22 #ifndef AVCODEC_SNOW_H
23 #define AVCODEC_SNOW_H
36 #define QROOT (1<<QSHIFT)
37 #define LOSSLESS_QLOG -128
39 #define MAX_REF_FRAMES 8
41 #define LOG2_OBMC_MAX 8
42 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
57 .
color= {128,128,128},
65 #define LOG2_MB_SIZE 4
66 #define MB_SIZE (1<<LOG2_MB_SIZE)
67 #define ENCODER_EXTRA_BITS 4
153 #define QBIAS_SHIFT 3
160 #define ME_CACHE_SIZE 1024
183 low[(*i)+1] = low[((*i)+1)>>1];
189 for (; (*i)>=0; (*i)-=2){
190 low[(*i)+1] = high[(*i)>>1];
191 low[*i] = low[(*i)>>1];
197 dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
200 if((width^lift_high)&1){
201 dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
207 dst[i] = src[i] + ((ref[i] + ref[(i+1)]+
W_BO + 4 * src[i]) >>
W_BS);
211 dst[w] = src[w] + ((2 * ref[w] +
W_BO + 4 * src[w]) >>
W_BS);
226 int plane_index,
int w,
int h);
230 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){
233 const int index= (x + y*w) << rem_depth;
234 const int block_w= 1<<rem_depth;
247 for(j=0; j<block_w; j++){
248 for(i=0; i<block_w; i++){
262 (top ->
mx * scale[top ->
ref] + 128) >>8,
263 (tr ->
mx * scale[tr ->
ref] + 128) >>8);
265 (top ->
my * scale[top ->
ref] + 128) >>8,
266 (tr ->
my * scale[tr ->
ref] + 128) >>8);
280 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){
283 const int b_stride= b_width;
297 }
else if(b_x + 1 >= b_width){
304 }
else if(b_y + 1 >= b_height){
312 if(!sliced && !offset_dst)
315 }
else if(src_x + b_w > w){
319 obmc -= src_y*obmc_stride;
321 if(!sliced && !offset_dst)
322 dst -= src_y*dst_stride;
324 }
else if(src_y + b_h> h){
328 if(b_w<=0 || b_h<=0)
return;
330 assert(src_stride > 2*
MB_SIZE + 5);
332 if(!sliced && offset_dst)
333 dst += src_x + src_y*dst_stride;
334 dst8+= src_x + src_y*src_stride;
337 ptmp= tmp + 3*tmp_step;
340 ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
347 ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
357 ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
368 ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
371 s->
dwt.
inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
373 for(y=0; y<b_h; y++){
375 const uint8_t *obmc1= obmc + y*obmc_stride;
376 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
377 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
379 for(x=0; x<b_w; x++){
380 int v= obmc1[x] * block[3][x + y*src_stride]
381 +obmc2[x] * block[2][x + y*src_stride]
382 +obmc3[x] * block[1][x + y*src_stride]
383 +obmc4[x] * block[0][x + y*src_stride];
390 v += dst[x + y*dst_stride];
392 if(v&(~255)) v= ~(v>>31);
393 dst8[x + y*src_stride] = v;
395 dst[x + y*dst_stride] -= v;
408 int block_w = plane_index ? block_size/2 : block_size;
410 const int obmc_stride= plane_index ? block_size : 2*block_size;
421 for(y=block_w*mb_y; y<
FFMIN(h,block_w*(mb_y+1)); y++){
425 if(v&(~255)) v= ~(v>>31);
426 dst8[x + y*ref_stride]= v;
430 for(y=block_w*mb_y; y<
FFMIN(h,block_w*(mb_y+1)); y++){
440 for(mb_x=0; mb_x<=mb_w; mb_x++){
442 block_w*mb_x - block_w/2,
443 block_w*mb_y - block_w/2,
446 w, ref_stride, obmc_stride,
448 add, 1, plane_index);
455 for(mb_y=0; mb_y<=mb_h; mb_y++)
459 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){
462 const int index= (x + y*w) << rem_depth;
463 const int block_w= 1<<rem_depth;
476 for(j=0; j<block_w; j++){
477 for(i=0; i<block_w; i++){
484 const int offset[3]= {
491 c->
src[0][i]= src [i];
492 c->
ref[0][i]= ref [i] + offset[i];
502 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
508 const int a=
FFABS(v);
510 const int el=
FFMIN(e, 10);
521 for(i=e-1; i>=el; i--){
522 put_rac(c, state+22+9, (a>>i)&1);
525 put_rac(c, state+22+i, (a>>i)&1);
529 put_rac(c, state+11 + el, v < 0);
546 for(i=e-1; i>=0; i--){
557 int r= log2>=0 ? 1<<log2 : 1;
570 for(i=log2-1; i>=0; i--){
571 put_rac(c, state+31-i, (v>>i)&1);
577 int r= log2>=0 ? 1<<log2 : 1;
582 while(
get_rac(c, state+4+log2)){
588 for(i=log2-1; i>=0; i--){
596 const int w= b->
width;
615 if(y && prev_xc->
x == 0){
627 if(prev_xc->
x == x + 1)
633 if(x>>1 > parent_xc->
x){
636 if(x>>1 == parent_xc->
x){
641 int context=
av_log2(3*(l>>1) + (lt>>1) + (
t&~1) + (rt>>1) + (p>>1));
665 if(y) max_run=
FFMIN(run, prev_xc->
x - x - 2);
666 else max_run=
FFMIN(run, w-x-1);
668 max_run=
FFMIN(max_run, 2*parent_xc->
x - x - 1);
680 while(parent_xc->
x != parent->
width+1)
683 prev_parent_xc= parent_xc;
685 parent_xc= prev_parent_xc;