22 #ifndef AVCODEC_SNOW_H
23 #define AVCODEC_SNOW_H
39 #define QROOT (1<<QSHIFT)
40 #define LOSSLESS_QLOG -128
42 #define MAX_REF_FRAMES 8
44 #define LOG2_OBMC_MAX 8
45 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
60 .
color= {128,128,128},
68 #define LOG2_MB_SIZE 4
69 #define MB_SIZE (1<<LOG2_MB_SIZE)
70 #define ENCODER_EXTRA_BITS 4
160 #define QBIAS_SHIFT 3
168 #define ME_CACHE_SIZE 1024
192 low[(*i)+1] = low[((*i)+1)>>1];
198 for (; (*i)>=0; (*i)-=2){
199 low[(*i)+1] = high[(*i)>>1];
200 low[*i] = low[(*i)>>1];
206 dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >>
shift);
209 if((width^lift_high)&1){
210 dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
216 dst[i] = src[i] + ((ref[i] + ref[(i+1)]+
W_BO + 4 * src[i]) >>
W_BS);
220 dst[w] = src[w] + ((2 * ref[w] +
W_BO + 4 * src[w]) >>
W_BS);
235 int plane_index,
int w,
int h);
240 static inline void snow_set_blocks(
SnowContext *
s,
int level,
int x,
int y,
int l,
int cb,
int cr,
int mx,
int my,
int ref,
int type){
243 const int index= (x + y*w) << rem_depth;
244 const int block_w= 1<<rem_depth;
257 for(j=0; j<block_w; j++){
258 for(i=0; i<block_w; i++){
272 (top ->
mx * scale[top ->
ref] + 128) >>8,
273 (tr ->
mx * scale[tr ->
ref] + 128) >>8);
275 (top ->
my * scale[top ->
ref] + 128) >>8,
276 (tr ->
my * scale[tr ->
ref] + 128) >>8);
290 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){
293 const int b_stride= b_width;
307 }
else if(b_x + 1 >= b_width){
314 }
else if(b_y + 1 >= b_height){
322 if(!sliced && !offset_dst)
330 obmc -= src_y*obmc_stride;
332 if(!sliced && !offset_dst)
333 dst -= src_y*dst_stride;
340 if(b_w<=0 || b_h<=0)
return;
344 if(!sliced && offset_dst)
345 dst += src_x + src_y*dst_stride;
346 dst8+= src_x + src_y*src_stride;
349 ptmp= tmp + 3*tmp_step;
352 ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
359 ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
369 ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
380 ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
383 s->
dwt.
inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
385 for(y=0; y<b_h; y++){
387 const uint8_t *obmc1= obmc + y*obmc_stride;
388 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
389 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
391 for(x=0; x<b_w; x++){
392 int v= obmc1[x] * block[3][x + y*src_stride]
393 +obmc2[x] * block[2][x + y*src_stride]
394 +obmc3[x] * block[1][x + y*src_stride]
395 +obmc4[x] * block[0][x + y*src_stride];
402 v += dst[x + y*dst_stride];
404 if(v&(~255)) v= ~(v>>31);
405 dst8[x + y*src_stride] =
v;
407 dst[x + y*dst_stride] -=
v;
420 int block_w = plane_index ? block_size>>s->
chroma_h_shift : block_size;
421 int block_h = plane_index ? block_size>>s->
chroma_v_shift : block_size;
423 const int obmc_stride= plane_index ? (2*block_size)>>s->
chroma_h_shift : 2*block_size;
434 for(y=block_h*mb_y; y<
FFMIN(h,block_h*(mb_y+1)); y++){
438 if(v&(~255)) v= ~(v>>31);
439 dst8[x + y*ref_stride]=
v;
443 for(y=block_h*mb_y; y<
FFMIN(h,block_h*(mb_y+1)); y++){
453 for(mb_x=0; mb_x<=mb_w; mb_x++){
455 block_w*mb_x - block_w/2,
456 block_h*mb_y - block_h/2,
459 w, ref_stride, obmc_stride,
461 add, 1, plane_index);
468 for(mb_y=0; mb_y<=mb_h; mb_y++)
472 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){
475 const int index= (x + y*w) << rem_depth;
476 const int block_w= 1<<rem_depth;
477 const int block_h= 1<<rem_depth;
490 for(j=0; j<block_h; j++){
491 for(i=0; i<block_w; i++){
506 c->
src[0][i]= src [i];
507 c->
ref[0][i]= ref [i] + offset[i];
517 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
525 const int el=
FFMIN(e, 10);
536 for(i=e-1; i>=el; i--){
537 put_rac(c, state+22+9, (a>>i)&1);
540 put_rac(c, state+22+i, (a>>i)&1);
544 put_rac(c, state+11 + el, v < 0);
561 for(i=e-1; i>=0; i--){
572 int r= log2>=0 ? 1<<log2 : 1;
585 for(i=log2-1; i>=0; i--){
586 put_rac(c, state+31-i, (v>>i)&1);
592 int r= log2>=0 ? 1<<log2 : 1;
597 while(log2<28 &&
get_rac(c, state+4+log2)){
603 for(i=log2-1; i>=0; i--){
611 const int w= b->
width;
630 if(y && prev_xc->
x == 0){
642 if(prev_xc->
x == x + 1)
648 if(x>>1 > parent_xc->
x){
651 if(x>>1 == parent_xc->
x){
656 int context=
av_log2(3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
680 if(y) max_run=
FFMIN(run, prev_xc->
x - x - 2);
681 else max_run=
FFMIN(run, w-x-1);
683 max_run=
FFMIN(max_run, 2*parent_xc->
x - x - 1);
697 while(parent_xc->
x != parent->
width+1)
700 prev_parent_xc= parent_xc;
702 parent_xc= prev_parent_xc;