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 inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *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);
158 #define LIFT(src, ref, inv) ((src) + ((inv) ? -(ref) : +(ref)))
160 dst[0] = LIFT(src[0], ((mul * 2 * ref[0] + add) >> shift), inverse);
165 for (i = 0; i < w; i++)
166 dst[i * dst_step] = LIFT(src[i * src_step],
167 ((mul * (ref[i * ref_step] +
168 ref[(i + 1) * ref_step]) +
173 dst[w * dst_step] = LIFT(src[w * src_step],
174 ((mul * 2 * ref[w * ref_step] + add) >> shift),
180 static av_always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
181 int dst_step, int src_step, int ref_step,
182 int width, int mul, int add, int shift,
183 int highpass, int inverse)
185 const int mirror_left = !highpass;
186 const int mirror_right = (width & 1) ^ highpass;
187 const int w = (width >> 1) - 1 + (highpass & width);
191 #define LIFTS(src, ref, inv) \
192 ((inv) ? (src) + (((ref) + 4 * (src)) >> shift) \
193 : -((-16 * (src) + (ref) + add / \
194 4 + 1 + (5 << 25)) / (5 * 4) - (1 << 23)))
196 dst[0] = LIFTS(src[0], mul * 2 * ref[0] + add, inverse);
201 for (i = 0; i < w; i++)
202 dst[i * dst_step] = LIFTS(src[i * src_step],
203 mul * (ref[i * ref_step] +
204 ref[(i + 1) * ref_step]) + add,
208 dst[w * dst_step] = LIFTS(src[w * src_step],
209 mul * 2 * ref[w * ref_step] + add,
213 static av_always_inline void inv_liftS(IDWTELEM *dst, IDWTELEM *src,
214 IDWTELEM *ref, int dst_step,
215 int src_step, int ref_step,
216 int width, int mul, int add, int shift,
217 int highpass, int inverse)
219 const int mirror_left = !highpass;
220 const int mirror_right = (width & 1) ^ highpass;
221 const int w = (width >> 1) - 1 + (highpass & width);
225 #define LIFTS(src, ref, inv) \
226 ((inv) ? (src) + (((ref) + 4 * (src)) >> shift) \
227 : -((-16 * (src) + (ref) + add / \
228 4 + 1 + (5 << 25)) / (5 * 4) - (1 << 23)))
230 dst[0] = LIFTS(src[0], mul * 2 * ref[0] + add, inverse);
235 for (i = 0; i < w; i++)
236 dst[i * dst_step] = LIFTS(src[i * src_step],
237 mul * (ref[i * ref_step] +
238 ref[(i + 1) * ref_step]) + add,
242 dst[w * dst_step] = LIFTS(src[w * src_step],
243 mul * 2 * ref[w * ref_step] + add, inverse);
247 static void horizontal_decompose53i(DWTELEM *b, DWTELEM *temp, int width)
249 const int width2 = width >> 1;
251 const int w2 = (width + 1) >> 1;
253 for (x = 0; x < width2; x++) {
255 temp[x + w2] = b[2 * x + 1];
264 A1 = temp[0 + width2];
265 A1 -= (A2 + A4) >> 1;
269 for (x = 1; x + 1 < width2; x += 2) {
270 A3 = temp[x + width2];
272 A3 -= (A2 + A4) >> 1;
273 A2 += (A1 + A3 + 2) >> 2;
277 A1 = temp[x + 1 + width2];
279 A1 -= (A2 + A4) >> 1;
280 A4 += (A1 + A3 + 2) >> 2;
281 b[x + 1 + width2] = A1;
284 A3 = temp[width - 1];
286 A2 += (A1 + A3 + 2) >> 2;
291 lift(b + w2, temp + w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
292 lift(b, temp, b + w2, 1, 1, 1, width, 1, 2, 2, 0, 0);
296 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
301 for (i = 0; i < width; i++)
302 b1[i] -= (b0[i] + b2[i]) >> 1;
305 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
310 for (i = 0; i < width; i++)
311 b1[i] += (b0[i] + b2[i] + 2) >> 2;
314 static void spatial_decompose53i(DWTELEM *buffer, DWTELEM *temp,
315 int width, int height, int stride)
318 DWTELEM *b0 = buffer + mirror(-2 - 1, height - 1) * stride;
319 DWTELEM *b1 = buffer + mirror(-2, height - 1) * stride;
321 for (y = -2; y < height; y += 2) {
322 DWTELEM *b2 = buffer + mirror(y + 1, height - 1) * stride;
323 DWTELEM *b3 = buffer + mirror(y + 2, height - 1) * stride;
325 if (y + 1 < (unsigned)height)
326 horizontal_decompose53i(b2, temp, width);
327 if (y + 2 < (unsigned)height)
328 horizontal_decompose53i(b3, temp, width);
330 if (y + 1 < (unsigned)height)
331 vertical_decompose53iH0(b1, b2, b3, width);
332 if (y + 0 < (unsigned)height)
333 vertical_decompose53iL0(b0, b1, b2, width);
340 static void horizontal_decompose97i(DWTELEM *b, DWTELEM *temp, int width)
342 const int w2 = (width + 1) >> 1;
344 lift(temp + w2, b + 1, b, 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1);
345 liftS(temp, b, temp + w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0);
346 lift(b + w2, temp + w2, temp, 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
347 lift(b, temp, b + w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0);
350 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
355 for (i = 0; i < width; i++)
356 b1[i] -= (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
359 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
364 for (i = 0; i < width; i++)
365 b1[i] += (W_CM * (b0[i] + b2[i]) + W_CO) >> W_CS;
368 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
373 for (i = 0; i < width; i++)
375 b1[i] -= (W_BM * (b0[i] + b2[i]) + W_BO) >> W_BS;
377 b1[i] = (16 * 4 * b1[i] - 4 * (b0[i] + b2[i]) + W_BO * 5 + (5 << 27)) /
378 (5 * 16) - (1 << 23);
382 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2,
387 for (i = 0; i < width; i++)
388 b1[i] += (W_DM * (b0[i] + b2[i]) + W_DO) >> W_DS;
391 static void spatial_decompose97i(DWTELEM *buffer, DWTELEM *temp,
392 int width, int height, int stride)
395 DWTELEM *b0 = buffer + mirror(-4 - 1, height - 1) * stride;
396 DWTELEM *b1 = buffer + mirror(-4, height - 1) * stride;
397 DWTELEM *b2 = buffer + mirror(-4 + 1, height - 1) * stride;
398 DWTELEM *b3 = buffer + mirror(-4 + 2, height - 1) * stride;
400 for (y = -4; y < height; y += 2) {
401 DWTELEM *b4 = buffer + mirror(y + 3, height - 1) * stride;
402 DWTELEM *b5 = buffer + mirror(y + 4, height - 1) * stride;
404 if (y + 3 < (unsigned)height)
405 horizontal_decompose97i(b4, temp, width);
406 if (y + 4 < (unsigned)height)
407 horizontal_decompose97i(b5, temp, width);
409 if (y + 3 < (unsigned)height)
410 vertical_decompose97iH0(b3, b4, b5, width);
411 if (y + 2 < (unsigned)height)
412 vertical_decompose97iL0(b2, b3, b4, width);
413 if (y + 1 < (unsigned)height)
414 vertical_decompose97iH1(b1, b2, b3, width);
415 if (y + 0 < (unsigned)height)
416 vertical_decompose97iL1(b0, b1, b2, width);
425 void ff_spatial_dwt(DWTELEM *buffer, DWTELEM *temp, int width, int height,
426 int stride, int type, int decomposition_count)
430 for (level = 0; level < decomposition_count; level++) {
433 spatial_decompose97i(buffer, temp,
434 width >> level, height >> level,
438 spatial_decompose53i(buffer, temp,
439 width >> level, height >> level,
446 static void horizontal_compose53i(IDWTELEM *b, IDWTELEM *temp, int width)
448 const int width2 = width >> 1;
449 const int w2 = (width + 1) >> 1;
452 for (x = 0; x < width2; x++) {
454 temp[2 * x + 1] = b[x + w2];
459 b[0] = temp[0] - ((temp[1] + 1) >> 1);
460 for (x = 2; x < width - 1; x += 2) {
461 b[x] = temp[x] - ((temp[x - 1] + temp[x + 1] + 2) >> 2);
462 b[x - 1] = temp[x - 1] + ((b[x - 2] + b[x] + 1) >> 1);
465 b[x] = temp[x] - ((temp[x - 1] + 1) >> 1);
466 b[x - 1] = temp[x - 1] + ((b[x - 2] + b[x] + 1) >> 1);
468 b[x - 1] = temp[x - 1] + b[x - 2];
471 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
476 for (i = 0; i < width; i++)
477 b1[i] += (b0[i] + b2[i]) >> 1;
480 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
485 for (i = 0; i < width; i++)
486 b1[i] -= (b0[i] + b2[i] + 2) >> 2;
489 static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer *sb,
490 int height, int stride_line)
492 cs->b0 = slice_buffer_get_line(sb,
493 mirror(-1 - 1, height - 1) * stride_line);
494 cs->b1 = slice_buffer_get_line(sb, mirror(-1, height - 1) * stride_line);
498 static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer,
499 int height, int stride)
501 cs->b0 = buffer + mirror(-1 - 1, height - 1) * stride;
502 cs->b1 = buffer + mirror(-1, height - 1) * stride;
506 static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer *sb,
508 int width, int height,
513 IDWTELEM *b0 = cs->b0;
514 IDWTELEM *b1 = cs->b1;
515 IDWTELEM *b2 = slice_buffer_get_line(sb,
516 mirror(y + 1, height - 1) *
518 IDWTELEM *b3 = slice_buffer_get_line(sb,
519 mirror(y + 2, height - 1) *
522 if (y + 1 < (unsigned)height && y < (unsigned)height) {
525 for (x = 0; x < width; x++) {
526 b2[x] -= (b1[x] + b3[x] + 2) >> 2;
527 b1[x] += (b0[x] + b2[x]) >> 1;
530 if (y + 1 < (unsigned)height)
531 vertical_compose53iL0(b1, b2, b3, width);
532 if (y + 0 < (unsigned)height)
533 vertical_compose53iH0(b0, b1, b2, width);
536 if (y - 1 < (unsigned)height)
537 horizontal_compose53i(b0, temp, width);
538 if (y + 0 < (unsigned)height)
539 horizontal_compose53i(b1, temp, width);
546 static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer,
547 IDWTELEM *temp, int width, int height,
551 IDWTELEM *b0 = cs->b0;
552 IDWTELEM *b1 = cs->b1;
553 IDWTELEM *b2 = buffer + mirror(y + 1, height - 1) * stride;
554 IDWTELEM *b3 = buffer + mirror(y + 2, height - 1) * stride;
556 if (y + 1 < (unsigned)height)
557 vertical_compose53iL0(b1, b2, b3, width);
558 if (y + 0 < (unsigned)height)
559 vertical_compose53iH0(b0, b1, b2, width);
561 if (y - 1 < (unsigned)height)
562 horizontal_compose53i(b0, temp, width);
563 if (y + 0 < (unsigned)height)
564 horizontal_compose53i(b1, temp, width);
571 static void av_unused spatial_compose53i(IDWTELEM *buffer, IDWTELEM *temp,
572 int width, int height, int stride)
575 spatial_compose53i_init(&cs, buffer, height, stride);
576 while (cs.y <= height)
577 spatial_compose53i_dy(&cs, buffer, temp, width, height, stride);
580 void ff_snow_horizontal_compose97i(IDWTELEM *b, IDWTELEM *temp, int width)
582 const int w2 = (width + 1) >> 1;
584 #if 0 //maybe more understadable but slower
585 inv_lift(temp, b, b + w2, 2, 1, 1, width, W_DM, W_DO, W_DS, 0, 1);
586 inv_lift(temp + 1, b + w2, temp, 2, 1, 2, width, W_CM, W_CO, W_CS, 1, 1);
588 inv_liftS(b, temp, temp + 1, 2, 2, 2, width, W_BM, W_BO, W_BS, 0, 1);
589 inv_lift(b + 1, temp + 1, b, 2, 2, 2, width, W_AM, W_AO, W_AS, 1, 0);
592 temp[0] = b[0] - ((3 * b[w2] + 2) >> 2);
593 for (x = 1; x < (width >> 1); x++) {
594 temp[2 * x] = b[x] - ((3 * (b[x + w2 - 1] + b[x + w2]) + 4) >> 3);
595 temp[2 * x - 1] = b[x + w2 - 1] - temp[2 * x - 2] - temp[2 * x];
598 temp[2 * x] = b[x] - ((3 * b[x + w2 - 1] + 2) >> 2);
599 temp[2 * x - 1] = b[x + w2 - 1] - temp[2 * x - 2] - temp[2 * x];
601 temp[2 * x - 1] = b[x + w2 - 1] - 2 * temp[2 * x - 2];
603 b[0] = temp[0] + ((2 * temp[0] + temp[1] + 4) >> 3);
604 for (x = 2; x < width - 1; x += 2) {
605 b[x] = temp[x] + ((4 * temp[x] + temp[x - 1] + temp[x + 1] + 8) >> 4);
606 b[x - 1] = temp[x - 1] + ((3 * (b[x - 2] + b[x])) >> 1);
609 b[x] = temp[x] + ((2 * temp[x] + temp[x - 1] + 4) >> 3);
610 b[x - 1] = temp[x - 1] + ((3 * (b[x - 2] + b[x])) >> 1);
612 b[x - 1] = temp[x - 1] + 3 * b[x - 2];
616 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
621 for (i = 0; i < width; i++)
622 b1[i] += (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
625 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
630 for (i = 0; i < width; i++)
631 b1[i] -= (W_CM * (b0[i] + b2[i]) + W_CO) >> W_CS;
634 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
639 for (i = 0; i < width; i++)
641 b1[i] += (W_BM * (b0[i] + b2[i]) + W_BO) >> W_BS;
643 b1[i] += (W_BM * (b0[i] + b2[i]) + 4 * b1[i] + W_BO) >> W_BS;
647 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
652 for (i = 0; i < width; i++)
653 b1[i] -= (W_DM * (b0[i] + b2[i]) + W_DO) >> W_DS;
656 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
657 IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5,
662 for (i = 0; i < width; i++) {
663 b4[i] -= (W_DM * (b3[i] + b5[i]) + W_DO) >> W_DS;
664 b3[i] -= (W_CM * (b2[i] + b4[i]) + W_CO) >> W_CS;
666 b2[i] += (W_BM * (b1[i] + b3[i]) + W_BO) >> W_BS;
668 b2[i] += (W_BM * (b1[i] + b3[i]) + 4 * b2[i] + W_BO) >> W_BS;
670 b1[i] += (W_AM * (b0[i] + b2[i]) + W_AO) >> W_AS;
674 static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer *sb,
675 int height, int stride_line)
677 cs->b0 = slice_buffer_get_line(sb, mirror(-3 - 1, height - 1) * stride_line);
678 cs->b1 = slice_buffer_get_line(sb, mirror(-3, height - 1) * stride_line);
679 cs->b2 = slice_buffer_get_line(sb, mirror(-3 + 1, height - 1) * stride_line);
680 cs->b3 = slice_buffer_get_line(sb, mirror(-3 + 2, height - 1) * stride_line);
684 static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height,
687 cs->b0 = buffer + mirror(-3 - 1, height - 1) * stride;
688 cs->b1 = buffer + mirror(-3, height - 1) * stride;
689 cs->b2 = buffer + mirror(-3 + 1, height - 1) * stride;
690 cs->b3 = buffer + mirror(-3 + 2, height - 1) * stride;
694 static void spatial_compose97i_dy_buffered(DWTContext *dsp, DWTCompose *cs,
695 slice_buffer * sb, IDWTELEM *temp,
696 int width, int height,
701 IDWTELEM *b0 = cs->b0;
702 IDWTELEM *b1 = cs->b1;
703 IDWTELEM *b2 = cs->b2;
704 IDWTELEM *b3 = cs->b3;
705 IDWTELEM *b4 = slice_buffer_get_line(sb,
706 mirror(y + 3, height - 1) *
708 IDWTELEM *b5 = slice_buffer_get_line(sb,
709 mirror(y + 4, height - 1) *
712 if (y > 0 && y + 4 < height) {
713 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
715 if (y + 3 < (unsigned)height)
716 vertical_compose97iL1(b3, b4, b5, width);
717 if (y + 2 < (unsigned)height)
718 vertical_compose97iH1(b2, b3, b4, width);
719 if (y + 1 < (unsigned)height)
720 vertical_compose97iL0(b1, b2, b3, width);
721 if (y + 0 < (unsigned)height)
722 vertical_compose97iH0(b0, b1, b2, width);
725 if (y - 1 < (unsigned)height)
726 dsp->horizontal_compose97i(b0, temp, width);
727 if (y + 0 < (unsigned)height)
728 dsp->horizontal_compose97i(b1, temp, width);
737 static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer,
738 IDWTELEM *temp, int width, int height,
742 IDWTELEM *b0 = cs->b0;
743 IDWTELEM *b1 = cs->b1;
744 IDWTELEM *b2 = cs->b2;
745 IDWTELEM *b3 = cs->b3;
746 IDWTELEM *b4 = buffer + mirror(y + 3, height - 1) * stride;
747 IDWTELEM *b5 = buffer + mirror(y + 4, height - 1) * stride;
749 if (y + 3 < (unsigned)height)
750 vertical_compose97iL1(b3, b4, b5, width);
751 if (y + 2 < (unsigned)height)
752 vertical_compose97iH1(b2, b3, b4, width);
753 if (y + 1 < (unsigned)height)
754 vertical_compose97iL0(b1, b2, b3, width);
755 if (y + 0 < (unsigned)height)
756 vertical_compose97iH0(b0, b1, b2, width);
758 if (y - 1 < (unsigned)height)
759 ff_snow_horizontal_compose97i(b0, temp, width);
760 if (y + 0 < (unsigned)height)
761 ff_snow_horizontal_compose97i(b1, temp, width);
770 static void av_unused spatial_compose97i(IDWTELEM *buffer, IDWTELEM *temp,
771 int width, int height, int stride)
774 spatial_compose97i_init(&cs, buffer, height, stride);
775 while (cs.y <= height)
776 spatial_compose97i_dy(&cs, buffer, temp, width, height, stride);
779 void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer *sb, int width,
780 int height, int stride_line, int type,
781 int decomposition_count)
784 for (level = decomposition_count - 1; level >= 0; level--) {
787 spatial_compose97i_buffered_init(cs + level, sb, height >> level,
788 stride_line << level);
791 spatial_compose53i_buffered_init(cs + level, sb, height >> level,
792 stride_line << level);
798 void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs,
799 slice_buffer *slice_buf, IDWTELEM *temp,
800 int width, int height, int stride_line,
801 int type, int decomposition_count, int y)
803 const int support = type == 1 ? 3 : 5;
808 for (level = decomposition_count - 1; level >= 0; level--)
809 while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
812 spatial_compose97i_dy_buffered(dsp, cs + level, slice_buf, temp,
815 stride_line << level);
818 spatial_compose53i_dy_buffered(cs + level, slice_buf, temp,
821 stride_line << level);
827 static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width,
828 int height, int stride, int type,
829 int decomposition_count)
832 for (level = decomposition_count - 1; level >= 0; level--) {
835 spatial_compose97i_init(cs + level, buffer, height >> level,
839 spatial_compose53i_init(cs + level, buffer, height >> level,
846 static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer,
847 IDWTELEM *temp, int width, int height,
848 int stride, int type,
849 int decomposition_count, int y)
851 const int support = type == 1 ? 3 : 5;
856 for (level = decomposition_count - 1; level >= 0; level--)
857 while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
860 spatial_compose97i_dy(cs + level, buffer, temp, width >> level,
861 height >> level, stride << level);
864 spatial_compose53i_dy(cs + level, buffer, temp, width >> level,
865 height >> level, stride << level);
871 void ff_spatial_idwt(IDWTELEM *buffer, IDWTELEM *temp, int width, int height,
872 int stride, int type, int decomposition_count)
874 DWTCompose cs[MAX_DECOMPOSITIONS];
876 ff_spatial_idwt_init(cs, buffer, width, height, stride, type,
877 decomposition_count);
878 for (y = 0; y < height; y += 4)
879 ff_spatial_idwt_slice(cs, buffer, temp, width, height, stride, type,
880 decomposition_count, y);
883 static inline int w_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size,
884 int w, int h, int type)
887 const int dec_count = w == 8 ? 3 : 4;
888 int tmp[32 * 32], tmp2[32];
890 static const int scale[2][2][4][4] = {
893 { 268, 239, 239, 213 },
894 { 0, 224, 224, 152 },
895 { 0, 135, 135, 110 },
897 { // 9/7 16x16 or 32x32 dec=4
898 { 344, 310, 310, 280 },
899 { 0, 320, 320, 228 },
900 { 0, 175, 175, 136 },
901 { 0, 129, 129, 102 },
906 { 275, 245, 245, 218 },
907 { 0, 230, 230, 156 },
908 { 0, 138, 138, 113 },
910 { // 5/3 16x16 or 32x32 dec=4
911 { 352, 317, 317, 286 },
912 { 0, 328, 328, 233 },
913 { 0, 180, 180, 140 },
914 { 0, 132, 132, 105 },
919 for (i = 0; i < h; i++) {
920 for (j = 0; j < w; j += 4) {
921 tmp[32 * i + j + 0] = (pix1[j + 0] - pix2[j + 0]) << 4;
922 tmp[32 * i + j + 1] = (pix1[j + 1] - pix2[j + 1]) << 4;
923 tmp[32 * i + j + 2] = (pix1[j + 2] - pix2[j + 2]) << 4;
924 tmp[32 * i + j + 3] = (pix1[j + 3] - pix2[j + 3]) << 4;
930 ff_spatial_dwt(tmp, tmp2, w, h, 32, type, dec_count);
934 for (level = 0; level < dec_count; level++)
935 for (ori = level ? 1 : 0; ori < 4; ori++) {
936 int size = w >> (dec_count - level);
937 int sx = (ori & 1) ? size : 0;
938 int stride = 32 << (dec_count - level);
939 int sy = (ori & 2) ? stride >> 1 : 0;
941 for (i = 0; i < size; i++)
942 for (j = 0; j < size; j++) {
943 int v = tmp[sx + sy + i * stride + j] *
944 scale[type][dec_count - 3][level][ori];
952 static int w53_8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
954 return w_c(v, pix1, pix2, line_size, 8, h, 1);
957 static int w97_8_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
959 return w_c(v, pix1, pix2, line_size, 8, h, 0);
962 static int w53_16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
964 return w_c(v, pix1, pix2, line_size, 16, h, 1);
967 static int w97_16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
969 return w_c(v, pix1, pix2, line_size, 16, h, 0);
972 int ff_w53_32_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
974 return w_c(v, pix1, pix2, line_size, 32, h, 1);
977 int ff_w97_32_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
979 return w_c(v, pix1, pix2, line_size, 32, h, 0);
982 void ff_dsputil_init_dwt(DSPContext *c)
984 c->w53[0] = w53_16_c;
986 c->w97[0] = w97_16_c;
990 void ff_dwt_init(DWTContext *c)
992 c->vertical_compose97i = ff_snow_vertical_compose97i;
993 c->horizontal_compose97i = ff_snow_horizontal_compose97i;
994 c->inner_add_yblock = ff_snow_inner_add_yblock;