FFmpeg  4.4.5
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  const int y_size = FFMIN(4, bi->image_height - bi->row * 4);
422 
423  for (int y = 0; y < y_size; y++) {
424  memcpy(dest_pixels, src_pixels, 8);
425  dest_pixels += bi->rowstride;
426  src_pixels += bi->rowstride;
427  }
428 }
429 
430 /*
431  * update statistics for the specified block. If first_block,
432  * it initializes the statistics. Otherwise it updates the statistics IF THIS
433  * BLOCK IS SUITABLE TO CONTINUE A 1-COLOR RUN. That is, it checks whether
434  * the range of colors (since the routine was called first_block != 0) are
435  * all close enough intensities to be represented by a single color.
436 
437  * The routine returns 0 if this block is too different to be part of
438  * the same run of 1-color blocks. The routine returns 1 if this
439  * block can be part of the same 1-color block run.
440 
441  * If the routine returns 1, it also updates its arguments to include
442  * the statistics of this block. Otherwise, the stats are unchanged
443  * and don't include the current block.
444  */
445 static int update_block_stats(RpzaContext *s, BlockInfo *bi, uint16_t *block,
446  uint8_t min_color[3], uint8_t max_color[3],
447  int *total_rgb, int *total_pixels,
448  uint8_t avg_color[3], int first_block)
449 {
450  int x, y;
451  int is_in_range;
452  int total_pixels_blk;
453  int threshold;
454 
455  uint8_t min_color_blk[3], max_color_blk[3];
456  int total_rgb_blk[3];
457  uint8_t avg_color_blk[3];
458 
459  if (first_block) {
460  min_color[0] = UINT8_MAX;
461  min_color[1] = UINT8_MAX;
462  min_color[2] = UINT8_MAX;
463  max_color[0] = 0;
464  max_color[1] = 0;
465  max_color[2] = 0;
466  total_rgb[0] = 0;
467  total_rgb[1] = 0;
468  total_rgb[2] = 0;
469  *total_pixels = 0;
470  threshold = s->start_one_color_thresh;
471  } else {
472  threshold = s->continue_one_color_thresh;
473  }
474 
475  /*
476  The *_blk variables will include the current block.
477  Initialize them based on the blocks so far.
478  */
479  min_color_blk[0] = min_color[0];
480  min_color_blk[1] = min_color[1];
481  min_color_blk[2] = min_color[2];
482  max_color_blk[0] = max_color[0];
483  max_color_blk[1] = max_color[1];
484  max_color_blk[2] = max_color[2];
485  total_rgb_blk[0] = total_rgb[0];
486  total_rgb_blk[1] = total_rgb[1];
487  total_rgb_blk[2] = total_rgb[2];
488  total_pixels_blk = *total_pixels + bi->block_height * bi->block_width;
489 
490  /*
491  Update stats for this block's pixels
492  */
493  for (y = 0; y < bi->block_height; y++) {
494  for (x = 0; x < bi->block_width; x++) {
495  total_rgb_blk[0] += R(block[x]);
496  total_rgb_blk[1] += G(block[x]);
497  total_rgb_blk[2] += B(block[x]);
498 
499  min_color_blk[0] = FFMIN(R(block[x]), min_color_blk[0]);
500  min_color_blk[1] = FFMIN(G(block[x]), min_color_blk[1]);
501  min_color_blk[2] = FFMIN(B(block[x]), min_color_blk[2]);
502 
503  max_color_blk[0] = FFMAX(R(block[x]), max_color_blk[0]);
504  max_color_blk[1] = FFMAX(G(block[x]), max_color_blk[1]);
505  max_color_blk[2] = FFMAX(B(block[x]), max_color_blk[2]);
506  }
507  block += bi->rowstride;
508  }
509 
510  /*
511  Calculate average color including current block.
512  */
513  avg_color_blk[0] = total_rgb_blk[0] / total_pixels_blk;
514  avg_color_blk[1] = total_rgb_blk[1] / total_pixels_blk;
515  avg_color_blk[2] = total_rgb_blk[2] / total_pixels_blk;
516 
517  /*
518  Are all the pixels within threshold of the average color?
519  */
520  is_in_range = (max_color_blk[0] - avg_color_blk[0] <= threshold &&
521  max_color_blk[1] - avg_color_blk[1] <= threshold &&
522  max_color_blk[2] - avg_color_blk[2] <= threshold &&
523  avg_color_blk[0] - min_color_blk[0] <= threshold &&
524  avg_color_blk[1] - min_color_blk[1] <= threshold &&
525  avg_color_blk[2] - min_color_blk[2] <= threshold);
526 
527  if (is_in_range) {
528  /*
529  Set the output variables to include this block.
530  */
531  min_color[0] = min_color_blk[0];
532  min_color[1] = min_color_blk[1];
533  min_color[2] = min_color_blk[2];
534  max_color[0] = max_color_blk[0];
535  max_color[1] = max_color_blk[1];
536  max_color[2] = max_color_blk[2];
537  total_rgb[0] = total_rgb_blk[0];
538  total_rgb[1] = total_rgb_blk[1];
539  total_rgb[2] = total_rgb_blk[2];
540  *total_pixels = total_pixels_blk;
541  avg_color[0] = avg_color_blk[0];
542  avg_color[1] = avg_color_blk[1];
543  avg_color[2] = avg_color_blk[2];
544  }
545 
546  return is_in_range;
547 }
548 
549 static void rpza_encode_stream(RpzaContext *s, const AVFrame *pict)
550 {
551  BlockInfo bi;
552  int block_counter = 0;
553  int n_blocks;
554  int total_blocks;
555  int prev_block_offset;
556  int block_offset = 0;
557  uint8_t min = 0, max = 0;
558  channel_offset chan;
559  int i;
560  int tmp_min, tmp_max;
561  int total_rgb[3];
562  uint8_t avg_color[3];
563  int pixel_count;
564  uint8_t min_color[3], max_color[3];
565  double slope, y_intercept, correlation_coef;
566  uint16_t *src_pixels = (uint16_t *)pict->data[0];
567  uint16_t *prev_pixels = (uint16_t *)s->prev_frame->data[0];
568 
569  /* Number of 4x4 blocks in frame. */
570  total_blocks = ((s->frame_width + 3) / 4) * ((s->frame_height + 3) / 4);
571 
572  bi.image_width = s->frame_width;
573  bi.image_height = s->frame_height;
574  bi.rowstride = pict->linesize[0] / 2;
575 
576  bi.blocks_per_row = (s->frame_width + 3) / 4;
577 
578  while (block_counter < total_blocks) {
579  // SKIP CHECK
580  // make sure we have a valid previous frame and we're not writing
581  // a key frame
582  if (!s->first_frame) {
583  n_blocks = 0;
584  prev_block_offset = 0;
585 
586  while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
587 
588  block_offset = get_block_info(&bi, block_counter + n_blocks);
589 
590  // multi-block opcodes cannot span multiple rows.
591  // If we're starting a new row, break out and write the opcode
592  /* TODO: Should eventually use bi.row here to determine when a
593  row break occurs, but that is currently breaking the
594  quicktime player. This is probably due to a bug in the
595  way I'm calculating the current row.
596  */
597  if (prev_block_offset && block_offset - prev_block_offset > 12) {
598  break;
599  }
600 
601  prev_block_offset = block_offset;
602 
603  if (compare_blocks(&prev_pixels[block_offset],
604  &src_pixels[block_offset], &bi, s->skip_frame_thresh) != 0) {
605  // write out skipable blocks
606  if (n_blocks) {
607 
608  // write skip opcode
609  put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
610  block_counter += n_blocks;
611 
612  goto post_skip;
613  }
614  break;
615  }
616 
617  /*
618  * NOTE: we don't update skipped blocks in the previous frame buffer
619  * since skipped needs always to be compared against the first skipped
620  * block to avoid artifacts during gradual fade in/outs.
621  */
622 
623  // update_block_in_prev_frame(&src_pixels[block_offset],
624  // &prev_pixels[block_offset], &bi, block_counter + n_blocks);
625 
626  n_blocks++;
627  }
628 
629  // we're either at the end of the frame or we've reached the maximum
630  // of 32 blocks in a run. Write out the run.
631  if (n_blocks) {
632  // write skip opcode
633  put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
634  block_counter += n_blocks;
635 
636  continue;
637  }
638 
639  } else {
640  block_offset = get_block_info(&bi, block_counter);
641  }
642 post_skip :
643 
644  // ONE COLOR CHECK
645  if (update_block_stats(s, &bi, &src_pixels[block_offset],
646  min_color, max_color,
647  total_rgb, &pixel_count, avg_color, 1)) {
648  prev_block_offset = block_offset;
649 
650  n_blocks = 1;
651 
652  /* update this block in the previous frame buffer */
653  update_block_in_prev_frame(&src_pixels[block_offset],
654  &prev_pixels[block_offset], &bi, block_counter + n_blocks);
655 
656  // check for subsequent blocks with the same color
657  while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
658  block_offset = get_block_info(&bi, block_counter + n_blocks);
659 
660  // multi-block opcodes cannot span multiple rows.
661  // If we've hit end of a row, break out and write the opcode
662  if (block_offset - prev_block_offset > 12) {
663  break;
664  }
665 
666  if (!update_block_stats(s, &bi, &src_pixels[block_offset],
667  min_color, max_color,
668  total_rgb, &pixel_count, avg_color, 0)) {
669  break;
670  }
671 
672  prev_block_offset = block_offset;
673 
674  /* update this block in the previous frame buffer */
675  update_block_in_prev_frame(&src_pixels[block_offset],
676  &prev_pixels[block_offset], &bi, block_counter + n_blocks);
677 
678  n_blocks++;
679  }
680 
681  // write one color opcode.
682  put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
683  // write color to encode.
684  put_bits(&s->pb, 16, rgb24_to_rgb555(avg_color));
685  // skip past the blocks we've just encoded.
686  block_counter += n_blocks;
687  } else { // FOUR COLOR CHECK
688  int err = 0;
689 
690  // get max component diff for block
691  get_max_component_diff(&bi, &src_pixels[block_offset], &min, &max, &chan);
692 
693  min_color[0] = 0;
694  max_color[0] = 0;
695  min_color[1] = 0;
696  max_color[1] = 0;
697  min_color[2] = 0;
698  max_color[2] = 0;
699 
700  // run least squares against other two components
701  for (i = 0; i < 3; i++) {
702  if (i == chan) {
703  min_color[i] = min;
704  max_color[i] = max;
705  continue;
706  }
707 
708  slope = y_intercept = correlation_coef = 0;
709 
710  if (leastsquares(&src_pixels[block_offset], &bi, chan, i,
711  &slope, &y_intercept, &correlation_coef)) {
712  min_color[i] = GET_CHAN(src_pixels[block_offset], i);
713  max_color[i] = GET_CHAN(src_pixels[block_offset], i);
714  } else {
715  tmp_min = (int)(0.5 + min * slope + y_intercept);
716  tmp_max = (int)(0.5 + max * slope + y_intercept);
717 
718  av_assert0(tmp_min <= tmp_max);
719  // clamp min and max color values
720  tmp_min = av_clip_uint8(tmp_min);
721  tmp_max = av_clip_uint8(tmp_max);
722 
723  err = FFMAX(calc_lsq_max_fit_error(&src_pixels[block_offset], &bi,
724  min, max, tmp_min, tmp_max, chan, i), err);
725 
726  min_color[i] = tmp_min;
727  max_color[i] = tmp_max;
728  }
729  }
730 
731  if (err > s->sixteen_color_thresh) { // DO SIXTEEN COLOR BLOCK
732  uint16_t *row_ptr;
733  int y_size, rgb555;
734 
735  block_offset = get_block_info(&bi, block_counter);
736 
737  row_ptr = &src_pixels[block_offset];
738  y_size = FFMIN(4, bi.image_height - bi.row * 4);
739 
740  for (int y = 0; y < y_size; y++) {
741  for (int x = 0; x < 4; x++) {
742  rgb555 = row_ptr[x] & ~0x8000;
743 
744  put_bits(&s->pb, 16, rgb555);
745  }
746  row_ptr += bi.rowstride;
747  }
748 
749  for (int y = y_size; y < 4; y++) {
750  for (int x = 0; x < 4; x++)
751  put_bits(&s->pb, 16, 0);
752  }
753 
754  block_counter++;
755  } else { // FOUR COLOR BLOCK
756  block_counter += encode_four_color_block(min_color, max_color,
757  &s->pb, &src_pixels[block_offset], &bi);
758  }
759 
760  /* update this block in the previous frame buffer */
761  update_block_in_prev_frame(&src_pixels[block_offset],
762  &prev_pixels[block_offset], &bi, block_counter);
763  }
764  }
765 }
766 
768 {
769  RpzaContext *s = avctx->priv_data;
770 
771  s->frame_width = avctx->width;
772  s->frame_height = avctx->height;
773 
774  s->prev_frame = av_frame_alloc();
775  if (!s->prev_frame)
776  return AVERROR(ENOMEM);
777 
778  return 0;
779 }
780 
782  const AVFrame *frame, int *got_packet)
783 {
784  RpzaContext *s = avctx->priv_data;
785  const AVFrame *pict = frame;
786  uint8_t *buf;
787  int ret;
788 
789  if ((ret = ff_alloc_packet2(avctx, pkt, 6LL * avctx->height * avctx->width, 0)) < 0)
790  return ret;
791 
792  init_put_bits(&s->pb, pkt->data, pkt->size);
793 
794  // skip 4 byte header, write it later once the size of the chunk is known
795  put_bits32(&s->pb, 0x00);
796 
797  if (!s->prev_frame->data[0]) {
798  s->first_frame = 1;
799  s->prev_frame->format = pict->format;
800  s->prev_frame->width = pict->width;
801  s->prev_frame->height = pict->height;
802  ret = av_frame_get_buffer(s->prev_frame, 0);
803  if (ret < 0)
804  return ret;
805  } else {
806  s->first_frame = 0;
807  }
808 
809  rpza_encode_stream(s, pict);
810 
811  flush_put_bits(&s->pb);
812 
813  av_shrink_packet(pkt, put_bits_count(&s->pb) >> 3);
814  buf = pkt->data;
815 
816  // write header opcode
817  buf[0] = 0xe1; // chunk opcode
818 
819  // write chunk length
820  AV_WB24(buf + 1, pkt->size);
821 
822  *got_packet = 1;
823 
824  return 0;
825 }
826 
827 static int rpza_encode_end(AVCodecContext *avctx)
828 {
829  RpzaContext *s = (RpzaContext *)avctx->priv_data;
830 
831  av_frame_free(&s->prev_frame);
832 
833  return 0;
834 }
835 
836 #define OFFSET(x) offsetof(RpzaContext, x)
837 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
838 static const AVOption options[] = {
839  { "skip_frame_thresh", NULL, OFFSET(skip_frame_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
840  { "start_one_color_thresh", NULL, OFFSET(start_one_color_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
841  { "continue_one_color_thresh", NULL, OFFSET(continue_one_color_thresh), AV_OPT_TYPE_INT, {.i64=0}, 0, 24, VE},
842  { "sixteen_color_thresh", NULL, OFFSET(sixteen_color_thresh), AV_OPT_TYPE_INT, {.i64=1}, 0, 24, VE},
843  { NULL },
844 };
845 
846 static const AVClass rpza_class = {
847  .class_name = "rpza",
848  .item_name = av_default_item_name,
849  .option = options,
850  .version = LIBAVUTIL_VERSION_INT,
851 };
852 
854  .name = "rpza",
855  .long_name = NULL_IF_CONFIG_SMALL("QuickTime video (RPZA)"),
856  .type = AVMEDIA_TYPE_VIDEO,
857  .id = AV_CODEC_ID_RPZA,
858  .priv_data_size = sizeof(RpzaContext),
859  .priv_class = &rpza_class,
861  .encode2 = rpza_encode_frame,
862  .close = rpza_encode_end,
863  .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
864  .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_RGB555,
866 };
uint8_t
simple assert() macros that are a bit more flexible than ISO C assert().
#define av_assert0(cond)
assert() equivalent, that is always enabled.
Definition: avassert.h:37
Libavcodec external API header.
static av_cold int init(AVCodecContext *avctx)
Definition: avrndec.c:31
#define s(width, name)
Definition: cbs_vp9.c:257
common internal and external API header
#define FFMIN(a, b)
Definition: common.h:105
#define FFMAX(a, b)
Definition: common.h:103
#define av_clip_uint8
Definition: common.h:128
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:72
#define NULL
Definition: coverity.c:32
static __device__ float floor(float a)
Definition: cuda_runtime.h:173
#define max(a, b)
Definition: cuda_runtime.h:33
static AVFrame * frame
channel
Use these values when setting the channel map with ebur128_set_channel().
Definition: ebur128.h:39
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
@ AV_OPT_TYPE_INT
Definition: opt.h:225
@ AV_CODEC_ID_RPZA
Definition: codec_id.h:91
void av_shrink_packet(AVPacket *pkt, int size)
Reduce packet size, correctly zeroing padding.
Definition: avpacket.c:114
#define AVERROR(e)
Definition: error.h:43
int av_frame_get_buffer(AVFrame *frame, int align)
Allocate new buffer(s) for audio or video data.
Definition: frame.c:337
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:203
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
Definition: frame.c:190
const char * av_default_item_name(void *ptr)
Return the context name.
Definition: log.c:235
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
#define LIBAVUTIL_VERSION_INT
Definition: version.h:85
int i
Definition: input.c:407
#define AV_WB24(p, d)
Definition: intreadwrite.h:450
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
Definition: j2kenc.c:218
#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
common internal API header
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
static enum AVPixelFormat pix_fmts[]
Definition: libkvazaar.c:309
AVOptions.
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
@ AV_PIX_FMT_NONE
Definition: pixfmt.h:65
#define AV_PIX_FMT_RGB555
Definition: pixfmt.h:387
bitstream writer API
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
Definition: put_bits.h:57
static void av_unused put_bits32(PutBitContext *s, uint32_t value)
Write exactly 32 bits into a bitstream.
Definition: put_bits.h:263
static int put_bits_count(PutBitContext *s)
Definition: put_bits.h:76
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
Definition: put_bits.h:110
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
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
static int rpza_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet)
Definition: rpzaenc.c:781
static int rpza_encode_end(AVCodecContext *avctx)
Definition: rpzaenc.c:827
channel_offset
Definition: rpzaenc.c:52
@ BLUE
Definition: rpzaenc.c:55
@ GREEN
Definition: rpzaenc.c:54
@ RED
Definition: rpzaenc.c:53
static const AVOption options[]
Definition: rpzaenc.c:838
#define VE
Definition: rpzaenc.c:837
static int rpza_encode_init(AVCodecContext *avctx)
Definition: rpzaenc.c:767
static int get_block_info(BlockInfo *bi, int block)
Definition: rpzaenc.c:115
static const AVClass rpza_class
Definition: rpzaenc.c:846
static int max_component_diff(uint16_t *colorA, uint16_t *colorB)
Definition: rpzaenc.c:170
#define G(color)
Definition: rpzaenc.c:69
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
static int diff_colors(uint8_t *colorA, uint8_t *colorB)
Definition: rpzaenc.c:156
AVCodec ff_rpza_encoder
Definition: rpzaenc.c:853
static int match_color(uint16_t *color, uint8_t colors[4][3])
Definition: rpzaenc.c:358
static int compare_blocks(uint16_t *block1, uint16_t *block2, BlockInfo *bi, int thresh)
Definition: rpzaenc.c:244
#define B(color)
Definition: rpzaenc.c:70
#define SQR(x)
Definition: rpzaenc.c:64
static void rpza_encode_stream(RpzaContext *s, const AVFrame *pict)
Definition: rpzaenc.c:549
static uint16_t rgb24_to_rgb555(uint8_t *rgb24)
Definition: rpzaenc.c:137
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:445
#define OFFSET(x)
Definition: rpzaenc.c:836
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
#define GET_CHAN(color, chan)
Definition: rpzaenc.c:67
#define R(color)
Definition: rpzaenc.c:68
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
static void get_colors(uint8_t *min, uint8_t *max, uint8_t color4[4][3])
Definition: rpzaenc.c:86
static const SheerTable rgb[2]
Describe the class of an AVClass context structure.
Definition: log.h:67
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
main external API structure.
Definition: avcodec.h:536
int width
picture width / height.
Definition: avcodec.h:709
void * priv_data
Definition: avcodec.h:563
AVCodec.
Definition: codec.h:197
const char * name
Name of the codec implementation.
Definition: codec.h:204
This structure describes decoded (raw) audio or video data.
Definition: frame.h:318
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:332
int width
Definition: frame.h:376
int height
Definition: frame.h:376
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:349
int format
format of the frame, -1 if unknown or unset Values correspond to enum AVPixelFormat for video frames,...
Definition: frame.h:391
AVOption.
Definition: opt.h:248
This structure stores compressed data.
Definition: packet.h:346
int size
Definition: packet.h:370
uint8_t * data
Definition: packet.h:369
int col
Definition: rpzaenc.c:74
int rowstride
Definition: rpzaenc.c:81
int image_width
Definition: rpzaenc.c:77
int block_width
Definition: rpzaenc.c:75
int block_index
Definition: rpzaenc.c:79
int row
Definition: rpzaenc.c:73
int image_height
Definition: rpzaenc.c:78
int total_blocks
Definition: rpzaenc.c:83
int block_height
Definition: rpzaenc.c:76
uint16_t start
Definition: rpzaenc.c:80
int blocks_per_row
Definition: rpzaenc.c:82
AVClass * avclass
Definition: rpzaenc.c:35
AVFrame * prev_frame
Definition: rpzaenc.c:42
int frame_width
Definition: rpzaenc.c:45
PutBitContext pb
Definition: rpzaenc.c:43
int continue_one_color_thresh
Definition: rpzaenc.c:39
int skip_frame_thresh
Definition: rpzaenc.c:37
int frame_height
Definition: rpzaenc.c:46
int first_frame
Definition: rpzaenc.c:48
int sixteen_color_thresh
Definition: rpzaenc.c:40
int start_one_color_thresh
Definition: rpzaenc.c:38
Definition: rpzaenc.c:58
uint8_t b
Definition: rpzaenc.c:61
uint8_t g
Definition: rpzaenc.c:60
uint8_t r
Definition: rpzaenc.c:59
static uint8_t tmp[11]
Definition: aes_ctr.c:27
static int16_t block[64]
Definition: dct.c:116
static int16_t block1[64]
Definition: dct.c:117
AVPacket * pkt
Definition: movenc.c:59
const char * b
Definition: vf_curves.c:118
const char * g
Definition: vf_curves.c:117
const char * r
Definition: vf_curves.c:116
static av_always_inline int diff(const uint32_t a, const uint32_t b)
float min