y2++) - memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); - for(y2= h; y2 w){ - for(y2= y; y2dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w); - } - - if(plane_index==0){ - BlockNode *b= &s->block[mb_x+mb_y*b_stride]; - int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1); - -/* ..RRRr - * .RXXx. - * .RXXx. - * rxxx. - */ - if(merged) - rate = get_block_bits(s, mb_x, mb_y, 2); - for(i=merged?4:0; i<9; i++){ - static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}}; - rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1); - } - } - return distortion + rate*penalty_factor; -} - -static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){ - const int w= b->width; - const int h= b->height; - int x, y; - - if(1){ - int run=0; - int runs[w*h]; - int run_index=0; - int max_index; - - for(y=0; y 1){ - if(orientation==1) ll= src[y + (x-2)*stride]; - else ll= src[x - 2 + y*stride]; - }*/ - } - if(parent){ - int px= x>>1; - int py= y>>1; - if(pxparent->width && pyparent->height) - p= parent[px + py*2*stride]; - } - if(!(/*ll|*/l|lt|t|rt|p)){ - if(v){ - runs[run_index++]= run; - run=0; - }else{ - run++; - } - } - } - } - max_index= run_index; - runs[run_index++]= run; - run_index=0; - run= runs[run_index++]; - - put_symbol2(&s->c, b->state[30], max_index, 0); - if(run_index <= max_index) - put_symbol2(&s->c, b->state[1], run, 3); - - for(y=0; yc.bytestream_end - s->c.bytestream < w*40){ - av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); - return -1; - } - for(x=0; x 1){ - if(orientation==1) ll= src[y + (x-2)*stride]; - else ll= src[x - 2 + y*stride]; - }*/ - } - if(parent){ - int px= x>>1; - int py= y>>1; - if(pxparent->width && pyparent->height) - p= parent[px + py*2*stride]; - } - if(/*ll|*/l|lt|t|rt|p){ - int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); - - put_rac(&s->c, &b->state[0][context], !!v); - }else{ - if(!run){ - run= runs[run_index++]; - - if(run_index <= max_index) - put_symbol2(&s->c, b->state[1], run, 3); - assert(v); - }else{ - run--; - assert(!v); - } - } - if(v){ - int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); - int l2= 2*FFABS(l) + (l<0); - int t2= 2*FFABS(t) + (t<0); - - put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4); - put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0); - } - } - } - } - return 0; -} - -static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){ -// encode_subband_qtree(s, b, src, parent, stride, orientation); -// encode_subband_z0run(s, b, src, parent, stride, orientation); - return encode_subband_c0run(s, b, src, parent, stride, orientation); -// encode_subband_dzr(s, b, src, parent, stride, orientation); -} - -static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){ - const int b_stride= s->b_width << s->block_max_depth; - BlockNode *block= &s->block[mb_x + mb_y * b_stride]; - BlockNode backup= *block; - int rd, index, value; - - assert(mb_x>=0 && mb_y>=0); - assert(mb_xcolor[0] = p[0]; - block->color[1] = p[1]; - block->color[2] = p[2]; - block->type |= BLOCK_INTRA; - }else{ - index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1); - value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12); - if(s->me_cache[index] == value) - return 0; - s->me_cache[index]= value; - - block->mx= p[0]; - block->my= p[1]; - block->type &= ~BLOCK_INTRA; - } - - rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged); - -//FIXME chroma - if(rd < *best_rd){ - *best_rd= rd; - return 1; - }else{ - *block= backup; - return 0; - } -} - -/* special case for int[2] args we discard afterwards, - * fixes compilation problem with gcc 2.95 */ -static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){ - int p[2] = {p0, p1}; - return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd); -} - -static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){ - const int b_stride= s->b_width << s->block_max_depth; - BlockNode *block= &s->block[mb_x + mb_y * b_stride]; - BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]}; - int rd, index, value; - - assert(mb_x>=0 && mb_y>=0); - assert(mb_xme_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12); - if(s->me_cache[index] == value) - return 0; - s->me_cache[index]= value; - - block->mx= p0; - block->my= p1; - block->ref= ref; - block->type &= ~BLOCK_INTRA; - block[1]= block[b_stride]= block[b_stride+1]= *block; - - rd= get_4block_rd(s, mb_x, mb_y, 0); - -//FIXME chroma - if(rd < *best_rd){ - *best_rd= rd; - return 1; - }else{ - block[0]= backup[0]; - block[1]= backup[1]; - block[b_stride]= backup[2]; - block[b_stride+1]= backup[3]; - return 0; - } -} - -static void iterative_me(SnowContext *s){ - int pass, mb_x, mb_y; - const int b_width = s->b_width << s->block_max_depth; - const int b_height= s->b_height << s->block_max_depth; - const int b_stride= b_width; - int color[3]; - - { - RangeCoder r = s->c; - uint8_t state[sizeof(s->block_state)]; - memcpy(state, s->block_state, sizeof(s->block_state)); - for(mb_y= 0; mb_yb_height; mb_y++) - for(mb_x= 0; mb_xb_width; mb_x++) - encode_q_branch(s, 0, mb_x, mb_y); - s->c = r; - memcpy(s->block_state, state, sizeof(s->block_state)); - } - - for(pass=0; pass<25; pass++){ - int change= 0; - - for(mb_y= 0; mb_yblock[index]; - BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL; - BlockNode *lb = mb_x ? &s->block[index -1] : NULL; - BlockNode *rb = mb_x+1block[index +1] : NULL; - BlockNode *bb = mb_y+1block[index+b_stride ] : NULL; - BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL; - BlockNode *trb= mb_x+1block[index-b_stride+1] : NULL; - BlockNode *blb= mb_x && mb_y+1block[index+b_stride-1] : NULL; - BlockNode *brb= mb_x+1block[index+b_stride+1] : NULL; - const int b_w= (MB_SIZE >> s->block_max_depth); - uint8_t obmc_edged[b_w*2][b_w*2]; - - if(pass && (block->type & BLOCK_OPT)) - continue; - block->type |= BLOCK_OPT; - - backup= *block; - - if(!s->me_cache_generation) - memset(s->me_cache, 0, sizeof(s->me_cache)); - s->me_cache_generation += 1<<22; - - //FIXME precalculate - { - int x, y; - memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4); - if(mb_x==0) - for(y=0; y input_picture.data[0]; - uint8_t *dst= s->current_picture.data[0]; - const int stride= s->current_picture.linesize[0]; - const int block_w= MB_SIZE >> s->block_max_depth; - const int sx= block_w*mb_x - block_w/2; - const int sy= block_w*mb_y - block_w/2; - const int w= s->plane[0].width; - const int h= s->plane[0].height; - int y; - - for(y=sy; y<0; y++) - memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); - for(y=h; y w){ - for(y=sy; y 0 && (block->type&BLOCK_INTRA)){ - int color0[3]= {block->color[0], block->color[1], block->color[2]}; - check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd); - }else - check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd); - - ref_b= *block; - ref_rd= best_rd; - for(ref=0; ref < s->ref_frames; ref++){ - int16_t (*mvr)[2]= &s->ref_mvs[ref][index]; - if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold - continue; - block->ref= ref; - best_rd= INT_MAX; - - check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd); - check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd); - if(tb) - check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd); - if(lb) - check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd); - if(rb) - check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd); - if(bb) - check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd); - - /* fullpel ME */ - //FIXME avoid subpel interpolation / round to nearest integer - do{ - dia_change=0; - for(i=0; iavctx->dia_size, 1); i++){ - for(j=0; jmx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); - dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); - dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); - dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); - } - } - }while(dia_change); - /* subpel ME */ - do{ - static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},}; - dia_change=0; - for(i=0; i<8; i++) - dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd); - }while(dia_change); - //FIXME or try the standard 2 pass qpel or similar - - mvr[0][0]= block->mx; - mvr[0][1]= block->my; - if(ref_rd > best_rd){ - ref_rd= best_rd; - ref_b= *block; - } - } - best_rd= ref_rd; - *block= ref_b; - check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd); - //FIXME RD style color selection - if(!same_block(block, &backup)){ - if(tb ) tb ->type &= ~BLOCK_OPT; - if(lb ) lb ->type &= ~BLOCK_OPT; - if(rb ) rb ->type &= ~BLOCK_OPT; - if(bb ) bb ->type &= ~BLOCK_OPT; - if(tlb) tlb->type &= ~BLOCK_OPT; - if(trb) trb->type &= ~BLOCK_OPT; - if(blb) blb->type &= ~BLOCK_OPT; - if(brb) brb->type &= ~BLOCK_OPT; - change ++; - } - } - } - av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change); - if(!change) - break; - } - - if(s->block_max_depth == 1){ - int change= 0; - for(mb_y= 0; mb_yblock[index]; - b[1]= b[0]+1; - b[2]= b[0]+b_stride; - b[3]= b[2]+1; - if(same_block(b[0], b[1]) && - same_block(b[0], b[2]) && - same_block(b[0], b[3])) - continue; - - if(!s->me_cache_generation) - memset(s->me_cache, 0, sizeof(s->me_cache)); - s->me_cache_generation += 1<<22; - - init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0); - - //FIXME more multiref search? - check_4block_inter(s, mb_x, mb_y, - (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2, - (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd); - - for(i=0; i<4; i++) - if(!(b[i]->type&BLOCK_INTRA)) - check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd); - - if(init_rd != best_rd) - change++; - } - } - av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4); - } -} - -static void encode_blocks(SnowContext *s, int search){ - int x, y; - int w= s->b_width; - int h= s->b_height; - - if(s->avctx->me_method == ME_ITER && !s->keyframe && search) - iterative_me(s); - - for(y=0; yc.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit - av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); - return; - } - for(x=0; xavctx->me_method == ME_ITER || !search) - encode_q_branch2(s, 0, x, y); - else - encode_q_branch (s, 0, x, y); - } - } -} - -static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){ - const int w= b->width; - const int h= b->height; - const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); - const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS); - int x,y, thres1, thres2; - - if(s->qlog == LOSSLESS_QLOG){ - for(y=0; y>3; - thres1= ((qmul - bias)>>QEXPSHIFT) - 1; - thres2= 2*thres1; - - if(!bias){ - for(y=0; y thres2){ - if(i>=0){ - i<<= QEXPSHIFT; - i/= qmul; //FIXME optimize - dst[x + y*stride]= i; - }else{ - i= -i; - i<<= QEXPSHIFT; - i/= qmul; //FIXME optimize - dst[x + y*stride]= -i; - } - }else - dst[x + y*stride]= 0; - } - } - }else{ - for(y=0; y thres2){ - if(i>=0){ - i<<= QEXPSHIFT; - i= (i + bias) / qmul; //FIXME optimize - dst[x + y*stride]= i; - }else{ - i= -i; - i<<= QEXPSHIFT; - i= (i + bias) / qmul; //FIXME optimize - dst[x + y*stride]= -i; - } - }else - dst[x + y*stride]= 0; - } - } - } -} - -static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){ - const int w= b->width; - const int h= b->height; - const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); - const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); - const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; - int x,y; - - if(s->qlog == LOSSLESS_QLOG) return; - - for(y=0; y>(QEXPSHIFT)); //FIXME try different bias - }else if(i>0){ - src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT)); - } - } - } -} - -static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){ - const int w= b->width; - const int h= b->height; - int x,y; - - for(y=h-1; y>=0; y--){ - for(x=w-1; x>=0; x--){ - int i= x + y*stride; - - if(x){ - if(use_median){ - if(y && x+1width; - const int h= b->height; - int x,y; - - for(y=0; yspatial_decomposition_count; level++){ - for(orientation=level ? 1:0; orientation<4; orientation++){ - if(orientation==2) continue; - put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1); - } - } - } -} - -static void encode_header(SnowContext *s){ - int plane_index, i; - uint8_t kstate[32]; - - memset(kstate, MID_STATE, sizeof(kstate)); - - put_rac(&s->c, kstate, s->keyframe); - if(s->keyframe || s->always_reset){ - reset_contexts(s); - s->last_spatial_decomposition_type= - s->last_qlog= - s->last_qbias= - s->last_mv_scale= - s->last_block_max_depth= 0; - for(plane_index=0; plane_index<2; plane_index++){ - Plane *p= &s->plane[plane_index]; - p->last_htaps=0; - p->last_diag_mc=0; - memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff)); - } - } - if(s->keyframe){ - put_symbol(&s->c, s->header_state, s->version, 0); - put_rac(&s->c, s->header_state, s->always_reset); - put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0); - put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0); - put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0); - put_symbol(&s->c, s->header_state, s->colorspace_type, 0); - put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0); - put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0); - put_rac(&s->c, s->header_state, s->spatial_scalability); -// put_rac(&s->c, s->header_state, s->rate_scalability); - put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0); - - encode_qlogs(s); - } - - if(!s->keyframe){ - int update_mc=0; - for(plane_index=0; plane_index<2; plane_index++){ - Plane *p= &s->plane[plane_index]; - update_mc |= p->last_htaps != p->htaps; - update_mc |= p->last_diag_mc != p->diag_mc; - update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff)); - } - put_rac(&s->c, s->header_state, update_mc); - if(update_mc){ - for(plane_index=0; plane_index<2; plane_index++){ - Plane *p= &s->plane[plane_index]; - put_rac(&s->c, s->header_state, p->diag_mc); - put_symbol(&s->c, s->header_state, p->htaps/2-1, 0); - for(i= p->htaps/2; i; i--) - put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0); - } - } - if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){ - put_rac(&s->c, s->header_state, 1); - put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0); - encode_qlogs(s); - }else - put_rac(&s->c, s->header_state, 0); - } - - put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1); - put_symbol(&s->c, s->header_state, s->qlog - s->last_qlog , 1); - put_symbol(&s->c, s->header_state, s->mv_scale - s->last_mv_scale, 1); - put_symbol(&s->c, s->header_state, s->qbias - s->last_qbias , 1); - put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1); - -} - -static void update_last_header_values(SnowContext *s){ - int plane_index; - - if(!s->keyframe){ - for(plane_index=0; plane_index<2; plane_index++){ - Plane *p= &s->plane[plane_index]; - p->last_diag_mc= p->diag_mc; - p->last_htaps = p->htaps; - memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff)); - } - } - - s->last_spatial_decomposition_type = s->spatial_decomposition_type; - s->last_qlog = s->qlog; - s->last_qbias = s->qbias; - s->last_mv_scale = s->mv_scale; - s->last_block_max_depth = s->block_max_depth; - s->last_spatial_decomposition_count = s->spatial_decomposition_count; -} - -static int qscale2qlog(int qscale){ - return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2)) - + 61*QROOT/8; ///< 64 > 60 -} - -static int ratecontrol_1pass(SnowContext *s, AVFrame *pict) -{ - /* Estimate the frame's complexity as a sum of weighted dwt coefficients. - * FIXME we know exact mv bits at this point, - * but ratecontrol isn't set up to include them. */ - uint32_t coef_sum= 0; - int level, orientation, delta_qlog; - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &s->plane[0].band[level][orientation]; - IDWTELEM *buf= b->ibuf; - const int w= b->width; - const int h= b->height; - const int stride= b->stride; - const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16); - const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); - const int qdiv= (1<<16)/qmul; - int x, y; - //FIXME this is ugly - for(y=0; ybuf[x+y*stride]; - if(orientation==0) - decorrelate(s, b, buf, stride, 1, 0); - for(y=0; y> 16; - } - } - - /* ugly, ratecontrol just takes a sqrt again */ - coef_sum = (uint64_t)coef_sum * coef_sum >> 16; - assert(coef_sum < INT_MAX); - - if(pict->pict_type == AV_PICTURE_TYPE_I){ - s->m.current_picture.mb_var_sum= coef_sum; - s->m.current_picture.mc_mb_var_sum= 0; - }else{ - s->m.current_picture.mc_mb_var_sum= coef_sum; - s->m.current_picture.mb_var_sum= 0; - } - - pict->quality= ff_rate_estimate_qscale(&s->m, 1); - if (pict->quality < 0) - return INT_MIN; - s->lambda= pict->quality * 3/2; - delta_qlog= qscale2qlog(pict->quality) - s->qlog; - s->qlog+= delta_qlog; - return delta_qlog; -} - -static void calculate_visual_weight(SnowContext *s, Plane *p){ - int width = p->width; - int height= p->height; - int level, orientation, x, y; - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - IDWTELEM *ibuf= b->ibuf; - int64_t error=0; - - memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height); - ibuf[b->width/2 + b->height/2*b->stride]= 256*16; - ff_spatial_idwt(s->spatial_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); - for(y=0; yspatial_idwt_buffer[x + y*width]*16; - error += d*d; - } - } - - b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5); - } - } -} - -static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){ - SnowContext *s = avctx->priv_data; - RangeCoder * const c= &s->c; - AVFrame *pict = data; - const int width= s->avctx->width; - const int height= s->avctx->height; - int level, orientation, plane_index, i, y; - uint8_t rc_header_bak[sizeof(s->header_state)]; - uint8_t rc_block_bak[sizeof(s->block_state)]; - - ff_init_range_encoder(c, buf, buf_size); - ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); - - for(i=0; i<3; i++){ - int shift= !!i; - for(y=0; y<(height>>shift); y++) - memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]], - &pict->data[i][y * pict->linesize[i]], - width>>shift); - } - s->new_picture = *pict; - - s->m.picture_number= avctx->frame_number; - if(avctx->flags&CODEC_FLAG_PASS2){ - s->m.pict_type = - pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type; - s->keyframe= pict->pict_type==AV_PICTURE_TYPE_I; - if(!(avctx->flags&CODEC_FLAG_QSCALE)) { - pict->quality= ff_rate_estimate_qscale(&s->m, 0); - if (pict->quality < 0) - return -1; - } - }else{ - s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; - s->m.pict_type= - pict->pict_type= s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P; - } - - if(s->pass1_rc && avctx->frame_number == 0) - pict->quality= 2*FF_QP2LAMBDA; - if(pict->quality){ - s->qlog= qscale2qlog(pict->quality); - s->lambda = pict->quality * 3/2; - } - if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){ - s->qlog= LOSSLESS_QLOG; - s->lambda = 0; - }//else keep previous frame's qlog until after motion estimation - - frame_start(s); - - s->m.current_picture_ptr= &s->m.current_picture; - s->m.last_picture.f.pts = s->m.current_picture.f.pts; - s->m.current_picture.f.pts = pict->pts; - if(pict->pict_type == AV_PICTURE_TYPE_P){ - int block_width = (width +15)>>4; - int block_height= (height+15)>>4; - int stride= s->current_picture.linesize[0]; - - assert(s->current_picture.data[0]); - assert(s->last_picture[0].data[0]); - - s->m.avctx= s->avctx; - s->m.current_picture.f.data[0] = s->current_picture.data[0]; - s->m. last_picture.f.data[0] = s->last_picture[0].data[0]; - s->m. new_picture.f.data[0] = s-> input_picture.data[0]; - s->m. last_picture_ptr= &s->m. last_picture; - s->m.linesize= - s->m. last_picture.f.linesize[0] = - s->m. new_picture.f.linesize[0] = - s->m.current_picture.f.linesize[0] = stride; - s->m.uvlinesize= s->current_picture.linesize[1]; - s->m.width = width; - s->m.height= height; - s->m.mb_width = block_width; - s->m.mb_height= block_height; - s->m.mb_stride= s->m.mb_width+1; - s->m.b8_stride= 2*s->m.mb_width+1; - s->m.f_code=1; - s->m.pict_type= pict->pict_type; - s->m.me_method= s->avctx->me_method; - s->m.me.scene_change_score=0; - s->m.flags= s->avctx->flags; - s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0; - s->m.out_format= FMT_H263; - s->m.unrestricted_mv= 1; - - s->m.lambda = s->lambda; - s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); - s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; - - s->m.dsp= s->dsp; //move - ff_init_me(&s->m); - s->dsp= s->m.dsp; - } - - if(s->pass1_rc){ - memcpy(rc_header_bak, s->header_state, sizeof(s->header_state)); - memcpy(rc_block_bak, s->block_state, sizeof(s->block_state)); - } - -redo_frame: - - if(pict->pict_type == AV_PICTURE_TYPE_I) - s->spatial_decomposition_count= 5; - else - s->spatial_decomposition_count= 5; - - s->m.pict_type = pict->pict_type; - s->qbias= pict->pict_type == AV_PICTURE_TYPE_P ? 2 : 0; - - common_init_after_header(avctx); - - if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){ - for(plane_index=0; plane_index<3; plane_index++){ - calculate_visual_weight(s, &s->plane[plane_index]); - } - } - - encode_header(s); - s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start); - encode_blocks(s, 1); - s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits; - - for(plane_index=0; plane_index<3; plane_index++){ - Plane *p= &s->plane[plane_index]; - int w= p->width; - int h= p->height; - int x, y; -// int bits= put_bits_count(&s->c.pb); - - if (!s->memc_only) { - //FIXME optimize - if(pict->data[plane_index]) //FIXME gray hack - for(y=0; yspatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<spatial_idwt_buffer, plane_index, 0); - - if( plane_index==0 - && pict->pict_type == AV_PICTURE_TYPE_P - && !(avctx->flags&CODEC_FLAG_PASS2) - && s->m.me.scene_change_score > s->avctx->scenechange_threshold){ - ff_init_range_encoder(c, buf, buf_size); - ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); - pict->pict_type= AV_PICTURE_TYPE_I; - s->keyframe=1; - s->current_picture.key_frame=1; - goto redo_frame; - } - - if(s->qlog == LOSSLESS_QLOG){ - for(y=0; yspatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS; - } - } - }else{ - for(y=0; yspatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type); - else*/ - ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); - - if(s->pass1_rc && plane_index==0){ - int delta_qlog = ratecontrol_1pass(s, pict); - if (delta_qlog <= INT_MIN) - return -1; - if(delta_qlog){ - //reordering qlog in the bitstream would eliminate this reset - ff_init_range_encoder(c, buf, buf_size); - memcpy(s->header_state, rc_header_bak, sizeof(s->header_state)); - memcpy(s->block_state, rc_block_bak, sizeof(s->block_state)); - encode_header(s); - encode_blocks(s, 0); - } - } - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - - if(!QUANTIZE2) - quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias); - if(orientation==0) - decorrelate(s, b, b->ibuf, b->stride, pict->pict_type == AV_PICTURE_TYPE_P, 0); - encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation); - assert(b->parent==NULL || b->parent->stride == b->stride*2); - if(orientation==0) - correlate(s, b, b->ibuf, b->stride, 1, 0); - } - } - - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1 : 0; orientation<4; orientation++){ - SubBand *b= &p->band[level][orientation]; - - dequantize(s, b, b->ibuf, b->stride); - } - } - - ff_spatial_idwt(s->spatial_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); - if(s->qlog == LOSSLESS_QLOG){ - for(y=0; yspatial_idwt_buffer[y*w + x]<<=FRAC_BITS; - } - } - } - predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); - }else{ - //ME/MC only - if(pict->pict_type == AV_PICTURE_TYPE_I){ - for(y=0; ycurrent_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]= - pict->data[plane_index][y*pict->linesize[plane_index] + x]; - } - } - }else{ - memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h); - predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); - } - } - if(s->avctx->flags&CODEC_FLAG_PSNR){ - int64_t error= 0; - - if(pict->data[plane_index]) //FIXME gray hack - for(y=0; ycurrent_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x]; - error += d*d; - } - } - s->avctx->error[plane_index] += error; - s->current_picture.error[plane_index] = error; - } - - } - - update_last_header_values(s); - - release_buffer(avctx); - - s->current_picture.coded_picture_number = avctx->frame_number; - s->current_picture.pict_type = pict->pict_type; - s->current_picture.quality = pict->quality; - s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start); - s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits; - s->m.current_picture.f.display_picture_number = - s->m.current_picture.f.coded_picture_number = avctx->frame_number; - s->m.current_picture.f.quality = pict->quality; - s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start); - if(s->pass1_rc) - if (ff_rate_estimate_qscale(&s->m, 0) < 0) - return -1; - if(avctx->flags&CODEC_FLAG_PASS1) - ff_write_pass1_stats(&s->m); - s->m.last_pict_type = s->m.pict_type; - avctx->frame_bits = s->m.frame_bits; - avctx->mv_bits = s->m.mv_bits; - avctx->misc_bits = s->m.misc_bits; - avctx->p_tex_bits = s->m.p_tex_bits; - - emms_c(); - - return ff_rac_terminate(c); -} - -static av_cold int encode_end(AVCodecContext *avctx) -{ - SnowContext *s = avctx->priv_data; - - common_end(s); - if (s->input_picture.data[0]) - avctx->release_buffer(avctx, &s->input_picture); - av_free(avctx->stats_out); - - return 0; -} - -#define OFFSET(x) offsetof(SnowContext, x) -#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM -static const AVOption options[] = { - { "memc_only", "Only do ME/MC (I frames -> ref, P frame -> ME+MC).", OFFSET(memc_only), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE }, - { NULL }, -}; - -static const AVClass snowenc_class = { - .class_name = "snow encoder", - .item_name = av_default_item_name, - .option = options, - .version = LIBAVUTIL_VERSION_INT, -}; - -AVCodec ff_snow_encoder = { - .name = "snow", - .type = AVMEDIA_TYPE_VIDEO, - .id = CODEC_ID_SNOW, - .priv_data_size = sizeof(SnowContext), - .init = encode_init, - .encode = encode_frame, - .close = encode_end, - .long_name = NULL_IF_CONFIG_SMALL("Snow"), - .priv_class = &snowenc_class, -}; -#endif - - -#ifdef TEST -#undef malloc -#undef free -#undef printf - -#include "libavutil/lfg.h" -#include "libavutil/mathematics.h" - -int main(void){ - int width=256; - int height=256; - int buffer[2][width*height]; - SnowContext s; - int i; - AVLFG prng; - s.spatial_decomposition_count=6; - s.spatial_decomposition_type=1; - - av_lfg_init(&prng, 1); - - printf("testing 5/3 DWT\n"); - for(i=0; i20) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]); - - { - int level, orientation, x, y; - int64_t errors[8][4]; - int64_t g=0; - - memset(errors, 0, sizeof(errors)); - s.spatial_decomposition_count=3; - s.spatial_decomposition_type=0; - for(level=0; level> (s.spatial_decomposition_count-level); - int h= height >> (s.spatial_decomposition_count-level); - int stride= width << (s.spatial_decomposition_count-level); - DWTELEM *buf= buffer[0]; - int64_t error=0; - - if(orientation&1) buf+=w; - if(orientation>1) buf+=stride>>1; - - memset(buffer[0], 0, sizeof(int)*width*height); - buf[w/2 + h/2*stride]= 256*256; - ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); - for(y=0; y> (s.spatial_decomposition_count-level); - //int h= height >> (s.spatial_decomposition_count-level); - int stride= width << (s.spatial_decomposition_count-level); - DWTELEM *buf= buffer[0]; - int64_t error=0; - - buf+=w; - buf+=stride>>1; - - memset(buffer[0], 0, sizeof(int)*width*height); - for(y=0; y