2 * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of Libav.
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/attributes.h"
22 #include "libavutil/common.h"
26 int ff_slice_buffer_init(slice_buffer *buf, int line_count,
27 int max_allocated_lines, int line_width,
28 IDWTELEM *base_buffer)
32 buf->base_buffer = base_buffer;
33 buf->line_count = line_count;
34 buf->line_width = line_width;
35 buf->data_count = max_allocated_lines;
36 buf->line = av_mallocz(sizeof(IDWTELEM *) * line_count);
38 return AVERROR(ENOMEM);
39 buf->data_stack = av_malloc(sizeof(IDWTELEM *) * max_allocated_lines);
40 if (!buf->data_stack) {
42 return AVERROR(ENOMEM);
45 for (i = 0; i < max_allocated_lines; i++) {
46 buf->data_stack[i] = av_malloc(sizeof(IDWTELEM) * line_width);
47 if (!buf->data_stack[i]) {
49 av_free(buf->data_stack[i]);
50 av_free(buf->data_stack);
52 return AVERROR(ENOMEM);
56 buf->data_stack_top = max_allocated_lines - 1;
60 IDWTELEM *ff_slice_buffer_load_line(slice_buffer *buf, int line)
64 assert(buf->data_stack_top >= 0);
65 // assert(!buf->line[line]);
67 return buf->line[line];
69 buffer = buf->data_stack[buf->data_stack_top];
70 buf->data_stack_top--;
71 buf->line[line] = buffer;
76 void ff_slice_buffer_release(slice_buffer *buf, int line)
80 assert(line >= 0 && line < buf->line_count);
81 assert(buf->line[line]);
83 buffer = buf->line[line];
84 buf->data_stack_top++;
85 buf->data_stack[buf->data_stack_top] = buffer;
86 buf->line[line] = NULL;
89 void ff_slice_buffer_flush(slice_buffer *buf)
92 for (i = 0; i < buf->line_count; i++)
94 ff_slice_buffer_release(buf, i);
97 void ff_slice_buffer_destroy(slice_buffer *buf)
100 ff_slice_buffer_flush(buf);
102 for (i = buf->data_count - 1; i >= 0; i--)
103 av_freep(&buf->data_stack[i]);
104 av_freep(&buf->data_stack);
105 av_freep(&buf->line);
108 static inline int mirror(int v, int m)
110 while ((unsigned)v > (unsigned)m) {
118 static av_always_inline void lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
119 int dst_step, int src_step, int ref_step,
120 int width, int mul, int add, int shift,
121 int highpass, int inverse)
123 const int mirror_left = !highpass;
124 const int mirror_right = (width & 1) ^ highpass;
125 const int w = (width >> 1) - 1 + (highpass & width);
128 #define LIFT(src, ref, inv) ((src) + ((inv) ? -(ref) : +(ref)))
130 dst[0] = LIFT(src[0], ((mul * 2 * ref[0] + add) >> shift), inverse);
135 for (i = 0; i < w; i++)
136 dst[i * dst_step] = LIFT(src[i * src_step],
137 ((mul * (ref[i * ref_step] +
138 ref[(i + 1) * ref_step]) +
143 dst[w * dst_step] = LIFT(src[w * src_step],
144 ((mul * 2 * ref[w * ref_step] + add) >> shift),
148 static av_always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
149 int dst_step, int src_step, int ref_step,
150 int width, int mul, int add, int shift,
151 int highpass, int inverse)
153 const int mirror_left = !highpass;
154 const int mirror_right = (width & 1) ^ highpass;
155 const int w = (width >> 1) - 1 + (highpass & width);
159 #define LIFTS(src, ref, inv) \
160 ((inv) ? (src) + (((ref) + 4 * (src)) >> shift) \
161 : -((-16 * (src) + (ref) + add / \
162 4 + 1 + (5 << 25)) / (5 * 4) - (1 << 23)))
164 dst[0] = LIFTS(src[0], mul * 2 * ref[0] + add, inverse);
169 for (i = 0; i < w; i++)
170 dst[i * dst_step] = LIFTS(src[i * src_step],
171 mul * (ref[i * ref_step] +
172 ref[(i + 1) * ref_step]) + add,
176 dst[w * dst_step] = LIFTS(src[w * src_step],
177 mul * 2 * ref[w * ref_step] + add,
181 static void horizontal_decompose53i(DWTELEM *b, DWTELEM *temp, int width)
183 const int width2 = width >> 1;
185 const int w2 = (width + 1) >> 1;
187 for (x = 0; x < width2; x++) {
189 temp[x + w2] = b[2 * x + 1];
193 lift(b + w2, temp + w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
194 lift(b, temp, b + w2, 1, 1, 1, width, 1, 2, 2, 0, 0);
197 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
202 for (i = 0; i < width; i++)
203 b1[i] -= (b0[i] + b2[i]) >> 1;
206 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
211 for (i = 0; i < width; i++)
212 b1[i] += (b0[i] + b2[i] + 2) >> 2;
215 static void spatial_decompose53i(DWTELEM *buffer, DWTELEM *temp,
216 int width, int height, int stride)
219 DWTELEM *b0 = buffer + mirror(-2 - 1, height - 1) * stride;
220 DWTELEM *b1 = buffer + mirror(-2, height - 1) * stride;
222 for (y = -2; y < height; y += 2) {
223 DWTELEM *b2 = buffer + mirror(y + 1, height - 1) * stride;
224 DWTELEM *b3 = buffer + mirror(y + 2, height - 1) * stride;
226 if (y + 1 < (unsigned)height)
227 horizontal_decompose53i(b2, temp, width);
228 if (y + 2 < (unsigned)height)
229 horizontal_decompose53i(b3, temp, width);
231 if (y + 1 < (unsigned)height)
232 vertical_decompose53iH0(b1, b2, b3, width);
233 if (y + 0 < (unsigned)height)
234 vertical_decompose53iL0(b0, b1, b2, width);
241 static void horizontal_decompose97i(DWTELEM *b, DWTELEM *temp, int width)
243 const int w2 = (width + 1) >> 1;
245 lift(temp + w2, b + 1, b, 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1);
246 liftS(temp, b, temp + w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0);
247 lift(b + w2, temp + w2, temp, 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
248 lift(b, temp, b + w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0);
251 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
256 for (i = 0; i < width; i++)
257 b1[i] -= (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
260 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
265 for (i = 0; i < width; i++)
266 b1[i] += (W_CM * (b0[i] + b2[i]) + W_CO) >> W_CS;
269 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
274 for (i = 0; i < width; i++)
275 b1[i] = (16 * 4 * b1[i] - 4 * (b0[i] + b2[i]) + W_BO * 5 + (5 << 27)) /
276 (5 * 16) - (1 << 23);
279 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
284 for (i = 0; i < width; i++)
285 b1[i] += (W_DM * (b0[i] + b2[i]) + W_DO) >> W_DS;
288 static void spatial_decompose97i(DWTELEM *buffer, DWTELEM *temp,
289 int width, int height, int stride)
292 DWTELEM *b0 = buffer + mirror(-4 - 1, height - 1) * stride;
293 DWTELEM *b1 = buffer + mirror(-4, height - 1) * stride;
294 DWTELEM *b2 = buffer + mirror(-4 + 1, height - 1) * stride;
295 DWTELEM *b3 = buffer + mirror(-4 + 2, height - 1) * stride;
297 for (y = -4; y < height; y += 2) {
298 DWTELEM *b4 = buffer + mirror(y + 3, height - 1) * stride;
299 DWTELEM *b5 = buffer + mirror(y + 4, height - 1) * stride;
301 if (y + 3 < (unsigned)height)
302 horizontal_decompose97i(b4, temp, width);
303 if (y + 4 < (unsigned)height)
304 horizontal_decompose97i(b5, temp, width);
306 if (y + 3 < (unsigned)height)
307 vertical_decompose97iH0(b3, b4, b5, width);
308 if (y + 2 < (unsigned)height)
309 vertical_decompose97iL0(b2, b3, b4, width);
310 if (y + 1 < (unsigned)height)
311 vertical_decompose97iH1(b1, b2, b3, width);
312 if (y + 0 < (unsigned)height)
313 vertical_decompose97iL1(b0, b1, b2, width);
322 void ff_spatial_dwt(DWTELEM *buffer, DWTELEM *temp, int width, int height,
323 int stride, int type, int decomposition_count)
327 for (level = 0; level < decomposition_count; level++) {
330 spatial_decompose97i(buffer, temp,
331 width >> level, height >> level,
335 spatial_decompose53i(buffer, temp,
336 width >> level, height >> level,
343 static void horizontal_compose53i(IDWTELEM *b, IDWTELEM *temp, int width)
345 const int width2 = width >> 1;
346 const int w2 = (width + 1) >> 1;
349 for (x = 0; x < width2; x++) {
351 temp[2 * x + 1] = b[x + w2];
356 b[0] = temp[0] - ((temp[1] + 1) >> 1);
357 for (x = 2; x < width - 1; x += 2) {
358 b[x] = temp[x] - ((temp[x - 1] + temp[x + 1] + 2) >> 2);
359 b[x - 1] = temp[x - 1] + ((b[x - 2] + b[x] + 1) >> 1);
362 b[x] = temp[x] - ((temp[x - 1] + 1) >> 1);
363 b[x - 1] = temp[x - 1] + ((b[x - 2] + b[x] + 1) >> 1);
365 b[x - 1] = temp[x - 1] + b[x - 2];
368 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
373 for (i = 0; i < width; i++)
374 b1[i] += (b0[i] + b2[i]) >> 1;
377 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
382 for (i = 0; i < width; i++)
383 b1[i] -= (b0[i] + b2[i] + 2) >> 2;
386 static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer *sb,
387 int height, int stride_line)
389 cs->b0 = slice_buffer_get_line(sb,
390 mirror(-1 - 1, height - 1) * stride_line);
391 cs->b1 = slice_buffer_get_line(sb, mirror(-1, height - 1) * stride_line);
395 static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer,
396 int height, int stride)
398 cs->b0 = buffer + mirror(-1 - 1, height - 1) * stride;
399 cs->b1 = buffer + mirror(-1, height - 1) * stride;
403 static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer *sb,
405 int width, int height,
410 IDWTELEM *b0 = cs->b0;
411 IDWTELEM *b1 = cs->b1;
412 IDWTELEM *b2 = slice_buffer_get_line(sb,
413 mirror(y + 1, height - 1) *
415 IDWTELEM *b3 = slice_buffer_get_line(sb,
416 mirror(y + 2, height - 1) *
419 if (y + 1 < (unsigned)height && y < (unsigned)height) {
422 for (x = 0; x < width; x++) {
423 b2[x] -= (b1[x] + b3[x] + 2) >> 2;
424 b1[x] += (b0[x] + b2[x]) >> 1;
427 if (y + 1 < (unsigned)height)
428 vertical_compose53iL0(b1, b2, b3, width);
429 if (y + 0 < (unsigned)height)
430 vertical_compose53iH0(b0, b1, b2, width);
433 if (y - 1 < (unsigned)height)
434 horizontal_compose53i(b0, temp, width);
435 if (y + 0 < (unsigned)height)
436 horizontal_compose53i(b1, temp, width);
443 static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer,
444 IDWTELEM *temp, int width, int height,
448 IDWTELEM *b0 = cs->b0;
449 IDWTELEM *b1 = cs->b1;
450 IDWTELEM *b2 = buffer + mirror(y + 1, height - 1) * stride;
451 IDWTELEM *b3 = buffer + mirror(y + 2, height - 1) * stride;
453 if (y + 1 < (unsigned)height)
454 vertical_compose53iL0(b1, b2, b3, width);
455 if (y + 0 < (unsigned)height)
456 vertical_compose53iH0(b0, b1, b2, width);
458 if (y - 1 < (unsigned)height)
459 horizontal_compose53i(b0, temp, width);
460 if (y + 0 < (unsigned)height)
461 horizontal_compose53i(b1, temp, width);
468 void ff_snow_horizontal_compose97i(IDWTELEM *b, IDWTELEM *temp, int width)
470 const int w2 = (width + 1) >> 1;
473 temp[0] = b[0] - ((3 * b[w2] + 2) >> 2);
474 for (x = 1; x < (width >> 1); x++) {
475 temp[2 * x] = b[x] - ((3 * (b[x + w2 - 1] + b[x + w2]) + 4) >> 3);
476 temp[2 * x - 1] = b[x + w2 - 1] - temp[2 * x - 2] - temp[2 * x];
479 temp[2 * x] = b[x] - ((3 * b[x + w2 - 1] + 2) >> 2);
480 temp[2 * x - 1] = b[x + w2 - 1] - temp[2 * x - 2] - temp[2 * x];
482 temp[2 * x - 1] = b[x + w2 - 1] - 2 * temp[2 * x - 2];
484 b[0] = temp[0] + ((2 * temp[0] + temp[1] + 4) >> 3);
485 for (x = 2; x < width - 1; x += 2) {
486 b[x] = temp[x] + ((4 * temp[x] + temp[x - 1] + temp[x + 1] + 8) >> 4);
487 b[x - 1] = temp[x - 1] + ((3 * (b[x - 2] + b[x])) >> 1);
490 b[x] = temp[x] + ((2 * temp[x] + temp[x - 1] + 4) >> 3);
491 b[x - 1] = temp[x - 1] + ((3 * (b[x - 2] + b[x])) >> 1);
493 b[x - 1] = temp[x - 1] + 3 * b[x - 2];
496 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
501 for (i = 0; i < width; i++)
502 b1[i] += (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
505 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
510 for (i = 0; i < width; i++)
511 b1[i] -= (W_CM * (b0[i] + b2[i]) + W_CO) >> W_CS;
514 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
519 for (i = 0; i < width; i++)
520 b1[i] += (W_BM * (b0[i] + b2[i]) + 4 * b1[i] + W_BO) >> W_BS;
523 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
528 for (i = 0; i < width; i++)
529 b1[i] -= (W_DM * (b0[i] + b2[i]) + W_DO) >> W_DS;
532 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
533 IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5,
538 for (i = 0; i < width; i++) {
539 b4[i] -= (W_DM * (b3[i] + b5[i]) + W_DO) >> W_DS;
540 b3[i] -= (W_CM * (b2[i] + b4[i]) + W_CO) >> W_CS;
541 b2[i] += (W_BM * (b1[i] + b3[i]) + 4 * b2[i] + W_BO) >> W_BS;
542 b1[i] += (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
546 static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer *sb,
547 int height, int stride_line)
549 cs->b0 = slice_buffer_get_line(sb, mirror(-3 - 1, height - 1) * stride_line);
550 cs->b1 = slice_buffer_get_line(sb, mirror(-3, height - 1) * stride_line);
551 cs->b2 = slice_buffer_get_line(sb, mirror(-3 + 1, height - 1) * stride_line);
552 cs->b3 = slice_buffer_get_line(sb, mirror(-3 + 2, height - 1) * stride_line);
556 static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height,
559 cs->b0 = buffer + mirror(-3 - 1, height - 1) * stride;
560 cs->b1 = buffer + mirror(-3, height - 1) * stride;
561 cs->b2 = buffer + mirror(-3 + 1, height - 1) * stride;
562 cs->b3 = buffer + mirror(-3 + 2, height - 1) * stride;
566 static void spatial_compose97i_dy_buffered(DWTContext *dsp, DWTCompose *cs,
567 slice_buffer * sb, IDWTELEM *temp,
568 int width, int height,
573 IDWTELEM *b0 = cs->b0;
574 IDWTELEM *b1 = cs->b1;
575 IDWTELEM *b2 = cs->b2;
576 IDWTELEM *b3 = cs->b3;
577 IDWTELEM *b4 = slice_buffer_get_line(sb,
578 mirror(y + 3, height - 1) *
580 IDWTELEM *b5 = slice_buffer_get_line(sb,
581 mirror(y + 4, height - 1) *
584 if (y > 0 && y + 4 < height) {
585 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
587 if (y + 3 < (unsigned)height)
588 vertical_compose97iL1(b3, b4, b5, width);
589 if (y + 2 < (unsigned)height)
590 vertical_compose97iH1(b2, b3, b4, width);
591 if (y + 1 < (unsigned)height)
592 vertical_compose97iL0(b1, b2, b3, width);
593 if (y + 0 < (unsigned)height)
594 vertical_compose97iH0(b0, b1, b2, width);
597 if (y - 1 < (unsigned)height)
598 dsp->horizontal_compose97i(b0, temp, width);
599 if (y + 0 < (unsigned)height)
600 dsp->horizontal_compose97i(b1, temp, width);
609 static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer,
610 IDWTELEM *temp, int width, int height,
614 IDWTELEM *b0 = cs->b0;
615 IDWTELEM *b1 = cs->b1;
616 IDWTELEM *b2 = cs->b2;
617 IDWTELEM *b3 = cs->b3;
618 IDWTELEM *b4 = buffer + mirror(y + 3, height - 1) * stride;
619 IDWTELEM *b5 = buffer + mirror(y + 4, height - 1) * stride;
621 if (y + 3 < (unsigned)height)
622 vertical_compose97iL1(b3, b4, b5, width);
623 if (y + 2 < (unsigned)height)
624 vertical_compose97iH1(b2, b3, b4, width);
625 if (y + 1 < (unsigned)height)
626 vertical_compose97iL0(b1, b2, b3, width);
627 if (y + 0 < (unsigned)height)
628 vertical_compose97iH0(b0, b1, b2, width);
630 if (y - 1 < (unsigned)height)
631 ff_snow_horizontal_compose97i(b0, temp, width);
632 if (y + 0 < (unsigned)height)
633 ff_snow_horizontal_compose97i(b1, temp, width);
642 void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer *sb, int width,
643 int height, int stride_line, int type,
644 int decomposition_count)
647 for (level = decomposition_count - 1; level >= 0; level--) {
650 spatial_compose97i_buffered_init(cs + level, sb, height >> level,
651 stride_line << level);
654 spatial_compose53i_buffered_init(cs + level, sb, height >> level,
655 stride_line << level);
661 void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs,
662 slice_buffer *slice_buf, IDWTELEM *temp,
663 int width, int height, int stride_line,
664 int type, int decomposition_count, int y)
666 const int support = type == 1 ? 3 : 5;
671 for (level = decomposition_count - 1; level >= 0; level--)
672 while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
675 spatial_compose97i_dy_buffered(dsp, cs + level, slice_buf, temp,
678 stride_line << level);
681 spatial_compose53i_dy_buffered(cs + level, slice_buf, temp,
684 stride_line << level);
690 static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width,
691 int height, int stride, int type,
692 int decomposition_count)
695 for (level = decomposition_count - 1; level >= 0; level--) {
698 spatial_compose97i_init(cs + level, buffer, height >> level,
702 spatial_compose53i_init(cs + level, buffer, height >> level,
709 static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer,
710 IDWTELEM *temp, int width, int height,
711 int stride, int type,
712 int decomposition_count, int y)
714 const int support = type == 1 ? 3 : 5;
719 for (level = decomposition_count - 1; level >= 0; level--)
720 while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
723 spatial_compose97i_dy(cs + level, buffer, temp, width >> level,
724 height >> level, stride << level);
727 spatial_compose53i_dy(cs + level, buffer, temp, width >> level,
728 height >> level, stride << level);
734 void ff_spatial_idwt(IDWTELEM *buffer, IDWTELEM *temp, int width, int height,
735 int stride, int type, int decomposition_count)
737 DWTCompose cs[MAX_DECOMPOSITIONS];
739 ff_spatial_idwt_init(cs, buffer, width, height, stride, type,
740 decomposition_count);
741 for (y = 0; y < height; y += 4)
742 ff_spatial_idwt_slice(cs, buffer, temp, width, height, stride, type,
743 decomposition_count, y);
746 static inline int w_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size,
747 int w, int h, int type)
750 const int dec_count = w == 8 ? 3 : 4;
751 int tmp[32 * 32], tmp2[32];
753 static const int scale[2][2][4][4] = {
756 { 268, 239, 239, 213 },
757 { 0, 224, 224, 152 },
758 { 0, 135, 135, 110 },
760 { // 9/7 16x16 or 32x32 dec=4
761 { 344, 310, 310, 280 },
762 { 0, 320, 320, 228 },
763 { 0, 175, 175, 136 },
764 { 0, 129, 129, 102 },
769 { 275, 245, 245, 218 },
770 { 0, 230, 230, 156 },
771 { 0, 138, 138, 113 },
773 { // 5/3 16x16 or 32x32 dec=4
774 { 352, 317, 317, 286 },
775 { 0, 328, 328, 233 },
776 { 0, 180, 180, 140 },
777 { 0, 132, 132, 105 },
782 for (i = 0; i < h; i++) {
783 for (j = 0; j < w; j += 4) {
784 tmp[32 * i + j + 0] = (pix1[j + 0] - pix2[j + 0]) << 4;
785 tmp[32 * i + j + 1] = (pix1[j + 1] - pix2[j + 1]) << 4;
786 tmp[32 * i + j + 2] = (pix1[j + 2] - pix2[j + 2]) << 4;
787 tmp[32 * i + j + 3] = (pix1[j + 3] - pix2[j + 3]) << 4;
793 ff_spatial_dwt(tmp, tmp2, w, h, 32, type, dec_count);
797 for (level = 0; level < dec_count; level++)
798 for (ori = level ? 1 : 0; ori < 4; ori++) {
799 int size = w >> (dec_count - level);
800 int sx = (ori & 1) ? size : 0;
801 int stride = 32 << (dec_count - level);
802 int sy = (ori & 2) ? stride >> 1 : 0;
804 for (i = 0; i < size; i++)
805 for (j = 0; j < size; j++) {
806 int v = tmp[sx + sy + i * stride + j] *
807 scale[type][dec_count - 3][level][ori];
815 static int w53_8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
817 return w_c(v, pix1, pix2, line_size, 8, h, 1);
820 static int w97_8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
822 return w_c(v, pix1, pix2, line_size, 8, h, 0);
825 static int w53_16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
827 return w_c(v, pix1, pix2, line_size, 16, h, 1);
830 static int w97_16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
832 return w_c(v, pix1, pix2, line_size, 16, h, 0);
835 int ff_w53_32_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
837 return w_c(v, pix1, pix2, line_size, 32, h, 1);
840 int ff_w97_32_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
842 return w_c(v, pix1, pix2, line_size, 32, h, 0);
845 void ff_dsputil_init_dwt(DSPContext *c)
847 c->w53[0] = w53_16_c;
849 c->w97[0] = w97_16_c;
853 void ff_dwt_init(DWTContext *c)
855 c->vertical_compose97i = ff_snow_vertical_compose97i;
856 c->horizontal_compose97i = ff_snow_horizontal_compose97i;
857 c->inner_add_yblock = ff_snow_inner_add_yblock;