2 * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (C) 2008 David Conrad
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/attributes.h"
25 #include "libavcodec/x86/dwt.h"
27 void ff_slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
31 buf->base_buffer = base_buffer;
32 buf->line_count = line_count;
33 buf->line_width = line_width;
34 buf->data_count = max_allocated_lines;
35 buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count);
36 buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines);
38 for(i = 0; i < max_allocated_lines; i++){
39 buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width);
42 buf->data_stack_top = max_allocated_lines - 1;
45 IDWTELEM * ff_slice_buffer_load_line(slice_buffer * buf, int line)
49 assert(buf->data_stack_top >= 0);
50 // assert(!buf->line[line]);
52 return buf->line[line];
54 buffer = buf->data_stack[buf->data_stack_top];
55 buf->data_stack_top--;
56 buf->line[line] = buffer;
61 void ff_slice_buffer_release(slice_buffer * buf, int line)
65 assert(line >= 0 && line < buf->line_count);
66 assert(buf->line[line]);
68 buffer = buf->line[line];
69 buf->data_stack_top++;
70 buf->data_stack[buf->data_stack_top] = buffer;
71 buf->line[line] = NULL;
74 void ff_slice_buffer_flush(slice_buffer * buf)
77 for(i = 0; i < buf->line_count; i++){
79 ff_slice_buffer_release(buf, i);
83 void ff_slice_buffer_destroy(slice_buffer * buf)
86 ff_slice_buffer_flush(buf);
88 for(i = buf->data_count - 1; i >= 0; i--){
89 av_freep(&buf->data_stack[i]);
91 av_freep(&buf->data_stack);
95 static inline int mirror(int v, int m){
96 while((unsigned)v > (unsigned)m){
103 static av_always_inline void
104 lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
105 int dst_step, int src_step, int ref_step,
106 int width, int mul, int add, int shift,
107 int highpass, int inverse){
108 const int mirror_left= !highpass;
109 const int mirror_right= (width&1) ^ highpass;
110 const int w= (width>>1) - 1 + (highpass & width);
113 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
115 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
122 LIFT(src[i*src_step],
123 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
129 LIFT(src[w*src_step],
130 ((mul*2*ref[w*ref_step]+add)>>shift),
135 static av_always_inline void
136 inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
137 int dst_step, int src_step, int ref_step,
138 int width, int mul, int add, int shift,
139 int highpass, int inverse){
140 const int mirror_left= !highpass;
141 const int mirror_right= (width&1) ^ highpass;
142 const int w= (width>>1) - 1 + (highpass & width);
145 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
147 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
154 LIFT(src[i*src_step],
155 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
161 LIFT(src[w*src_step],
162 ((mul*2*ref[w*ref_step]+add)>>shift),
168 static av_always_inline void
169 liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
170 int dst_step, int src_step, int ref_step,
171 int width, int mul, int add, int shift,
172 int highpass, int inverse){
173 const int mirror_left= !highpass;
174 const int mirror_right= (width&1) ^ highpass;
175 const int w= (width>>1) - 1 + (highpass & width);
179 #define LIFTS(src, ref, inv) \
181 (src) + (((ref) + 4*(src))>>shift): \
182 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
184 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
191 LIFTS(src[i*src_step],
192 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
198 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
201 static av_always_inline void
202 inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
203 int dst_step, int src_step, int ref_step,
204 int width, int mul, int add, int shift,
205 int highpass, int inverse){
206 const int mirror_left= !highpass;
207 const int mirror_right= (width&1) ^ highpass;
208 const int w= (width>>1) - 1 + (highpass & width);
212 #define LIFTS(src, ref, inv) \
214 (src) + (((ref) + 4*(src))>>shift): \
215 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
217 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
224 LIFTS(src[i*src_step],
225 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
231 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
236 static void horizontal_decompose53i(DWTELEM *b, int width){
238 const int width2= width>>1;
240 const int w2= (width+1)>>1;
242 for(x=0; x<width2; x++){
244 temp[x+w2]= b[2*x + 1];
258 for(x=1; x+1<width2; x+=2){
262 A2 += (A1 + A3 + 2)>>2;
266 A1= temp[x+1+width2];
269 A4 += (A1 + A3 + 2)>>2;
275 A2 += (A1 + A3 + 2)>>2;
280 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
281 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0);
285 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
288 for(i=0; i<width; i++){
289 b1[i] -= (b0[i] + b2[i])>>1;
293 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
296 for(i=0; i<width; i++){
297 b1[i] += (b0[i] + b2[i] + 2)>>2;
301 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
303 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
304 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride;
306 for(y=-2; y<height; y+=2){
307 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
308 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
310 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width);
311 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width);
313 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width);
314 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width);
321 static void horizontal_decompose97i(DWTELEM *b, int width){
323 const int w2= (width+1)>>1;
325 lift (temp+w2, b +1, b , 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1);
326 liftS(temp , b , temp+w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0);
327 lift (b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
328 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0);
332 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
335 for(i=0; i<width; i++){
336 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
340 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
343 for(i=0; i<width; i++){
344 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
348 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
351 for(i=0; i<width; i++){
353 b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
355 b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23);
360 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
363 for(i=0; i<width; i++){
364 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
368 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){
370 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
371 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride;
372 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
373 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
375 for(y=-4; y<height; y+=2){
376 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
377 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
379 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width);
380 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width);
382 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width);
383 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width);
384 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width);
385 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width);
394 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
397 for(level=0; level<decomposition_count; level++){
399 case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
400 case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
405 static void horizontal_compose53i(IDWTELEM *b, int width){
406 IDWTELEM temp[width];
407 const int width2= width>>1;
408 const int w2= (width+1)>>1;
411 for(x=0; x<width2; x++){
413 temp[2*x + 1]= b[x+w2];
418 b[0] = temp[0] - ((temp[1]+1)>>1);
419 for(x=2; x<width-1; x+=2){
420 b[x ] = temp[x ] - ((temp[x-1] + temp[x+1]+2)>>2);
421 b[x-1] = temp[x-1] + ((b [x-2] + b [x ]+1)>>1);
424 b[x ] = temp[x ] - ((temp[x-1]+1)>>1);
425 b[x-1] = temp[x-1] + ((b [x-2] + b [x ]+1)>>1);
427 b[x-1] = temp[x-1] + b[x-2];
430 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
433 for(i=0; i<width; i++){
434 b1[i] += (b0[i] + b2[i])>>1;
438 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
441 for(i=0; i<width; i++){
442 b1[i] -= (b0[i] + b2[i] + 2)>>2;
446 static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
447 cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line);
448 cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line);
452 static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
453 cs->b0 = buffer + mirror(-1-1, height-1)*stride;
454 cs->b1 = buffer + mirror(-1 , height-1)*stride;
458 static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
461 IDWTELEM *b0= cs->b0;
462 IDWTELEM *b1= cs->b1;
463 IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
464 IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
466 if(y+1<(unsigned)height && y<(unsigned)height){
469 for(x=0; x<width; x++){
470 b2[x] -= (b1[x] + b3[x] + 2)>>2;
471 b1[x] += (b0[x] + b2[x])>>1;
474 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
475 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
478 if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
479 if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
486 static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
488 IDWTELEM *b0= cs->b0;
489 IDWTELEM *b1= cs->b1;
490 IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
491 IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
493 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
494 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
496 if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
497 if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
504 static void av_unused spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){
506 spatial_compose53i_init(&cs, buffer, height, stride);
507 while(cs.y <= height)
508 spatial_compose53i_dy(&cs, buffer, width, height, stride);
512 void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){
513 IDWTELEM temp[width];
514 const int w2= (width+1)>>1;
516 #if 0 //maybe more understadable but slower
517 inv_lift (temp , b , b +w2, 2, 1, 1, width, W_DM, W_DO, W_DS, 0, 1);
518 inv_lift (temp+1 , b +w2, temp , 2, 1, 2, width, W_CM, W_CO, W_CS, 1, 1);
520 inv_liftS(b , temp , temp+1 , 2, 2, 2, width, W_BM, W_BO, W_BS, 0, 1);
521 inv_lift (b+1 , temp+1 , b , 2, 2, 2, width, W_AM, W_AO, W_AS, 1, 0);
524 temp[0] = b[0] - ((3*b[w2]+2)>>2);
525 for(x=1; x<(width>>1); x++){
526 temp[2*x ] = b[x ] - ((3*(b [x+w2-1] + b[x+w2])+4)>>3);
527 temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
530 temp[2*x ] = b[x ] - ((3*b [x+w2-1]+2)>>2);
531 temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
533 temp[2*x-1] = b[x+w2-1] - 2*temp[2*x-2];
535 b[0] = temp[0] + ((2*temp[0] + temp[1]+4)>>3);
536 for(x=2; x<width-1; x+=2){
537 b[x ] = temp[x ] + ((4*temp[x ] + temp[x-1] + temp[x+1]+8)>>4);
538 b[x-1] = temp[x-1] + ((3*(b [x-2] + b [x ] ))>>1);
541 b[x ] = temp[x ] + ((2*temp[x ] + temp[x-1]+4)>>3);
542 b[x-1] = temp[x-1] + ((3*(b [x-2] + b [x ] ))>>1);
544 b[x-1] = temp[x-1] + 3*b [x-2];
548 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
551 for(i=0; i<width; i++){
552 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
556 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
559 for(i=0; i<width; i++){
560 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
564 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
567 for(i=0; i<width; i++){
569 b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
571 b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;
576 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
579 for(i=0; i<width; i++){
580 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
584 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
587 for(i=0; i<width; i++){
588 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
589 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
591 b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS;
593 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
595 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
599 static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
600 cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line);
601 cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line);
602 cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line);
603 cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line);
607 static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
608 cs->b0 = buffer + mirror(-3-1, height-1)*stride;
609 cs->b1 = buffer + mirror(-3 , height-1)*stride;
610 cs->b2 = buffer + mirror(-3+1, height-1)*stride;
611 cs->b3 = buffer + mirror(-3+2, height-1)*stride;
615 static void spatial_compose97i_dy_buffered(DWTContext *dsp, DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
618 IDWTELEM *b0= cs->b0;
619 IDWTELEM *b1= cs->b1;
620 IDWTELEM *b2= cs->b2;
621 IDWTELEM *b3= cs->b3;
622 IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
623 IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
625 if(y>0 && y+4<height){
626 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
628 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
629 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
630 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
631 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
634 if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width);
635 if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width);
644 static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
646 IDWTELEM *b0= cs->b0;
647 IDWTELEM *b1= cs->b1;
648 IDWTELEM *b2= cs->b2;
649 IDWTELEM *b3= cs->b3;
650 IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
651 IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
653 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
654 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
655 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
656 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
658 if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width);
659 if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width);
668 static void av_unused spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){
670 spatial_compose97i_init(&cs, buffer, height, stride);
671 while(cs.y <= height)
672 spatial_compose97i_dy(&cs, buffer, width, height, stride);
675 void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){
677 for(level=decomposition_count-1; level>=0; level--){
679 case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
680 case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
685 void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
686 const int support = type==1 ? 3 : 5;
690 for(level=decomposition_count-1; level>=0; level--){
691 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
693 case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
695 case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
702 static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
704 for(level=decomposition_count-1; level>=0; level--){
706 case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
707 case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
712 static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
713 const int support = type==1 ? 3 : 5;
717 for(level=decomposition_count-1; level>=0; level--){
718 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
720 case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
722 case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
729 void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
730 DWTCompose cs[MAX_DECOMPOSITIONS];
732 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
733 for(y=0; y<height; y+=4)
734 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
737 static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){
739 const int dec_count= w==8 ? 3 : 4;
742 static const int scale[2][2][4][4]={
746 {268, 239, 239, 213},
750 // 9/7 16x16 or 32x32 dec=4
751 {344, 310, 310, 280},
759 {275, 245, 245, 218},
763 // 5/3 16x16 or 32x32 dec=4
764 {352, 317, 317, 286},
772 for (i = 0; i < h; i++) {
773 for (j = 0; j < w; j+=4) {
774 tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4;
775 tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4;
776 tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4;
777 tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4;
783 ff_spatial_dwt(tmp, w, h, 32, type, dec_count);
787 for(level=0; level<dec_count; level++){
788 for(ori= level ? 1 : 0; ori<4; ori++){
789 int size= w>>(dec_count-level);
790 int sx= (ori&1) ? size : 0;
791 int stride= 32<<(dec_count-level);
792 int sy= (ori&2) ? stride>>1 : 0;
794 for(i=0; i<size; i++){
795 for(j=0; j<size; j++){
796 int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori];
806 static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
807 return w_c(v, pix1, pix2, line_size, 8, h, 1);
810 static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
811 return w_c(v, pix1, pix2, line_size, 8, h, 0);
814 static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
815 return w_c(v, pix1, pix2, line_size, 16, h, 1);
818 static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
819 return w_c(v, pix1, pix2, line_size, 16, h, 0);
822 int ff_w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
823 return w_c(v, pix1, pix2, line_size, 32, h, 1);
826 int ff_w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
827 return w_c(v, pix1, pix2, line_size, 32, h, 0);
830 void ff_dsputil_init_dwt(DSPContext *c)
838 void ff_dwt_init(DWTContext *c)
840 c->vertical_compose97i = ff_snow_vertical_compose97i;
841 c->horizontal_compose97i = ff_snow_horizontal_compose97i;
842 c->inner_add_yblock = ff_snow_inner_add_yblock;
844 if (HAVE_MMX) ff_dwt_init_x86(c);
848 static av_always_inline
849 void interleave(IDWTELEM *dst, IDWTELEM *src0, IDWTELEM *src1, int w2, int add, int shift)
852 for (i = 0; i < w2; i++) {
853 dst[2*i ] = (src0[i] + add) >> shift;
854 dst[2*i+1] = (src1[i] + add) >> shift;
858 static void horizontal_compose_dirac53i(IDWTELEM *b, IDWTELEM *temp, int w)
860 const int w2 = w >> 1;
863 temp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
864 for (x = 1; x < w2; x++) {
865 temp[x ] = COMPOSE_53iL0 (b[x+w2-1], b[x ], b[x+w2]);
866 temp[x+w2-1] = COMPOSE_DIRAC53iH0(temp[x-1], b[x+w2-1], temp[x]);
868 temp[w-1] = COMPOSE_DIRAC53iH0(temp[w2-1], b[w-1], temp[w2-1]);
870 interleave(b, temp, temp+w2, w2, 1, 1);
873 static void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int w)
875 const int w2 = w >> 1;
878 tmp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
879 for (x = 1; x < w2; x++)
880 tmp[x] = COMPOSE_53iL0(b[x+w2-1], b[x], b[x+w2]);
884 tmp[w2+1] = tmp[w2] = tmp[w2-1];
886 for (x = 0; x < w2; x++) {
887 b[2*x ] = (tmp[x] + 1)>>1;
888 b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
892 static void horizontal_compose_dd137i(IDWTELEM *b, IDWTELEM *tmp, int w)
894 const int w2 = w >> 1;
897 tmp[0] = COMPOSE_DD137iL0(b[w2], b[w2], b[0], b[w2 ], b[w2+1]);
898 tmp[1] = COMPOSE_DD137iL0(b[w2], b[w2], b[1], b[w2+1], b[w2+2]);
899 for (x = 2; x < w2-1; x++)
900 tmp[x] = COMPOSE_DD137iL0(b[x+w2-2], b[x+w2-1], b[x], b[x+w2], b[x+w2+1]);
901 tmp[w2-1] = COMPOSE_DD137iL0(b[w-3], b[w-2], b[w2-1], b[w-1], b[w-1]);
905 tmp[w2+1] = tmp[w2] = tmp[w2-1];
907 for (x = 0; x < w2; x++) {
908 b[2*x ] = (tmp[x] + 1)>>1;
909 b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
913 static av_always_inline
914 void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *temp, int w, int shift)
916 const int w2 = w >> 1;
919 for (x = 0; x < w2; x++) {
920 temp[x ] = COMPOSE_HAARiL0(b[x ], b[x+w2]);
921 temp[x+w2] = COMPOSE_HAARiH0(b[x+w2], temp[x]);
924 interleave(b, temp, temp+w2, w2, shift, shift);
927 static void horizontal_compose_haar0i(IDWTELEM *b, IDWTELEM *temp, int w)
929 horizontal_compose_haari(b, temp, w, 0);
932 static void horizontal_compose_haar1i(IDWTELEM *b, IDWTELEM *temp, int w)
934 horizontal_compose_haari(b, temp, w, 1);
937 static void horizontal_compose_fidelityi(IDWTELEM *b, IDWTELEM *tmp, int w)
939 const int w2 = w >> 1;
943 for (x = 0; x < w2; x++) {
944 for (i = 0; i < 8; i++)
945 v[i] = b[av_clip(x-3+i, 0, w2-1)];
946 tmp[x] = COMPOSE_FIDELITYiH0(v[0], v[1], v[2], v[3], b[x+w2], v[4], v[5], v[6], v[7]);
949 for (x = 0; x < w2; x++) {
950 for (i = 0; i < 8; i++)
951 v[i] = tmp[av_clip(x-4+i, 0, w2-1)];
952 tmp[x+w2] = COMPOSE_FIDELITYiL0(v[0], v[1], v[2], v[3], b[x], v[4], v[5], v[6], v[7]);
955 interleave(b, tmp+w2, tmp, w2, 0, 0);
958 static void horizontal_compose_daub97i(IDWTELEM *b, IDWTELEM *temp, int w)
960 const int w2 = w >> 1;
963 temp[0] = COMPOSE_DAUB97iL1(b[w2], b[0], b[w2]);
964 for (x = 1; x < w2; x++) {
965 temp[x ] = COMPOSE_DAUB97iL1(b[x+w2-1], b[x ], b[x+w2]);
966 temp[x+w2-1] = COMPOSE_DAUB97iH1(temp[x-1], b[x+w2-1], temp[x]);
968 temp[w-1] = COMPOSE_DAUB97iH1(temp[w2-1], b[w-1], temp[w2-1]);
970 // second stage combined with interleave and shift
971 b0 = b2 = COMPOSE_DAUB97iL0(temp[w2], temp[0], temp[w2]);
972 b[0] = (b0 + 1) >> 1;
973 for (x = 1; x < w2; x++) {
974 b2 = COMPOSE_DAUB97iL0(temp[x+w2-1], temp[x ], temp[x+w2]);
975 b1 = COMPOSE_DAUB97iH0( b0, temp[x+w2-1], b2 );
976 b[2*x-1] = (b1 + 1) >> 1;
977 b[2*x ] = (b2 + 1) >> 1;
980 b[w-1] = (COMPOSE_DAUB97iH0(b2, temp[w-1], b2) + 1) >> 1;
983 static void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width)
987 for(i=0; i<width; i++){
988 b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]);
992 static void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
993 IDWTELEM *b3, IDWTELEM *b4, int width)
997 for(i=0; i<width; i++){
998 b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]);
1002 static void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
1003 IDWTELEM *b3, IDWTELEM *b4, int width)
1007 for(i=0; i<width; i++){
1008 b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]);
1012 static void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
1016 for (i = 0; i < width; i++) {
1017 b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]);
1018 b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]);
1022 static void vertical_compose_fidelityiH0(IDWTELEM *dst, IDWTELEM *b[8], int width)
1026 for(i=0; i<width; i++){
1027 dst[i] = COMPOSE_FIDELITYiH0(b[0][i], b[1][i], b[2][i], b[3][i], dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
1031 static void vertical_compose_fidelityiL0(IDWTELEM *dst, IDWTELEM *b[8], int width)
1035 for(i=0; i<width; i++){
1036 dst[i] = COMPOSE_FIDELITYiL0(b[0][i], b[1][i], b[2][i], b[3][i], dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
1040 static void vertical_compose_daub97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width)
1044 for(i=0; i<width; i++){
1045 b1[i] = COMPOSE_DAUB97iH0(b0[i], b1[i], b2[i]);
1049 static void vertical_compose_daub97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width)
1053 for(i=0; i<width; i++){
1054 b1[i] = COMPOSE_DAUB97iH1(b0[i], b1[i], b2[i]);
1058 static void vertical_compose_daub97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width)
1062 for(i=0; i<width; i++){
1063 b1[i] = COMPOSE_DAUB97iL0(b0[i], b1[i], b2[i]);
1067 static void vertical_compose_daub97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width)
1071 for(i=0; i<width; i++){
1072 b1[i] = COMPOSE_DAUB97iL1(b0[i], b1[i], b2[i]);
1077 static void spatial_compose_dd97i_dy(DWTContext *d, int level, int width, int height, int stride)
1079 vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
1080 vertical_compose_5tap vertical_compose_h0 = d->vertical_compose_h0;
1081 DWTCompose *cs = d->cs + level;
1085 for (i = 0; i < 6; i++)
1087 b[6] = d->buffer + av_clip(y+5, 0, height-2)*stride;
1088 b[7] = d->buffer + av_clip(y+6, 1, height-1)*stride;
1090 if(y+5<(unsigned)height) vertical_compose_l0( b[5], b[6], b[7], width);
1091 if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
1093 if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
1094 if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
1096 for (i = 0; i < 6; i++)
1101 static void spatial_compose_dirac53i_dy(DWTContext *d, int level, int width, int height, int stride)
1103 vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
1104 vertical_compose_3tap vertical_compose_h0 = d->vertical_compose_h0;
1105 DWTCompose *cs = d->cs + level;
1108 IDWTELEM *b[4] = { cs->b[0], cs->b[1] };
1109 b[2] = d->buffer + mirror(y+1, height-1)*stride;
1110 b[3] = d->buffer + mirror(y+2, height-1)*stride;
1112 if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
1113 if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
1115 if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
1116 if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
1124 static void spatial_compose_dd137i_dy(DWTContext *d, int level, int width, int height, int stride)
1126 vertical_compose_5tap vertical_compose_l0 = d->vertical_compose_l0;
1127 vertical_compose_5tap vertical_compose_h0 = d->vertical_compose_h0;
1128 DWTCompose *cs = d->cs + level;
1132 for (i = 0; i < 8; i++)
1134 b[8] = d->buffer + av_clip(y+7, 0, height-2)*stride;
1135 b[9] = d->buffer + av_clip(y+8, 1, height-1)*stride;
1137 if(y+5<(unsigned)height) vertical_compose_l0(b[3], b[5], b[6], b[7], b[9], width);
1138 if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
1140 if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
1141 if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
1143 for (i = 0; i < 8; i++)
1148 // haar makes the assumption that height is even (always true for dirac)
1149 static void spatial_compose_haari_dy(DWTContext *d, int level, int width, int height, int stride)
1151 vertical_compose_2tap vertical_compose = d->vertical_compose;
1152 int y = d->cs[level].y;
1153 IDWTELEM *b0 = d->buffer + (y-1)*stride;
1154 IDWTELEM *b1 = d->buffer + (y )*stride;
1156 vertical_compose(b0, b1, width);
1157 d->horizontal_compose(b0, d->temp, width);
1158 d->horizontal_compose(b1, d->temp, width);
1160 d->cs[level].y += 2;
1163 // Don't do sliced idwt for fidelity; the 9 tap filter makes it a bit annoying
1164 // Fortunately, this filter isn't used in practice.
1165 static void spatial_compose_fidelity(DWTContext *d, int level, int width, int height, int stride)
1167 vertical_compose_9tap vertical_compose_l0 = d->vertical_compose_l0;
1168 vertical_compose_9tap vertical_compose_h0 = d->vertical_compose_h0;
1172 for (y = 1; y < height; y += 2) {
1173 for (i = 0; i < 8; i++)
1174 b[i] = d->buffer + av_clip((y-7 + 2*i), 0, height-2)*stride;
1175 vertical_compose_h0(d->buffer + y*stride, b, width);
1178 for (y = 0; y < height; y += 2) {
1179 for (i = 0; i < 8; i++)
1180 b[i] = d->buffer + av_clip((y-7 + 2*i), 1, height-1)*stride;
1181 vertical_compose_l0(d->buffer + y*stride, b, width);
1184 for (y = 0; y < height; y++)
1185 d->horizontal_compose(d->buffer + y*stride, d->temp, width);
1187 d->cs[level].y = height+1;
1190 static void spatial_compose_daub97i_dy(DWTContext *d, int level, int width, int height, int stride)
1192 vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
1193 vertical_compose_3tap vertical_compose_h0 = d->vertical_compose_h0;
1194 vertical_compose_3tap vertical_compose_l1 = d->vertical_compose_l1;
1195 vertical_compose_3tap vertical_compose_h1 = d->vertical_compose_h1;
1196 DWTCompose *cs = d->cs + level;
1200 for (i = 0; i < 4; i++)
1202 b[4] = d->buffer + mirror(y+3, height-1)*stride;
1203 b[5] = d->buffer + mirror(y+4, height-1)*stride;
1205 if(y+3<(unsigned)height) vertical_compose_l1(b[3], b[4], b[5], width);
1206 if(y+2<(unsigned)height) vertical_compose_h1(b[2], b[3], b[4], width);
1207 if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
1208 if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
1210 if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
1211 if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
1213 for (i = 0; i < 4; i++)
1219 static void spatial_compose97i_init2(DWTCompose *cs, IDWTELEM *buffer, int height, int stride)
1221 cs->b[0] = buffer + mirror(-3-1, height-1)*stride;
1222 cs->b[1] = buffer + mirror(-3 , height-1)*stride;
1223 cs->b[2] = buffer + mirror(-3+1, height-1)*stride;
1224 cs->b[3] = buffer + mirror(-3+2, height-1)*stride;
1228 static void spatial_compose53i_init2(DWTCompose *cs, IDWTELEM *buffer, int height, int stride)
1230 cs->b[0] = buffer + mirror(-1-1, height-1)*stride;
1231 cs->b[1] = buffer + mirror(-1 , height-1)*stride;
1235 static void spatial_compose_dd97i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride)
1237 cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
1238 cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
1239 cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
1240 cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
1241 cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
1242 cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
1246 static void spatial_compose_dd137i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride)
1248 cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
1249 cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
1250 cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
1251 cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
1252 cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
1253 cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
1254 cs->b[6] = buffer + av_clip(-5+5, 0, height-2)*stride;
1255 cs->b[7] = buffer + av_clip(-5+6, 1, height-1)*stride;
1259 int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width, int height,
1260 int stride, enum dwt_type type, int decomposition_count,
1269 d->decomposition_count = decomposition_count;
1272 for(level=decomposition_count-1; level>=0; level--){
1273 int hl = height >> level;
1274 int stride_l = stride << level;
1277 case DWT_DIRAC_DD9_7:
1278 spatial_compose_dd97i_init(d->cs+level, buffer, hl, stride_l);
1280 case DWT_DIRAC_LEGALL5_3:
1281 spatial_compose53i_init2(d->cs+level, buffer, hl, stride_l);
1283 case DWT_DIRAC_DD13_7:
1284 spatial_compose_dd137i_init(d->cs+level, buffer, hl, stride_l);
1286 case DWT_DIRAC_HAAR0:
1287 case DWT_DIRAC_HAAR1:
1290 case DWT_DIRAC_DAUB9_7:
1291 spatial_compose97i_init2(d->cs+level, buffer, hl, stride_l);
1300 case DWT_DIRAC_DD9_7:
1301 d->spatial_compose = spatial_compose_dd97i_dy;
1302 d->vertical_compose_l0 = vertical_compose53iL0;
1303 d->vertical_compose_h0 = vertical_compose_dd97iH0;
1304 d->horizontal_compose = horizontal_compose_dd97i;
1307 case DWT_DIRAC_LEGALL5_3:
1308 d->spatial_compose = spatial_compose_dirac53i_dy;
1309 d->vertical_compose_l0 = vertical_compose53iL0;
1310 d->vertical_compose_h0 = vertical_compose_dirac53iH0;
1311 d->horizontal_compose = horizontal_compose_dirac53i;
1314 case DWT_DIRAC_DD13_7:
1315 d->spatial_compose = spatial_compose_dd137i_dy;
1316 d->vertical_compose_l0 = vertical_compose_dd137iL0;
1317 d->vertical_compose_h0 = vertical_compose_dd97iH0;
1318 d->horizontal_compose = horizontal_compose_dd137i;
1321 case DWT_DIRAC_HAAR0:
1322 case DWT_DIRAC_HAAR1:
1323 d->spatial_compose = spatial_compose_haari_dy;
1324 d->vertical_compose = vertical_compose_haar;
1325 if (type == DWT_DIRAC_HAAR0)
1326 d->horizontal_compose = horizontal_compose_haar0i;
1328 d->horizontal_compose = horizontal_compose_haar1i;
1331 case DWT_DIRAC_FIDELITY:
1332 d->spatial_compose = spatial_compose_fidelity;
1333 d->vertical_compose_l0 = vertical_compose_fidelityiL0;
1334 d->vertical_compose_h0 = vertical_compose_fidelityiH0;
1335 d->horizontal_compose = horizontal_compose_fidelityi;
1337 case DWT_DIRAC_DAUB9_7:
1338 d->spatial_compose = spatial_compose_daub97i_dy;
1339 d->vertical_compose_l0 = vertical_compose_daub97iL0;
1340 d->vertical_compose_h0 = vertical_compose_daub97iH0;
1341 d->vertical_compose_l1 = vertical_compose_daub97iL1;
1342 d->vertical_compose_h1 = vertical_compose_daub97iH1;
1343 d->horizontal_compose = horizontal_compose_daub97i;
1347 av_log(NULL, AV_LOG_ERROR, "Unknown wavelet type %d\n", type);
1351 if (HAVE_MMX) ff_spatial_idwt_init_mmx(d, type);
1356 void ff_spatial_idwt_slice2(DWTContext *d, int y)
1358 int level, support = d->support;
1360 for (level = d->decomposition_count-1; level >= 0; level--) {
1361 int wl = d->width >> level;
1362 int hl = d->height >> level;
1363 int stride_l = d->stride << level;
1365 while (d->cs[level].y <= FFMIN((y>>level)+support, hl))
1366 d->spatial_compose(d, level, wl, hl, stride_l);
1370 int ff_spatial_idwt2(IDWTELEM *buffer, int width, int height, int stride,
1371 enum dwt_type type, int decomposition_count, IDWTELEM *temp)
1376 if (ff_spatial_idwt_init2(&d, buffer, width, height, stride, type, decomposition_count, temp))
1379 for (y = 0; y < d.height; y += 4)
1380 ff_spatial_idwt_slice2(&d, y);