FFmpeg
rpzaenc.c
Go to the documentation of this file.
1 /*
2  * QuickTime RPZA Video Encoder
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg 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.
10  *
11  * FFmpeg 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.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 /**
22  * @file rpzaenc.c
23  * QT RPZA Video Encoder by Todd Kirby <doubleshot@pacbell.net> and David Adler
24  */
25 
26 #include "libavutil/avassert.h"
27 #include "libavutil/common.h"
28 #include "libavutil/opt.h"
29 
30 #include "avcodec.h"
31 #include "internal.h"
32 #include "put_bits.h"
33 
34 typedef struct RpzaContext {
36 
41 
42  AVFrame *prev_frame; // buffer for previous source frame
43  PutBitContext pb; // buffer for encoded frame data.
44 
45  int frame_width; // width in pixels of source frame
46  int frame_height; // height in pixesl of source frame
47 
48  int first_frame; // flag set to one when the first frame is being processed
49  // so that comparisons with previous frame data in not attempted
50 } RpzaContext;
51 
52 typedef enum channel_offset {
53  RED = 2,
54  GREEN = 1,
55  BLUE = 0,
57 
58 typedef struct rgb {
62 } rgb;
63 
64 #define SQR(x) ((x) * (x))
65 
66 /* 15 bit components */
67 #define GET_CHAN(color, chan) (((color) >> ((chan) * 5) & 0x1F) * 8)
68 #define R(color) GET_CHAN(color, RED)
69 #define G(color) GET_CHAN(color, GREEN)
70 #define B(color) GET_CHAN(color, BLUE)
71 
72 typedef struct BlockInfo {
73  int row;
74  int col;
80  uint16_t start;
81  int rowstride;
84 } BlockInfo;
85 
86 static void get_colors(uint8_t *min, uint8_t *max, uint8_t color4[4][3])
87 {
88  uint8_t step;
89 
90  color4[0][0] = min[0];
91  color4[0][1] = min[1];
92  color4[0][2] = min[2];
93 
94  color4[3][0] = max[0];
95  color4[3][1] = max[1];
96  color4[3][2] = max[2];
97 
98  // red components
99  step = (color4[3][0] - color4[0][0] + 1) / 3;
100  color4[1][0] = color4[0][0] + step;
101  color4[2][0] = color4[3][0] - step;
102 
103  // green components
104  step = (color4[3][1] - color4[0][1] + 1) / 3;
105  color4[1][1] = color4[0][1] + step;
106  color4[2][1] = color4[3][1] - step;
107 
108  // blue components
109  step = (color4[3][2] - color4[0][2] + 1) / 3;
110  color4[1][2] = color4[0][2] + step;
111  color4[2][2] = color4[3][2] - step;
112 }
113 
114 /* Fill BlockInfo struct with information about a 4x4 block of the image */
115 static int get_block_info(BlockInfo *bi, int block)
116 {
117  bi->row = block / bi->blocks_per_row;
118  bi->col = block % bi->blocks_per_row;
119 
120  // test for right edge block
121  if (bi->col == bi->blocks_per_row - 1 && (bi->image_width % 4) != 0) {
122  bi->block_width = bi->image_width % 4;
123  } else {
124  bi->block_width = 4;
125  }
126 
127  // test for bottom edge block
128  if (bi->row == (bi->image_height / 4) && (bi->image_height % 4) != 0) {
129  bi->block_height = bi->image_height % 4;
130  } else {
131  bi->block_height = 4;
132  }
133 
134  return block ? (bi->col * 4) + (bi->row * bi->rowstride * 4) : 0;
135 }
136 
137 static uint16_t rgb24_to_rgb555(uint8_t *rgb24)
138 {
139  uint16_t rgb555 = 0;
140  uint32_t r, g, b;
141 
142  r = rgb24[0] >> 3;
143  g = rgb24[1] >> 3;
144  b = rgb24[2] >> 3;
145 
146  rgb555 |= (r << 10);
147  rgb555 |= (g << 5);
148  rgb555 |= (b << 0);
149 
150  return rgb555;
151 }
152 
153 /*
154  * Returns the total difference between two 24 bit color values
155  */
156 static int diff_colors(uint8_t *colorA, uint8_t *colorB)
157 {
158  int tot;
159 
160  tot = SQR(colorA[0] - colorB[0]);
161  tot += SQR(colorA[1] - colorB[1]);
162  tot += SQR(colorA[2] - colorB[2]);
163 
164  return tot;
165 }
166 
167 /*
168  * Returns the maximum channel difference
169  */
170 static int max_component_diff(uint16_t *colorA, uint16_t *colorB)
171 {
172  int diff, max = 0;
173 
174  diff = FFABS(R(colorA[0]) - R(colorB[0]));
175  if (diff > max) {
176  max = diff;
177  }
178  diff = FFABS(G(colorA[0]) - G(colorB[0]));
179  if (diff > max) {
180  max = diff;
181  }
182  diff = FFABS(B(colorA[0]) - B(colorB[0]));
183  if (diff > max) {
184  max = diff;
185  }
186  return max * 8;
187 }
188 
189 /*
190  * Find the channel that has the largest difference between minimum and maximum
191  * color values. Put the minimum value in min, maximum in max and the channel
192  * in chan.
193  */
194 static void get_max_component_diff(BlockInfo *bi, uint16_t *block_ptr,
196 {
197  int x, y;
198  uint8_t min_r, max_r, min_g, max_g, min_b, max_b;
199  uint8_t r, g, b;
200 
201  // fix warning about uninitialized vars
202  min_r = min_g = min_b = UINT8_MAX;
203  max_r = max_g = max_b = 0;
204 
205  // loop thru and compare pixels
206  for (y = 0; y < bi->block_height; y++) {
207  for (x = 0; x < bi->block_width; x++){
208  // TODO: optimize
209  min_r = FFMIN(R(block_ptr[x]), min_r);
210  min_g = FFMIN(G(block_ptr[x]), min_g);
211  min_b = FFMIN(B(block_ptr[x]), min_b);
212 
213  max_r = FFMAX(R(block_ptr[x]), max_r);
214  max_g = FFMAX(G(block_ptr[x]), max_g);
215  max_b = FFMAX(B(block_ptr[x]), max_b);
216  }
217  block_ptr += bi->rowstride;
218  }
219 
220  r = max_r - min_r;
221  g = max_g - min_g;
222  b = max_b - min_b;
223 
224  if (r > g && r > b) {
225  *max = max_r;
226  *min = min_r;
227  *chan = RED;
228  } else if (g > b && g >= r) {
229  *max = max_g;
230  *min = min_g;
231  *chan = GREEN;
232  } else {
233  *max = max_b;
234  *min = min_b;
235  *chan = BLUE;
236  }
237 }
238 
239 /*
240  * Compare two 4x4 blocks to determine if the total difference between the
241  * blocks is greater than the thresh parameter. Returns -1 if difference
242  * exceeds threshold or zero otherwise.
243  */
244 static int compare_blocks(uint16_t *block1, uint16_t *block2, BlockInfo *bi, int thresh)
245 {
246  int x, y, diff = 0;
247  for (y = 0; y < bi->block_height; y++) {
248  for (x = 0; x < bi->block_width; x++) {
249  diff = max_component_diff(&block1[x], &block2[x]);
250  if (diff >= thresh) {
251  return -1;
252  }
253  }
254  block1 += bi->rowstride;
255  block2 += bi->rowstride;
256  }
257  return 0;
258 }
259 
260 /*
261  * Determine the fit of one channel to another within a 4x4 block. This
262  * is used to determine the best palette choices for 4-color encoding.
263  */
264 static int leastsquares(uint16_t *block_ptr, BlockInfo *bi,
265  channel_offset xchannel, channel_offset ychannel,
266  double *slope, double *y_intercept, double *correlation_coef)
267 {
268  double sumx = 0, sumy = 0, sumx2 = 0, sumy2 = 0, sumxy = 0,
269  sumx_sq = 0, sumy_sq = 0, tmp, tmp2;
270  int i, j, count;
271  uint8_t x, y;
272 
273  count = bi->block_height * bi->block_width;
274 
275  if (count < 2)
276  return -1;
277 
278  for (i = 0; i < bi->block_height; i++) {
279  for (j = 0; j < bi->block_width; j++){
280  x = GET_CHAN(block_ptr[j], xchannel);
281  y = GET_CHAN(block_ptr[j], ychannel);
282  sumx += x;
283  sumy += y;
284  sumx2 += x * x;
285  sumy2 += y * y;
286  sumxy += x * y;
287  }
288  block_ptr += bi->rowstride;
289  }
290 
291  sumx_sq = sumx * sumx;
292  tmp = (count * sumx2 - sumx_sq);
293 
294  // guard against div/0
295  if (tmp == 0)
296  return -2;
297 
298  sumy_sq = sumy * sumy;
299 
300  *slope = (sumx * sumy - sumxy) / tmp;
301  *y_intercept = (sumy - (*slope) * sumx) / count;
302 
303  tmp2 = count * sumy2 - sumy_sq;
304  if (tmp2 == 0) {
305  *correlation_coef = 0.0;
306  } else {
307  *correlation_coef = (count * sumxy - sumx * sumy) /
308  sqrt(tmp * tmp2);
309  }
310 
311  return 0; // success
312 }
313 
314 /*
315  * Determine the amount of error in the leastsquares fit.
316  */
317 static int calc_lsq_max_fit_error(uint16_t *block_ptr, BlockInfo *bi,
318  int min, int max, int tmp_min, int tmp_max,
319  channel_offset xchannel, channel_offset ychannel)
320 {
321  int i, j, x, y;
322  int err;
323  int max_err = 0;
324 
325  for (i = 0; i < bi->block_height; i++) {
326  for (j = 0; j < bi->block_width; j++){
327  int x_inc, lin_y, lin_x;
328  x = GET_CHAN(block_ptr[j], xchannel);
329  y = GET_CHAN(block_ptr[j], ychannel);
330 
331  /* calculate x_inc as the 4-color index (0..3) */
332  x_inc = floor( (x - min) * 3.0 / (max - min) + 0.5);
333  x_inc = FFMAX(FFMIN(3, x_inc), 0);
334 
335  /* calculate lin_y corresponding to x_inc */
336  lin_y = (int)(tmp_min + (tmp_max - tmp_min) * x_inc / 3.0 + 0.5);
337 
338  err = FFABS(lin_y - y);
339  if (err > max_err)
340  max_err = err;
341 
342  /* calculate lin_x corresponding to x_inc */
343  lin_x = (int)(min + (max - min) * x_inc / 3.0 + 0.5);
344 
345  err = FFABS(lin_x - x);
346  if (err > max_err)
347  max_err += err;
348  }
349  block_ptr += bi->rowstride;
350  }
351 
352  return max_err;
353 }
354 
355 /*
356  * Find the closest match to a color within the 4-color palette
357  */
358 static int match_color(uint16_t *color, uint8_t colors[4][3])
359 {
360  int ret = 0;
361  int smallest_variance = INT_MAX;
362  uint8_t dithered_color[3];
363 
364  for (int channel = 0; channel < 3; channel++) {
365  dithered_color[channel] = GET_CHAN(color[0], channel);
366  }
367 
368  for (int palette_entry = 0; palette_entry < 4; palette_entry++) {
369  int variance = diff_colors(dithered_color, colors[palette_entry]);
370 
371  if (variance < smallest_variance) {
372  smallest_variance = variance;
373  ret = palette_entry;
374  }
375  }
376 
377  return ret;
378 }
379 
380 /*
381  * Encode a block using the 4-color opcode and palette. return number of
382  * blocks encoded (until we implement multi-block 4 color runs this will
383  * always be 1)
384  */
385 static int encode_four_color_block(uint8_t *min_color, uint8_t *max_color,
386  PutBitContext *pb, uint16_t *block_ptr, BlockInfo *bi)
387 {
388  int x, y, idx;
389  uint8_t color4[4][3];
390  uint16_t rounded_max, rounded_min;
391 
392  // round min and max wider
393  rounded_min = rgb24_to_rgb555(min_color);
394  rounded_max = rgb24_to_rgb555(max_color);
395 
396  // put a and b colors
397  // encode 4 colors = first 16 bit color with MSB zeroed and...
398  put_bits(pb, 16, rounded_max & ~0x8000);
399  // ...second 16 bit color with MSB on.
400  put_bits(pb, 16, rounded_min | 0x8000);
401 
402  get_colors(min_color, max_color, color4);
403 
404  for (y = 0; y < 4; y++) {
405  for (x = 0; x < 4; x++) {
406  idx = match_color(&block_ptr[x], color4);
407  put_bits(pb, 2, idx);
408  }
409  block_ptr += bi->rowstride;
410  }
411  return 1; // num blocks encoded
412 }
413 
414 /*
415  * Copy a 4x4 block from the current frame buffer to the previous frame buffer.
416  */
417 static void update_block_in_prev_frame(const uint16_t *src_pixels,
418  uint16_t *dest_pixels,
419  const BlockInfo *bi, int block_counter)
420 {
421  for (int y = 0; y < 4; y++) {
422  memcpy(dest_pixels, src_pixels, 8);
423  dest_pixels += bi->rowstride;
424  src_pixels += bi->rowstride;
425  }
426 }
427 
428 /*
429  * update statistics for the specified block. If first_block,
430  * it initializes the statistics. Otherwise it updates the statistics IF THIS
431  * BLOCK IS SUITABLE TO CONTINUE A 1-COLOR RUN. That is, it checks whether
432  * the range of colors (since the routine was called first_block != 0) are
433  * all close enough intensities to be represented by a single color.
434 
435  * The routine returns 0 if this block is too different to be part of
436  * the same run of 1-color blocks. The routine returns 1 if this
437  * block can be part of the same 1-color block run.
438 
439  * If the routine returns 1, it also updates its arguments to include
440  * the statistics of this block. Otherwise, the stats are unchanged
441  * and don't include the current block.
442  */
443 static int update_block_stats(RpzaContext *s, BlockInfo *bi, uint16_t *block,
444  uint8_t min_color[3], uint8_t max_color[3],
445  int *total_rgb, int *total_pixels,
446  uint8_t avg_color[3], int first_block)
447 {
448  int x, y;
449  int is_in_range;
450  int total_pixels_blk;
451  int threshold;
452 
453  uint8_t min_color_blk[3], max_color_blk[3];
454  int total_rgb_blk[3];
455  uint8_t avg_color_blk[3];
456 
457  if (first_block) {
458  min_color[0] = UINT8_MAX;
459  min_color[1] = UINT8_MAX;
460  min_color[2] = UINT8_MAX;
461  max_color[0] = 0;
462  max_color[1] = 0;
463  max_color[2] = 0;
464  total_rgb[0] = 0;
465  total_rgb[1] = 0;
466  total_rgb[2] = 0;
467  *total_pixels = 0;
468  threshold = s->start_one_color_thresh;
469  } else {
470  threshold = s->continue_one_color_thresh;
471  }
472 
473  /*
474  The *_blk variables will include the current block.
475  Initialize them based on the blocks so far.
476  */
477  min_color_blk[0] = min_color[0];
478  min_color_blk[1] = min_color[1];
479  min_color_blk[2] = min_color[2];
480  max_color_blk[0] = max_color[0];
481  max_color_blk[1] = max_color[1];
482  max_color_blk[2] = max_color[2];
483  total_rgb_blk[0] = total_rgb[0];
484  total_rgb_blk[1] = total_rgb[1];
485  total_rgb_blk[2] = total_rgb[2];
486  total_pixels_blk = *total_pixels + bi->block_height * bi->block_width;
487 
488  /*
489  Update stats for this block's pixels
490  */
491  for (y = 0; y < bi->block_height; y++) {
492  for (x = 0; x < bi->block_width; x++) {
493  total_rgb_blk[0] += R(block[x]);
494  total_rgb_blk[1] += G(block[x]);
495  total_rgb_blk[2] += B(block[x]);
496 
497  min_color_blk[0] = FFMIN(R(block[x]), min_color_blk[0]);
498  min_color_blk[1] = FFMIN(G(block[x]), min_color_blk[1]);
499  min_color_blk[2] = FFMIN(B(block[x]), min_color_blk[2]);
500 
501  max_color_blk[0] = FFMAX(R(block[x]), max_color_blk[0]);
502  max_color_blk[1] = FFMAX(G(block[x]), max_color_blk[1]);
503  max_color_blk[2] = FFMAX(B(block[x]), max_color_blk[2]);
504  }
505  block += bi->rowstride;
506  }
507 
508  /*
509  Calculate average color including current block.
510  */
511  avg_color_blk[0] = total_rgb_blk[0] / total_pixels_blk;
512  avg_color_blk[1] = total_rgb_blk[1] / total_pixels_blk;
513  avg_color_blk[2] = total_rgb_blk[2] / total_pixels_blk;
514 
515  /*
516  Are all the pixels within threshold of the average color?
517  */
518  is_in_range = (max_color_blk[0] - avg_color_blk[0] <= threshold &&
519  max_color_blk[1] - avg_color_blk[1] <= threshold &&
520  max_color_blk[2] - avg_color_blk[2] <= threshold &&
521  avg_color_blk[0] - min_color_blk[0] <= threshold &&
522  avg_color_blk[1] - min_color_blk[1] <= threshold &&
523  avg_color_blk[2] - min_color_blk[2] <= threshold);
524 
525  if (is_in_range) {
526  /*
527  Set the output variables to include this block.
528  */
529  min_color[0] = min_color_blk[0];
530  min_color[1] = min_color_blk[1];
531  min_color[2] = min_color_blk[2];
532  max_color[0] = max_color_blk[0];
533  max_color[1] = max_color_blk[1];
534  max_color[2] = max_color_blk[2];
535  total_rgb[0] = total_rgb_blk[0];
536  total_rgb[1] = total_rgb_blk[1];
537  total_rgb[2] = total_rgb_blk[2];
538  *total_pixels = total_pixels_blk;
539  avg_color[0] = avg_color_blk[0];
540  avg_color[1] = avg_color_blk[1];
541  avg_color[2] = avg_color_blk[2];
542  }
543 
544  return is_in_range;
545 }
546 
547 static void rpza_encode_stream(RpzaContext *s, const AVFrame *pict)
548 {
549  BlockInfo bi;
550  int block_counter = 0;
551  int n_blocks;
552  int total_blocks;
553  int prev_block_offset;
554  int block_offset = 0;
555  uint8_t min = 0, max = 0;
556  channel_offset chan;
557  int i;
558  int tmp_min, tmp_max;
559  int total_rgb[3];
560  uint8_t avg_color[3];
561  int pixel_count;
562  uint8_t min_color[3], max_color[3];
563  double slope, y_intercept, correlation_coef;
564  uint16_t *src_pixels = (uint16_t *)pict->data[0];
565  uint16_t *prev_pixels = (uint16_t *)s->prev_frame->data[0];
566 
567  /* Number of 4x4 blocks in frame. */
568  total_blocks = ((s->frame_width + 3) / 4) * ((s->frame_height + 3) / 4);
569 
570  bi.image_width = s->frame_width;
571  bi.image_height = s->frame_height;
572  bi.rowstride = pict->linesize[0] / 2;
573 
574  bi.blocks_per_row = (s->frame_width + 3) / 4;
575 
576  while (block_counter < total_blocks) {
577  // SKIP CHECK
578  // make sure we have a valid previous frame and we're not writing
579  // a key frame
580  if (!s->first_frame) {
581  n_blocks = 0;
582  prev_block_offset = 0;
583 
584  while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
585 
586  block_offset = get_block_info(&bi, block_counter + n_blocks);
587 
588  // multi-block opcodes cannot span multiple rows.
589  // If we're starting a new row, break out and write the opcode
590  /* TODO: Should eventually use bi.row here to determine when a
591  row break occurs, but that is currently breaking the
592  quicktime player. This is probably due to a bug in the
593  way I'm calculating the current row.
594  */
595  if (prev_block_offset && block_offset - prev_block_offset > 12) {
596  break;
597  }
598 
599  prev_block_offset = block_offset;
600 
601  if (compare_blocks(&prev_pixels[block_offset],
602  &src_pixels[block_offset], &bi, s->skip_frame_thresh) != 0) {
603  // write out skipable blocks
604  if (n_blocks) {
605 
606  // write skip opcode
607  put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
608  block_counter += n_blocks;
609 
610  goto post_skip;
611  }
612  break;
613  }
614 
615  /*
616  * NOTE: we don't update skipped blocks in the previous frame buffer
617  * since skipped needs always to be compared against the first skipped
618  * block to avoid artifacts during gradual fade in/outs.
619  */
620 
621  // update_block_in_prev_frame(&src_pixels[block_offset],
622  // &prev_pixels[block_offset], &bi, block_counter + n_blocks);
623 
624  n_blocks++;
625  }
626 
627  // we're either at the end of the frame or we've reached the maximum
628  // of 32 blocks in a run. Write out the run.
629  if (n_blocks) {
630  // write skip opcode
631  put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
632  block_counter += n_blocks;
633 
634  continue;
635  }
636 
637  } else {
638  block_offset = get_block_info(&bi, block_counter);
639  }
640 post_skip :
641 
642  // ONE COLOR CHECK
643  if (update_block_stats(s, &bi, &src_pixels[block_offset],
644  min_color, max_color,
645  total_rgb, &pixel_count, avg_color, 1)) {
646  prev_block_offset = block_offset;
647 
648  n_blocks = 1;
649 
650  /* update this block in the previous frame buffer */
651  update_block_in_prev_frame(&src_pixels[block_offset],
652  &prev_pixels[block_offset], &bi, block_counter + n_blocks);
653 
654  // check for subsequent blocks with the same color
655  while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
656  block_offset = get_block_info(&bi, block_counter + n_blocks);
657 
658  // multi-block opcodes cannot span multiple rows.
659  // If we've hit end of a row, break out and write the opcode
660  if (block_offset - prev_block_offset > 12) {
661  break;
662  }
663 
664  if (!update_block_stats(s, &bi, &src_pixels[block_offset],
665  min_color, max_color,
666  total_rgb, &pixel_count, avg_color, 0)) {
667  break;
668  }
669 
670  prev_block_offset = block_offset;
671 
672  /* update this block in the previous frame buffer */
673  update_block_in_prev_frame(&src_pixels[block_offset],
674  &prev_pixels[block_offset], &bi, block_counter + n_blocks);
675 
676  n_blocks++;
677  }
678 
679  // write one color opcode.
680  put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
681  // write color to encode.
682  put_bits(&s->pb, 16, rgb24_to_rgb555(avg_color));
683  // skip past the blocks we've just encoded.
684  block_counter += n_blocks;
685  } else { // FOUR COLOR CHECK
686  int err = 0;
687 
688  // get max component diff for block
689  get_max_component_diff(&bi, &src_pixels[block_offset], &min, &max, &chan);
690 
691  min_color[0] = 0;
692  max_color[0] = 0;
693  min_color[1] = 0;
694  max_color[1] = 0;
695  min_color[2] = 0;
696  max_color[2] = 0;
697 
698  // run least squares against other two components
699  for (i = 0; i < 3; i++) {
700  if (i == chan) {
701  min_color[i] = min;
702  max_color[i] = max;
703  continue;
704  }
705 
706  slope = y_intercept = correlation_coef = 0;
707 
708  if (leastsquares(&src_pixels[block_offset], &bi, chan, i,
709  &slope, &y_intercept, &correlation_coef)) {
710  min_color[i] = GET_CHAN(src_pixels[block_offset], i);
711  max_color[i] = GET_CHAN(src_pixels[block_offset], i);
712  } else {
713  tmp_min = (int)(0.5 + min * slope + y_intercept);
714  tmp_max = (int)(0.5 + max * slope + y_intercept);
715 
716  av_assert0(tmp_min <= tmp_max);
717  // clamp min and max color values
718  tmp_min = av_clip_uint8(tmp_min);
719  tmp_max = av_clip_uint8(tmp_max);
720 
721  err = FFMAX(calc_lsq_max_fit_error(&src_pixels[block_offset], &bi,
722  min, max, tmp_min, tmp_max, chan, i), err);
723 
724  min_color[i] = tmp_min;
725  max_color[i] = tmp_max;
726  }
727  }
728 
729  if (err > s->sixteen_color_thresh) { // DO SIXTEEN COLOR BLOCK
730  uint16_t *row_ptr;
731  int rgb555;
732 
733  block_offset = get_block_info(&bi, block_counter);
734 
735  row_ptr = &src_pixels[block_offset];
736 
737  for (int y = 0; y < 4; y++) {
738  for (int x = 0; x < 4; x++){
739  rgb555 = row_ptr[x] & ~0x8000;
740 
741  put_bits(&s->pb, 16, rgb555);
742  }
743  row_ptr += bi.rowstride;
744  }
745 
746  block_counter++;
747  } else { // FOUR COLOR BLOCK
748  block_counter += encode_four_color_block(min_color, max_color,
749  &s->pb, &src_pixels[block_offset], &bi);
750  }
751 
752  /* update this block in the previous frame buffer */
753  update_block_in_prev_frame(&src_pixels[block_offset],
754  &prev_pixels[block_offset], &bi, block_counter);
755  }
756  }
757 }
758 
760 {
761  RpzaContext *s = avctx->priv_data;
762 
763  s->frame_width = avctx->width;
764  s->frame_height = avctx->height;
765 
766  s->prev_frame = av_frame_alloc();
767  if (!s->prev_frame)
768  return AVERROR(ENOMEM);
769 
770  return 0;
771 }
772 
774  const AVFrame *frame, int *got_packet)
775 {
776  RpzaContext *s = avctx->priv_data;
777  const AVFrame *pict = frame;
778  uint8_t *buf;
779  int ret;
780 
781  if ((ret = ff_alloc_packet2(avctx, pkt, 6LL * avctx->height * avctx->width, 0)) < 0)
782  return ret;
783 
784  init_put_bits(&s->pb, pkt->data, pkt->size);
785 
786  // skip 4 byte header, write it later once the size of the chunk is known
787  put_bits32(&s->pb, 0x00);
788 
789  if (!s->prev_frame->data[0]) {
790  s->first_frame = 1;
791  s->prev_frame->format = pict->format;
792  s->prev_frame->width = pict->width;
793  s->prev_frame->height = pict->height;
794  ret = av_frame_get_buffer(s->prev_frame, 0);
795  if (ret < 0)
796  return ret;
797  } else {
798  s->first_frame = 0;
799  }
800 
801  rpza_encode_stream(s, pict);
802 
803  flush_put_bits(&s->pb);
804 
805  av_shrink_packet(pkt, put_bits_count(&s->pb) >> 3);
806  buf = pkt->data;
807 
808  // write header opcode
809  buf[0] = 0xe1; // chunk opcode
810 
811  // write chunk length
812  AV_WB24(buf + 1, pkt->size);
813 
814  *got_packet = 1;
815 
816  return 0;
817 }
818 
819 static int rpza_encode_end(AVCodecContext *avctx)
820 {
821  RpzaContext *s = (RpzaContext *)avctx->priv_data;
822 
823  av_frame_free(&s->prev_frame);
824 
825  return 0;
826 }
827 
828 #define OFFSET(x) offsetof(RpzaContext, x)
829 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
830 static const AVOption options[] = {
831  { "skip_frame_thresh", NULL, OFFSET(skip_frame_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
832  { "start_one_color_thresh", NULL, OFFSET(start_one_color_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
833  { "continue_one_color_thresh", NULL, OFFSET(continue_one_color_thresh), AV_OPT_TYPE_INT, {.i64=0}, 0, 24, VE},
834  { "sixteen_color_thresh", NULL, OFFSET(sixteen_color_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
835  { NULL },
836 };
837 
838 static const AVClass rpza_class = {
839  .class_name = "rpza",
840  .item_name = av_default_item_name,
841  .option = options,
842  .version = LIBAVUTIL_VERSION_INT,
843 };
844 
846  .name = "rpza",
847  .long_name = NULL_IF_CONFIG_SMALL("QuickTime video (RPZA)"),
848  .type = AVMEDIA_TYPE_VIDEO,
849  .id = AV_CODEC_ID_RPZA,
850  .priv_data_size = sizeof(RpzaContext),
851  .priv_class = &rpza_class,
853  .encode2 = rpza_encode_frame,
854  .close = rpza_encode_end,
855  .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
856  .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_RGB555,
858 };
OFFSET
#define OFFSET(x)
Definition: rpzaenc.c:828
AVCodec
AVCodec.
Definition: codec.h:197
rgb::b
uint8_t b
Definition: rpzaenc.c:61
FF_CODEC_CAP_INIT_THREADSAFE
#define FF_CODEC_CAP_INIT_THREADSAFE
The codec does not modify any global variables in the init function, allowing to call the init functi...
Definition: internal.h:41
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
RED
@ RED
Definition: rpzaenc.c:53
init
static av_cold int init(AVCodecContext *avctx)
Definition: avrndec.c:31
r
const char * r
Definition: vf_curves.c:116
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
opt.h
put_bits32
static void av_unused put_bits32(PutBitContext *s, uint32_t value)
Write exactly 32 bits into a bitstream.
Definition: put_bits.h:263
av_frame_get_buffer
int av_frame_get_buffer(AVFrame *frame, int align)
Allocate new buffer(s) for audio or video data.
Definition: frame.c:337
color
Definition: vf_paletteuse.c:583
BlockInfo::block_index
int block_index
Definition: rpzaenc.c:79
init_put_bits
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
Definition: put_bits.h:57
av_frame_free
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:203
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:318
put_bits
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
Definition: j2kenc.c:218
tmp
static uint8_t tmp[11]
Definition: aes_ctr.c:27
step
trying all byte sequences megabyte in length and selecting the best looking sequence will yield cases to try But a word about which is also called distortion Distortion can be quantified by almost any quality measurement one chooses the sum of squared differences is used but more complex methods that consider psychovisual effects can be used as well It makes no difference in this discussion First step
Definition: rate_distortion.txt:58
AVFrame::width
int width
Definition: frame.h:376
RpzaContext::first_frame
int first_frame
Definition: rpzaenc.c:48
internal.h
AVPacket::data
uint8_t * data
Definition: packet.h:369
BlockInfo::start
uint16_t start
Definition: rpzaenc.c:80
AVOption
AVOption.
Definition: opt.h:248
b
#define b
Definition: input.c:41
RpzaContext::skip_frame_thresh
int skip_frame_thresh
Definition: rpzaenc.c:37
B
#define B(color)
Definition: rpzaenc.c:70
RpzaContext
Definition: rpza.c:47
max
#define max(a, b)
Definition: cuda_runtime.h:33
BlockInfo::row
int row
Definition: rpzaenc.c:73
BlockInfo::rowstride
int rowstride
Definition: rpzaenc.c:81
SQR
#define SQR(x)
Definition: rpzaenc.c:64
BlockInfo::image_width
int image_width
Definition: rpzaenc.c:77
GREEN
@ GREEN
Definition: rpzaenc.c:54
AVFrame::data
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:332
rgb
Definition: rpzaenc.c:58
av_shrink_packet
void av_shrink_packet(AVPacket *pkt, int size)
Reduce packet size, correctly zeroing padding.
Definition: avpacket.c:114
get_colors
static void get_colors(uint8_t *min, uint8_t *max, uint8_t color4[4][3])
Definition: rpzaenc.c:86
BlockInfo::col
int col
Definition: rpzaenc.c:74
ff_rpza_encoder
AVCodec ff_rpza_encoder
Definition: rpzaenc.c:845
get_block_info
static int get_block_info(BlockInfo *bi, int block)
Definition: rpzaenc.c:115
av_frame_alloc
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
Definition: frame.c:190
update_block_in_prev_frame
static void update_block_in_prev_frame(const uint16_t *src_pixels, uint16_t *dest_pixels, const BlockInfo *bi, int block_counter)
Definition: rpzaenc.c:417
avassert.h
pkt
AVPacket * pkt
Definition: movenc.c:59
BlockInfo::image_height
int image_height
Definition: rpzaenc.c:78
compare_blocks
static int compare_blocks(uint16_t *block1, uint16_t *block2, BlockInfo *bi, int thresh)
Definition: rpzaenc.c:244
s
#define s(width, name)
Definition: cbs_vp9.c:257
encode_four_color_block
static int encode_four_color_block(uint8_t *min_color, uint8_t *max_color, PutBitContext *pb, uint16_t *block_ptr, BlockInfo *bi)
Definition: rpzaenc.c:385
floor
static __device__ float floor(float a)
Definition: cuda_runtime.h:173
g
const char * g
Definition: vf_curves.c:117
rpza_encode_end
static int rpza_encode_end(AVCodecContext *avctx)
Definition: rpzaenc.c:819
rpza_encode_init
static int rpza_encode_init(AVCodecContext *avctx)
Definition: rpzaenc.c:759
av_assert0
#define av_assert0(cond)
assert() equivalent, that is always enabled.
Definition: avassert.h:37
RpzaContext::avclass
AVClass * avclass
Definition: rpzaenc.c:35
pix_fmts
static enum AVPixelFormat pix_fmts[]
Definition: libkvazaar.c:309
BlockInfo::block_height
int block_height
Definition: rpzaenc.c:76
G
#define G(color)
Definition: rpzaenc.c:69
PutBitContext
Definition: put_bits.h:44
BlockInfo::blocks_per_row
int blocks_per_row
Definition: rpzaenc.c:82
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:72
BLUE
@ BLUE
Definition: rpzaenc.c:55
LIBAVUTIL_VERSION_INT
#define LIBAVUTIL_VERSION_INT
Definition: version.h:85
AVClass
Describe the class of an AVClass context structure.
Definition: log.h:67
NULL
#define NULL
Definition: coverity.c:32
rpza_class
static const AVClass rpza_class
Definition: rpzaenc.c:838
av_default_item_name
const char * av_default_item_name(void *ptr)
Return the context name.
Definition: log.c:235
VE
#define VE
Definition: rpzaenc.c:829
BlockInfo
Definition: dvdec.c:54
leastsquares
static int leastsquares(uint16_t *block_ptr, BlockInfo *bi, channel_offset xchannel, channel_offset ychannel, double *slope, double *y_intercept, double *correlation_coef)
Definition: rpzaenc.c:264
AVPacket::size
int size
Definition: packet.h:370
NULL_IF_CONFIG_SMALL
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
FFMAX
#define FFMAX(a, b)
Definition: common.h:103
RpzaContext::start_one_color_thresh
int start_one_color_thresh
Definition: rpzaenc.c:38
AV_WB24
#define AV_WB24(p, d)
Definition: intreadwrite.h:450
AVFrame::format
int format
format of the frame, -1 if unknown or unset Values correspond to enum AVPixelFormat for video frames,...
Definition: frame.h:391
calc_lsq_max_fit_error
static int calc_lsq_max_fit_error(uint16_t *block_ptr, BlockInfo *bi, int min, int max, int tmp_min, int tmp_max, channel_offset xchannel, channel_offset ychannel)
Definition: rpzaenc.c:317
RpzaContext::frame_height
int frame_height
Definition: rpzaenc.c:46
FFMIN
#define FFMIN(a, b)
Definition: common.h:105
rgb::g
uint8_t g
Definition: rpzaenc.c:60
channel_offset
channel_offset
Definition: rpzaenc.c:52
i
int i
Definition: input.c:407
RpzaContext::pb
PutBitContext pb
Definition: rpzaenc.c:43
put_bits_count
static int put_bits_count(PutBitContext *s)
Definition: put_bits.h:76
common.h
AV_PIX_FMT_RGB555
#define AV_PIX_FMT_RGB555
Definition: pixfmt.h:387
uint8_t
uint8_t
Definition: audio_convert.c:194
rgb::r
uint8_t r
Definition: rpzaenc.c:59
AVCodec::name
const char * name
Name of the codec implementation.
Definition: codec.h:204
AVCodecContext::height
int height
Definition: avcodec.h:709
get_max_component_diff
static void get_max_component_diff(BlockInfo *bi, uint16_t *block_ptr, uint8_t *min, uint8_t *max, channel_offset *chan)
Definition: rpzaenc.c:194
avcodec.h
ret
ret
Definition: filter_design.txt:187
rgb24_to_rgb555
static uint16_t rgb24_to_rgb555(uint8_t *rgb24)
Definition: rpzaenc.c:137
AVClass::class_name
const char * class_name
The name of the class; usually it is the same name as the context structure type to which the AVClass...
Definition: log.h:72
frame
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
Definition: filter_design.txt:264
AVCodecContext
main external API structure.
Definition: avcodec.h:536
AVFrame::height
int height
Definition: frame.h:376
BlockInfo::total_blocks
int total_blocks
Definition: rpzaenc.c:83
AV_PIX_FMT_NONE
@ AV_PIX_FMT_NONE
Definition: pixfmt.h:65
AV_OPT_TYPE_INT
@ AV_OPT_TYPE_INT
Definition: opt.h:225
options
static const AVOption options[]
Definition: rpzaenc.c:830
BlockInfo::block_width
int block_width
Definition: rpzaenc.c:75
av_clip_uint8
#define av_clip_uint8
Definition: common.h:128
RpzaContext::frame_width
int frame_width
Definition: rpzaenc.c:45
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
AV_CODEC_ID_RPZA
@ AV_CODEC_ID_RPZA
Definition: codec_id.h:91
flush_put_bits
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
Definition: put_bits.h:110
match_color
static int match_color(uint16_t *color, uint8_t colors[4][3])
Definition: rpzaenc.c:358
diff
static av_always_inline int diff(const uint32_t a, const uint32_t b)
Definition: vf_palettegen.c:136
AVPacket
This structure stores compressed data.
Definition: packet.h:346
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:563
AVCodecContext::width
int width
picture width / height.
Definition: avcodec.h:709
R
#define R(color)
Definition: rpzaenc.c:68
rpza_encode_stream
static void rpza_encode_stream(RpzaContext *s, const AVFrame *pict)
Definition: rpzaenc.c:547
AVFrame::linesize
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:349
rgb
static const SheerTable rgb[2]
Definition: sheervideodata.h:32
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207
update_block_stats
static int update_block_stats(RpzaContext *s, BlockInfo *bi, uint16_t *block, uint8_t min_color[3], uint8_t max_color[3], int *total_rgb, int *total_pixels, uint8_t avg_color[3], int first_block)
Definition: rpzaenc.c:443
diff_colors
static int diff_colors(uint8_t *colorA, uint8_t *colorB)
Definition: rpzaenc.c:156
ff_alloc_packet2
int ff_alloc_packet2(AVCodecContext *avctx, AVPacket *avpkt, int64_t size, int64_t min_size)
Check AVPacket size and/or allocate data.
Definition: encode.c:33
int
int
Definition: ffmpeg_filter.c:170
put_bits.h
rpza_encode_frame
static int rpza_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet)
Definition: rpzaenc.c:773
RpzaContext::prev_frame
AVFrame * prev_frame
Definition: rpzaenc.c:42
block1
static int16_t block1[64]
Definition: dct.c:117
RpzaContext::sixteen_color_thresh
int sixteen_color_thresh
Definition: rpzaenc.c:40
max_component_diff
static int max_component_diff(uint16_t *colorA, uint16_t *colorB)
Definition: rpzaenc.c:170
RpzaContext::continue_one_color_thresh
int continue_one_color_thresh
Definition: rpzaenc.c:39
channel
channel
Definition: ebur128.h:39
min
float min
Definition: vorbis_enc_data.h:456
GET_CHAN
#define GET_CHAN(color, chan)
Definition: rpzaenc.c:67