/src/ffmpeg/libavcodec/flashsv2enc.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Flash Screen Video Version 2 encoder |
3 | | * Copyright (C) 2009 Joshua Warner |
4 | | * |
5 | | * This file is part of FFmpeg. |
6 | | * |
7 | | * FFmpeg is free software; you can redistribute it and/or |
8 | | * modify it under the terms of the GNU Lesser General Public |
9 | | * License as published by the Free Software Foundation; either |
10 | | * version 2.1 of the License, or (at your option) any later version. |
11 | | * |
12 | | * FFmpeg is distributed in the hope that it will be useful, |
13 | | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | | * Lesser General Public License for more details. |
16 | | * |
17 | | * You should have received a copy of the GNU Lesser General Public |
18 | | * License along with FFmpeg; if not, write to the Free Software |
19 | | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 | | */ |
21 | | |
22 | | /** |
23 | | * @file |
24 | | * Flash Screen Video Version 2 encoder |
25 | | * @author Joshua Warner |
26 | | */ |
27 | | |
28 | | /* Differences from version 1 stream: |
29 | | * NOTE: Currently, the only player that supports version 2 streams is Adobe Flash Player itself. |
30 | | * * Supports sending only a range of scanlines in a block, |
31 | | * indicating a difference from the corresponding block in the last keyframe. |
32 | | * * Supports initializing the zlib dictionary with data from the corresponding |
33 | | * block in the last keyframe, to improve compression. |
34 | | * * Supports a hybrid 15-bit rgb / 7-bit palette color space. |
35 | | */ |
36 | | |
37 | | /* TODO: |
38 | | * Don't keep Block structures for both current frame and keyframe. |
39 | | * Make better heuristics for deciding stream parameters (optimum_* functions). Currently these return constants. |
40 | | * Figure out how to encode palette information in the stream, choose an optimum palette at each keyframe. |
41 | | * Figure out how the zlibPrimeCompressCurrent flag works, implement support. |
42 | | * Find other sample files (that weren't generated here), develop a decoder. |
43 | | */ |
44 | | |
45 | | #include <stdio.h> |
46 | | #include <stdlib.h> |
47 | | #include <zlib.h> |
48 | | |
49 | | #include "libavutil/imgutils.h" |
50 | | #include "libavutil/mem.h" |
51 | | #include "avcodec.h" |
52 | | #include "codec_internal.h" |
53 | | #include "encode.h" |
54 | | #include "put_bits.h" |
55 | | #include "zlib_wrapper.h" |
56 | | |
57 | | #define HAS_IFRAME_IMAGE 0x02 |
58 | 9.54k | #define HAS_PALLET_INFO 0x01 |
59 | | |
60 | | #define COLORSPACE_BGR 0x00 |
61 | 48.4k | #define COLORSPACE_15_7 0x10 |
62 | 50.5k | #define HAS_DIFF_BLOCKS 0x04 |
63 | 47.1k | #define ZLIB_PRIME_COMPRESS_CURRENT 0x02 |
64 | 9.65k | #define ZLIB_PRIME_COMPRESS_PREVIOUS 0x01 |
65 | | |
66 | | // Disables experimental "smart" parameter-choosing code, as well as the statistics that it depends on. |
67 | | // At the moment, the "smart" code is a great example of how the parameters *shouldn't* be chosen. |
68 | | #define FLASHSV2_DUMB |
69 | | |
70 | | typedef struct Block { |
71 | | uint8_t *enc; |
72 | | uint8_t *sl_begin, *sl_end; |
73 | | int enc_size; |
74 | | uint8_t *data; |
75 | | unsigned long data_size; |
76 | | |
77 | | uint8_t start, len; |
78 | | uint8_t dirty; |
79 | | uint8_t col, row, width, height; |
80 | | uint8_t flags; |
81 | | } Block; |
82 | | |
83 | | typedef struct Palette { |
84 | | unsigned colors[128]; |
85 | | uint8_t index[1 << 15]; |
86 | | } Palette; |
87 | | |
88 | | typedef struct FlashSV2Context { |
89 | | AVCodecContext *avctx; |
90 | | uint8_t *current_frame; |
91 | | uint8_t *key_frame; |
92 | | uint8_t *encbuffer; |
93 | | uint8_t *keybuffer; |
94 | | uint8_t *databuffer; |
95 | | |
96 | | uint8_t *blockbuffer; |
97 | | int blockbuffer_size; |
98 | | |
99 | | Block *frame_blocks; |
100 | | Block *key_blocks; |
101 | | int frame_size; |
102 | | int blocks_size; |
103 | | |
104 | | int use15_7, dist, comp; |
105 | | |
106 | | int rows, cols; |
107 | | |
108 | | int64_t last_key_frame; |
109 | | |
110 | | int image_width, image_height; |
111 | | int block_width, block_height; |
112 | | uint8_t flags; |
113 | | uint8_t use_custom_palette; |
114 | | uint8_t palette_type; ///< 0=>default, 1=>custom - changed when palette regenerated. |
115 | | Palette palette; |
116 | | FFZStream zstream; |
117 | | #ifndef FLASHSV2_DUMB |
118 | | double tot_blocks; ///< blocks encoded since last keyframe |
119 | | double diff_blocks; ///< blocks that were different since last keyframe |
120 | | double tot_lines; ///< total scanlines in image since last keyframe |
121 | | double diff_lines; ///< scanlines that were different since last keyframe |
122 | | double raw_size; ///< size of raw frames since last keyframe |
123 | | double comp_size; ///< size of compressed data since last keyframe |
124 | | double uncomp_size; ///< size of uncompressed data since last keyframe |
125 | | |
126 | | double total_bits; ///< total bits written to stream so far |
127 | | #endif |
128 | | } FlashSV2Context; |
129 | | |
130 | | static av_cold void cleanup(FlashSV2Context * s) |
131 | 1.21k | { |
132 | 1.21k | av_freep(&s->encbuffer); |
133 | 1.21k | av_freep(&s->keybuffer); |
134 | 1.21k | av_freep(&s->databuffer); |
135 | 1.21k | av_freep(&s->blockbuffer); |
136 | 1.21k | av_freep(&s->current_frame); |
137 | 1.21k | av_freep(&s->key_frame); |
138 | | |
139 | 1.21k | av_freep(&s->frame_blocks); |
140 | 1.21k | av_freep(&s->key_blocks); |
141 | 1.21k | ff_deflate_end(&s->zstream); |
142 | 1.21k | } |
143 | | |
144 | | static void init_blocks(FlashSV2Context * s, Block * blocks, |
145 | | uint8_t * encbuf, uint8_t * databuf) |
146 | 2.33k | { |
147 | 2.33k | int row, col; |
148 | 2.33k | Block *b; |
149 | 2.33k | memset(blocks, 0, s->cols * s->rows * sizeof(*blocks)); |
150 | 14.3k | for (col = 0; col < s->cols; col++) { |
151 | 39.0k | for (row = 0; row < s->rows; row++) { |
152 | 26.9k | b = blocks + (col + row * s->cols); |
153 | 26.9k | b->width = (col < s->cols - 1) ? |
154 | 19.7k | s->block_width : |
155 | 26.9k | s->image_width - col * s->block_width; |
156 | | |
157 | 26.9k | b->height = (row < s->rows - 1) ? |
158 | 14.9k | s->block_height : |
159 | 26.9k | s->image_height - row * s->block_height; |
160 | | |
161 | 26.9k | b->row = row; |
162 | 26.9k | b->col = col; |
163 | 26.9k | b->enc = encbuf; |
164 | 26.9k | b->data = databuf; |
165 | 26.9k | encbuf += b->width * b->height * 3; |
166 | 26.9k | databuf = databuf ? databuf + b->width * b->height * 6 : NULL; |
167 | 26.9k | } |
168 | 12.0k | } |
169 | 2.33k | } |
170 | | |
171 | | static void reset_stats(FlashSV2Context * s) |
172 | 2.54k | { |
173 | | #ifndef FLASHSV2_DUMB |
174 | | s->diff_blocks = 0.1; |
175 | | s->tot_blocks = 1; |
176 | | s->diff_lines = 0.1; |
177 | | s->tot_lines = 1; |
178 | | s->raw_size = s->comp_size = s->uncomp_size = 10; |
179 | | #endif |
180 | 2.54k | } |
181 | | |
182 | | static int update_block_dimensions(FlashSV2Context *s, int block_width, int block_height) |
183 | 1.16k | { |
184 | 1.16k | s->block_width = block_width; |
185 | 1.16k | s->block_height = block_height; |
186 | 1.16k | s->rows = (s->image_height + s->block_height - 1) / s->block_height; |
187 | 1.16k | s->cols = (s->image_width + s->block_width - 1) / s->block_width; |
188 | 1.16k | if (s->rows * s->cols > s->blocks_size / sizeof(Block)) { |
189 | 1.16k | s->frame_blocks = av_realloc_array(s->frame_blocks, s->rows, s->cols * sizeof(Block)); |
190 | 1.16k | s->key_blocks = av_realloc_array(s->key_blocks, s->cols, s->rows * sizeof(Block)); |
191 | 1.16k | if (!s->frame_blocks || !s->key_blocks) { |
192 | 0 | av_log(s->avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); |
193 | 0 | return AVERROR(ENOMEM); |
194 | 0 | } |
195 | 1.16k | s->blocks_size = s->rows * s->cols * sizeof(Block); |
196 | 1.16k | } |
197 | 1.16k | init_blocks(s, s->frame_blocks, s->encbuffer, s->databuffer); |
198 | 1.16k | init_blocks(s, s->key_blocks, s->keybuffer, 0); |
199 | | |
200 | 1.16k | av_fast_malloc(&s->blockbuffer, &s->blockbuffer_size, block_width * block_height * 6); |
201 | 1.16k | if (!s->blockbuffer) { |
202 | 0 | av_log(s->avctx, AV_LOG_ERROR, "Could not allocate block buffer.\n"); |
203 | 0 | return AVERROR(ENOMEM); |
204 | 0 | } |
205 | 1.16k | return 0; |
206 | 1.16k | } |
207 | | |
208 | | |
209 | | static av_cold int flashsv2_encode_init(AVCodecContext * avctx) |
210 | 1.21k | { |
211 | 1.21k | FlashSV2Context *s = avctx->priv_data; |
212 | 1.21k | int ret; |
213 | | |
214 | 1.21k | s->avctx = avctx; |
215 | | |
216 | 1.21k | s->comp = avctx->compression_level; |
217 | 1.21k | if (s->comp == -1) |
218 | 1.21k | s->comp = 9; |
219 | 1.21k | if (s->comp < 0 || s->comp > 9) { |
220 | 0 | av_log(avctx, AV_LOG_ERROR, |
221 | 0 | "Compression level should be 0-9, not %d\n", s->comp); |
222 | 0 | return AVERROR(EINVAL); |
223 | 0 | } |
224 | | |
225 | | |
226 | 1.21k | if ((avctx->width > 4095) || (avctx->height > 4095)) { |
227 | 32 | av_log(avctx, AV_LOG_ERROR, |
228 | 32 | "Input dimensions too large, input must be max 4095x4095 !\n"); |
229 | 32 | return AVERROR(EINVAL); |
230 | 32 | } |
231 | 1.17k | if ((avctx->width < 16) || (avctx->height < 16)) { |
232 | 13 | av_log(avctx, AV_LOG_ERROR, |
233 | 13 | "Input dimensions too small, input must be at least 16x16 !\n"); |
234 | 13 | return AVERROR(EINVAL); |
235 | 13 | } |
236 | | |
237 | 1.16k | if ((ret = av_image_check_size(avctx->width, avctx->height, 0, avctx)) < 0) |
238 | 0 | return ret; |
239 | | |
240 | 1.16k | ret = ff_deflate_init(&s->zstream, s->comp, avctx); |
241 | 1.16k | if (ret < 0) |
242 | 0 | return ret; |
243 | 1.16k | s->last_key_frame = 0; |
244 | | |
245 | 1.16k | s->image_width = avctx->width; |
246 | 1.16k | s->image_height = avctx->height; |
247 | | |
248 | 1.16k | s->frame_size = s->image_width * s->image_height * 3; |
249 | | |
250 | 1.16k | s->encbuffer = av_mallocz(s->frame_size); |
251 | 1.16k | s->keybuffer = av_mallocz(s->frame_size); |
252 | 1.16k | s->databuffer = av_mallocz(s->frame_size * 6); |
253 | 1.16k | s->current_frame = av_mallocz(s->frame_size); |
254 | 1.16k | s->key_frame = av_mallocz(s->frame_size); |
255 | 1.16k | if (!s->encbuffer || !s->keybuffer || !s->databuffer |
256 | 1.16k | || !s->current_frame || !s->key_frame) { |
257 | 0 | av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); |
258 | 0 | return AVERROR(ENOMEM); |
259 | 0 | } |
260 | | |
261 | 1.16k | reset_stats(s); |
262 | | #ifndef FLASHSV2_DUMB |
263 | | s->total_bits = 1; |
264 | | #endif |
265 | | |
266 | 1.16k | s->use_custom_palette = 0; |
267 | 1.16k | s->palette_type = -1; // so that the palette will be generated in reconfigure_at_keyframe |
268 | | |
269 | 1.16k | return update_block_dimensions(s, 64, 64); |
270 | 1.16k | } |
271 | | |
272 | | static int new_key_frame(FlashSV2Context * s) |
273 | 1.37k | { |
274 | 1.37k | int i; |
275 | 1.37k | memcpy(s->key_blocks, s->frame_blocks, s->blocks_size); |
276 | 1.37k | memcpy(s->key_frame, s->current_frame, s->frame_size); |
277 | | |
278 | 15.3k | for (i = 0; i < s->rows * s->cols; i++) { |
279 | 13.9k | s->key_blocks[i].enc += (s->keybuffer - s->encbuffer); |
280 | 13.9k | s->key_blocks[i].sl_begin = 0; |
281 | 13.9k | s->key_blocks[i].sl_end = 0; |
282 | 13.9k | s->key_blocks[i].data = 0; |
283 | 13.9k | } |
284 | 1.37k | memcpy(s->keybuffer, s->encbuffer, s->frame_size); |
285 | | |
286 | 1.37k | return 0; |
287 | 1.37k | } |
288 | | |
289 | | static int write_palette(FlashSV2Context * s, uint8_t * buf, int buf_size) |
290 | 0 | { |
291 | | //this isn't implemented yet! Default palette only! |
292 | 0 | return -1; |
293 | 0 | } |
294 | | |
295 | | static int write_header(FlashSV2Context * s, uint8_t * buf, int buf_size) |
296 | 9.54k | { |
297 | 9.54k | PutBitContext pb; |
298 | 9.54k | int buf_pos, len; |
299 | | |
300 | 9.54k | if (buf_size < 5) |
301 | 0 | return -1; |
302 | | |
303 | 9.54k | init_put_bits(&pb, buf, buf_size); |
304 | | |
305 | 9.54k | put_bits(&pb, 4, (s->block_width >> 4) - 1); |
306 | 9.54k | put_bits(&pb, 12, s->image_width); |
307 | 9.54k | put_bits(&pb, 4, (s->block_height >> 4) - 1); |
308 | 9.54k | put_bits(&pb, 12, s->image_height); |
309 | | |
310 | 9.54k | flush_put_bits(&pb); |
311 | 9.54k | buf_pos = 4; |
312 | | |
313 | 9.54k | buf[buf_pos++] = s->flags; |
314 | | |
315 | 9.54k | if (s->flags & HAS_PALLET_INFO) { |
316 | 0 | len = write_palette(s, buf + buf_pos, buf_size - buf_pos); |
317 | 0 | if (len < 0) |
318 | 0 | return -1; |
319 | 0 | buf_pos += len; |
320 | 0 | } |
321 | | |
322 | 9.54k | return buf_pos; |
323 | 9.54k | } |
324 | | |
325 | | static int write_block(Block * b, uint8_t * buf, int buf_size) |
326 | 23.7k | { |
327 | 23.7k | int buf_pos = 0; |
328 | 23.7k | unsigned block_size = b->data_size; |
329 | | |
330 | 23.7k | if (b->flags & HAS_DIFF_BLOCKS) |
331 | 3.35k | block_size += 2; |
332 | 23.7k | if (b->flags & ZLIB_PRIME_COMPRESS_CURRENT) |
333 | 0 | block_size += 2; |
334 | 23.7k | if (block_size > 0) |
335 | 23.4k | block_size += 1; |
336 | 23.7k | if (buf_size < block_size + 2) |
337 | 0 | return -1; |
338 | | |
339 | 23.7k | buf[buf_pos++] = block_size >> 8; |
340 | 23.7k | buf[buf_pos++] = block_size; |
341 | | |
342 | 23.7k | if (block_size == 0) |
343 | 358 | return buf_pos; |
344 | | |
345 | 23.4k | buf[buf_pos++] = b->flags; |
346 | | |
347 | 23.4k | if (b->flags & HAS_DIFF_BLOCKS) { |
348 | 3.35k | buf[buf_pos++] = (b->start); |
349 | 3.35k | buf[buf_pos++] = (b->len); |
350 | 3.35k | } |
351 | | |
352 | 23.4k | if (b->flags & ZLIB_PRIME_COMPRESS_CURRENT) { |
353 | | //This feature of the format is poorly understood, and as of now, unused. |
354 | 0 | buf[buf_pos++] = (b->col); |
355 | 0 | buf[buf_pos++] = (b->row); |
356 | 0 | } |
357 | | |
358 | 23.4k | memcpy(buf + buf_pos, b->data, b->data_size); |
359 | | |
360 | 23.4k | buf_pos += b->data_size; |
361 | | |
362 | 23.4k | return buf_pos; |
363 | 23.7k | } |
364 | | |
365 | | static int encode_zlib(Block *b, uint8_t *buf, unsigned long *buf_size, |
366 | | z_stream *zstream) |
367 | 23.9k | { |
368 | 23.9k | int res; |
369 | | |
370 | 23.9k | if (deflateReset(zstream) != Z_OK) |
371 | 0 | return AVERROR_EXTERNAL; |
372 | 23.9k | zstream->next_out = buf; |
373 | 23.9k | zstream->avail_out = *buf_size; |
374 | 23.9k | zstream->next_in = b->sl_begin; |
375 | 23.9k | zstream->avail_in = b->sl_end - b->sl_begin; |
376 | 23.9k | res = deflate(zstream, Z_FINISH); |
377 | 23.9k | if (res != Z_STREAM_END) |
378 | 26 | return AVERROR_EXTERNAL; |
379 | 23.9k | *buf_size -= zstream->avail_out; |
380 | 23.9k | return 0; |
381 | 23.9k | } |
382 | | |
383 | | static int encode_zlibprime(Block * b, Block * prime, uint8_t * buf, |
384 | | int *buf_size, z_stream *zstream) |
385 | 10.0k | { |
386 | 10.0k | int res; |
387 | | |
388 | 10.0k | if (deflateReset(zstream) != Z_OK) |
389 | 0 | return AVERROR_EXTERNAL; |
390 | 10.0k | zstream->next_in = prime->enc; |
391 | 10.0k | zstream->avail_in = prime->enc_size; |
392 | 20.0k | while (zstream->avail_in > 0) { |
393 | 10.0k | zstream->next_out = buf; |
394 | 10.0k | zstream->avail_out = *buf_size; |
395 | 10.0k | res = deflate(zstream, Z_SYNC_FLUSH); |
396 | 10.0k | if (res < 0) |
397 | 0 | return -1; |
398 | 10.0k | } |
399 | | |
400 | 10.0k | zstream->next_in = b->sl_begin; |
401 | 10.0k | zstream->avail_in = b->sl_end - b->sl_begin; |
402 | 10.0k | zstream->next_out = buf; |
403 | 10.0k | zstream->avail_out = *buf_size; |
404 | 10.0k | res = deflate(zstream, Z_FINISH); |
405 | 10.0k | *buf_size -= zstream->avail_out; |
406 | 10.0k | if (res != Z_STREAM_END) |
407 | 0 | return -1; |
408 | 10.0k | return 0; |
409 | 10.0k | } |
410 | | |
411 | | static int encode_bgr(Block * b, const uint8_t * src, int stride) |
412 | 0 | { |
413 | 0 | int i; |
414 | 0 | uint8_t *ptr = b->enc; |
415 | 0 | for (i = 0; i < b->start; i++) |
416 | 0 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
417 | 0 | b->sl_begin = ptr + i * b->width * 3; |
418 | 0 | for (; i < b->start + b->len; i++) |
419 | 0 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
420 | 0 | b->sl_end = ptr + i * b->width * 3; |
421 | 0 | for (; i < b->height; i++) |
422 | 0 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
423 | 0 | b->enc_size = ptr + i * b->width * 3 - b->enc; |
424 | 0 | return b->enc_size; |
425 | 0 | } |
426 | | |
427 | | static inline unsigned pixel_color15(const uint8_t * src) |
428 | 52.4M | { |
429 | 52.4M | return (src[0] >> 3) | ((src[1] & 0xf8) << 2) | ((src[2] & 0xf8) << 7); |
430 | 52.4M | } |
431 | | |
432 | | static inline unsigned int chroma_diff(unsigned int c1, unsigned int c2) |
433 | 4.99G | { |
434 | 19.9G | #define ABSDIFF(a,b) (abs((int)(a)-(int)(b))) |
435 | | |
436 | 4.99G | unsigned int t1 = (c1 & 0x000000ff) + ((c1 & 0x0000ff00) >> 8) + ((c1 & 0x00ff0000) >> 16); |
437 | 4.99G | unsigned int t2 = (c2 & 0x000000ff) + ((c2 & 0x0000ff00) >> 8) + ((c2 & 0x00ff0000) >> 16); |
438 | | |
439 | 4.99G | return ABSDIFF(t1, t2) + ABSDIFF(c1 & 0x000000ff, c2 & 0x000000ff) + |
440 | 4.99G | ABSDIFF((c1 & 0x0000ff00) >> 8 , (c2 & 0x0000ff00) >> 8) + |
441 | 4.99G | ABSDIFF((c1 & 0x00ff0000) >> 16, (c2 & 0x00ff0000) >> 16); |
442 | 4.99G | } |
443 | | |
444 | | static inline int pixel_color7_fast(Palette * palette, unsigned c15) |
445 | 52.4M | { |
446 | 52.4M | return palette->index[c15]; |
447 | 52.4M | } |
448 | | |
449 | | static int pixel_color7_slow(Palette * palette, unsigned color) |
450 | 38.2M | { |
451 | 38.2M | int i, min = 0x7fffffff; |
452 | 38.2M | int minc = -1; |
453 | 4.92G | for (i = 0; i < 128; i++) { |
454 | 4.89G | int c1 = palette->colors[i]; |
455 | 4.89G | int diff = chroma_diff(c1, color); |
456 | 4.89G | if (diff < min) { |
457 | 237M | min = diff; |
458 | 237M | minc = i; |
459 | 237M | } |
460 | 4.89G | } |
461 | 38.2M | return minc; |
462 | 38.2M | } |
463 | | |
464 | | static inline unsigned pixel_bgr(const uint8_t * src) |
465 | 52.4M | { |
466 | 52.4M | return (src[0]) | (src[1] << 8) | (src[2] << 16); |
467 | 52.4M | } |
468 | | |
469 | | static int write_pixel_15_7(Palette * palette, uint8_t * dest, const uint8_t * src, |
470 | | int dist) |
471 | 52.4M | { |
472 | 52.4M | unsigned c15 = pixel_color15(src); |
473 | 52.4M | unsigned color = pixel_bgr(src); |
474 | 52.4M | int d15 = chroma_diff(color, color & 0x00f8f8f8); |
475 | 52.4M | int c7 = pixel_color7_fast(palette, c15); |
476 | 52.4M | int d7 = chroma_diff(color, palette->colors[c7]); |
477 | 52.4M | if (dist + d15 >= d7) { |
478 | 42.8M | dest[0] = c7; |
479 | 42.8M | return 1; |
480 | 42.8M | } else { |
481 | 9.54M | dest[0] = 0x80 | (c15 >> 8); |
482 | 9.54M | dest[1] = c15 & 0xff; |
483 | 9.54M | return 2; |
484 | 9.54M | } |
485 | 52.4M | } |
486 | | |
487 | | static int update_palette_index(Palette * palette) |
488 | 1.16k | { |
489 | 1.16k | int r, g, b; |
490 | 1.16k | unsigned int bgr, c15, index; |
491 | 38.4k | for (r = 4; r < 256; r += 8) { |
492 | 1.23M | for (g = 4; g < 256; g += 8) { |
493 | 39.4M | for (b = 4; b < 256; b += 8) { |
494 | 38.2M | bgr = b | (g << 8) | (r << 16); |
495 | 38.2M | c15 = (b >> 3) | ((g & 0xf8) << 2) | ((r & 0xf8) << 7); |
496 | 38.2M | index = pixel_color7_slow(palette, bgr); |
497 | | |
498 | 38.2M | palette->index[c15] = index; |
499 | 38.2M | } |
500 | 1.19M | } |
501 | 37.3k | } |
502 | 1.16k | return 0; |
503 | 1.16k | } |
504 | | |
505 | | static const unsigned int default_screen_video_v2_palette[128] = { |
506 | | 0x00000000, 0x00333333, 0x00666666, 0x00999999, 0x00CCCCCC, 0x00FFFFFF, |
507 | | 0x00330000, 0x00660000, 0x00990000, 0x00CC0000, 0x00FF0000, 0x00003300, |
508 | | 0x00006600, 0x00009900, 0x0000CC00, 0x0000FF00, 0x00000033, 0x00000066, |
509 | | 0x00000099, 0x000000CC, 0x000000FF, 0x00333300, 0x00666600, 0x00999900, |
510 | | 0x00CCCC00, 0x00FFFF00, 0x00003333, 0x00006666, 0x00009999, 0x0000CCCC, |
511 | | 0x0000FFFF, 0x00330033, 0x00660066, 0x00990099, 0x00CC00CC, 0x00FF00FF, |
512 | | 0x00FFFF33, 0x00FFFF66, 0x00FFFF99, 0x00FFFFCC, 0x00FF33FF, 0x00FF66FF, |
513 | | 0x00FF99FF, 0x00FFCCFF, 0x0033FFFF, 0x0066FFFF, 0x0099FFFF, 0x00CCFFFF, |
514 | | 0x00CCCC33, 0x00CCCC66, 0x00CCCC99, 0x00CCCCFF, 0x00CC33CC, 0x00CC66CC, |
515 | | 0x00CC99CC, 0x00CCFFCC, 0x0033CCCC, 0x0066CCCC, 0x0099CCCC, 0x00FFCCCC, |
516 | | 0x00999933, 0x00999966, 0x009999CC, 0x009999FF, 0x00993399, 0x00996699, |
517 | | 0x0099CC99, 0x0099FF99, 0x00339999, 0x00669999, 0x00CC9999, 0x00FF9999, |
518 | | 0x00666633, 0x00666699, 0x006666CC, 0x006666FF, 0x00663366, 0x00669966, |
519 | | 0x0066CC66, 0x0066FF66, 0x00336666, 0x00996666, 0x00CC6666, 0x00FF6666, |
520 | | 0x00333366, 0x00333399, 0x003333CC, 0x003333FF, 0x00336633, 0x00339933, |
521 | | 0x0033CC33, 0x0033FF33, 0x00663333, 0x00993333, 0x00CC3333, 0x00FF3333, |
522 | | 0x00003366, 0x00336600, 0x00660033, 0x00006633, 0x00330066, 0x00663300, |
523 | | 0x00336699, 0x00669933, 0x00993366, 0x00339966, 0x00663399, 0x00996633, |
524 | | 0x006699CC, 0x0099CC66, 0x00CC6699, 0x0066CC99, 0x009966CC, 0x00CC9966, |
525 | | 0x0099CCFF, 0x00CCFF99, 0x00FF99CC, 0x0099FFCC, 0x00CC99FF, 0x00FFCC99, |
526 | | 0x00111111, 0x00222222, 0x00444444, 0x00555555, 0x00AAAAAA, 0x00BBBBBB, |
527 | | 0x00DDDDDD, 0x00EEEEEE |
528 | | }; |
529 | | |
530 | | static int generate_default_palette(Palette * palette) |
531 | 1.16k | { |
532 | 1.16k | memcpy(palette->colors, default_screen_video_v2_palette, |
533 | 1.16k | sizeof(default_screen_video_v2_palette)); |
534 | | |
535 | 1.16k | return update_palette_index(palette); |
536 | 1.16k | } |
537 | | |
538 | | static int generate_optimum_palette(Palette * palette, const uint8_t * image, |
539 | | int width, int height, int stride) |
540 | 0 | { |
541 | | //this isn't implemented yet! Default palette only! |
542 | 0 | return -1; |
543 | 0 | } |
544 | | |
545 | | static inline int encode_15_7_sl(Palette * palette, uint8_t * dest, |
546 | | const uint8_t * src, int width, int dist) |
547 | 999k | { |
548 | 999k | int len = 0, x; |
549 | 53.4M | for (x = 0; x < width; x++) { |
550 | 52.4M | len += write_pixel_15_7(palette, dest + len, src + 3 * x, dist); |
551 | 52.4M | } |
552 | 999k | return len; |
553 | 999k | } |
554 | | |
555 | | static int encode_15_7(Palette * palette, Block * b, const uint8_t * src, |
556 | | int stride, int dist) |
557 | 24.0k | { |
558 | 24.0k | int i; |
559 | 24.0k | uint8_t *ptr = b->enc; |
560 | 40.9k | for (i = 0; i < b->start; i++) |
561 | 16.9k | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
562 | 24.0k | b->sl_begin = ptr; |
563 | 992k | for (; i < b->start + b->len; i++) |
564 | 968k | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
565 | 24.0k | b->sl_end = ptr; |
566 | 38.0k | for (; i < b->height; i++) |
567 | 14.0k | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
568 | 24.0k | b->enc_size = ptr - b->enc; |
569 | 24.0k | return b->enc_size; |
570 | 24.0k | } |
571 | | |
572 | | static int encode_block(FlashSV2Context *s, Palette * palette, Block * b, |
573 | | Block *prev, const uint8_t *src, int stride, |
574 | | int dist, int keyframe) |
575 | 24.0k | { |
576 | 24.0k | unsigned buf_size = b->width * b->height * 6; |
577 | 24.0k | uint8_t *buf = s->blockbuffer; |
578 | 24.0k | int res; |
579 | | |
580 | 24.0k | if (b->flags & COLORSPACE_15_7) { |
581 | 24.0k | encode_15_7(palette, b, src, stride, dist); |
582 | 24.0k | } else { |
583 | 0 | encode_bgr(b, src, stride); |
584 | 0 | } |
585 | | |
586 | 24.0k | if (b->len > 0) { |
587 | 23.9k | b->data_size = buf_size; |
588 | 23.9k | res = encode_zlib(b, b->data, &b->data_size, &s->zstream.zstream); |
589 | 23.9k | if (res) |
590 | 26 | return res; |
591 | | |
592 | 23.9k | if (!keyframe) { |
593 | 10.0k | res = encode_zlibprime(b, prev, buf, &buf_size, &s->zstream.zstream); |
594 | 10.0k | if (res) |
595 | 0 | return res; |
596 | | |
597 | 10.0k | if (buf_size < b->data_size) { |
598 | 9.65k | b->data_size = buf_size; |
599 | 9.65k | memcpy(b->data, buf, buf_size); |
600 | 9.65k | b->flags |= ZLIB_PRIME_COMPRESS_PREVIOUS; |
601 | 9.65k | } |
602 | 10.0k | } |
603 | 23.9k | } else { |
604 | 126 | b->data_size = 0; |
605 | 126 | } |
606 | 24.0k | return 0; |
607 | 24.0k | } |
608 | | |
609 | | static int compare_sl(FlashSV2Context * s, Block * b, const uint8_t * src, |
610 | | uint8_t * frame, uint8_t * key, int y, int keyframe) |
611 | 1.00M | { |
612 | 1.00M | if (memcmp(src, frame, b->width * 3) != 0) { |
613 | 267k | b->dirty = 1; |
614 | 267k | memcpy(frame, src, b->width * 3); |
615 | | #ifndef FLASHSV2_DUMB |
616 | | s->diff_lines++; |
617 | | #endif |
618 | 267k | } |
619 | 1.00M | if (memcmp(src, key, b->width * 3) != 0) { |
620 | 287k | if (b->len == 0) |
621 | 15.6k | b->start = y; |
622 | 287k | b->len = y + 1 - b->start; |
623 | 287k | } |
624 | 1.00M | return 0; |
625 | 1.00M | } |
626 | | |
627 | | static int mark_all_blocks(FlashSV2Context * s, const uint8_t * src, int stride, |
628 | | int keyframe) |
629 | 9.57k | { |
630 | 9.57k | int sl, rsl, col, pos, possl; |
631 | 9.57k | Block *b; |
632 | 449k | for (sl = s->image_height - 1; sl >= 0; sl--) { |
633 | 1.44M | for (col = 0; col < s->cols; col++) { |
634 | 1.00M | rsl = s->image_height - sl - 1; |
635 | 1.00M | b = s->frame_blocks + col + rsl / s->block_height * s->cols; |
636 | 1.00M | possl = stride * sl + col * s->block_width * 3; |
637 | 1.00M | pos = s->image_width * rsl * 3 + col * s->block_width * 3; |
638 | 1.00M | compare_sl(s, b, src + possl, s->current_frame + pos, |
639 | 1.00M | s->key_frame + pos, rsl % s->block_height, keyframe); |
640 | 1.00M | } |
641 | 439k | } |
642 | | #ifndef FLASHSV2_DUMB |
643 | | s->tot_lines += s->image_height * s->cols; |
644 | | #endif |
645 | 9.57k | return 0; |
646 | 9.57k | } |
647 | | |
648 | | static int encode_all_blocks(FlashSV2Context * s, int keyframe) |
649 | 9.57k | { |
650 | 9.57k | int row, col, res; |
651 | 9.57k | uint8_t *data; |
652 | 9.57k | Block *b, *prev; |
653 | 21.6k | for (row = 0; row < s->rows; row++) { |
654 | 36.4k | for (col = 0; col < s->cols; col++) { |
655 | 24.4k | b = s->frame_blocks + (row * s->cols + col); |
656 | 24.4k | prev = s->key_blocks + (row * s->cols + col); |
657 | 24.4k | b->flags = s->use15_7 ? COLORSPACE_15_7 : 0; |
658 | 24.4k | if (keyframe) { |
659 | 13.9k | b->start = 0; |
660 | 13.9k | b->len = b->height; |
661 | 13.9k | } else if (!b->dirty) { |
662 | 358 | b->start = 0; |
663 | 358 | b->len = 0; |
664 | 358 | b->data_size = 0; |
665 | 358 | continue; |
666 | 10.1k | } else if (b->start != 0 || b->len != b->height) { |
667 | 3.35k | b->flags |= HAS_DIFF_BLOCKS; |
668 | 3.35k | } |
669 | 24.0k | data = s->current_frame + s->image_width * 3 * s->block_height * row + s->block_width * col * 3; |
670 | 24.0k | res = encode_block(s, &s->palette, b, prev, data, |
671 | 24.0k | s->image_width * 3, s->dist, keyframe); |
672 | | #ifndef FLASHSV2_DUMB |
673 | | if (b->dirty) |
674 | | s->diff_blocks++; |
675 | | s->comp_size += b->data_size; |
676 | | s->uncomp_size += b->enc_size; |
677 | | #endif |
678 | 24.0k | if (res) |
679 | 26 | return res; |
680 | 24.0k | } |
681 | 12.0k | } |
682 | | #ifndef FLASHSV2_DUMB |
683 | | s->raw_size += s->image_width * s->image_height * 3; |
684 | | s->tot_blocks += s->rows * s->cols; |
685 | | #endif |
686 | 9.54k | return 0; |
687 | 9.57k | } |
688 | | |
689 | | static int write_all_blocks(FlashSV2Context * s, uint8_t * buf, |
690 | | int buf_size) |
691 | 9.54k | { |
692 | 9.54k | int row, col, buf_pos = 0, len; |
693 | 9.54k | Block *b; |
694 | 21.5k | for (row = 0; row < s->rows; row++) { |
695 | 35.7k | for (col = 0; col < s->cols; col++) { |
696 | 23.7k | b = s->frame_blocks + row * s->cols + col; |
697 | 23.7k | len = write_block(b, buf + buf_pos, buf_size - buf_pos); |
698 | 23.7k | b->start = b->len = b->dirty = 0; |
699 | 23.7k | if (len < 0) |
700 | 0 | return len; |
701 | 23.7k | buf_pos += len; |
702 | 23.7k | } |
703 | 11.9k | } |
704 | 9.54k | return buf_pos; |
705 | 9.54k | } |
706 | | |
707 | | static int write_bitstream(FlashSV2Context * s, const uint8_t * src, int stride, |
708 | | uint8_t * buf, int buf_size, int keyframe) |
709 | 9.57k | { |
710 | 9.57k | int buf_pos, res; |
711 | | |
712 | 9.57k | res = mark_all_blocks(s, src, stride, keyframe); |
713 | 9.57k | if (res) |
714 | 0 | return res; |
715 | 9.57k | res = encode_all_blocks(s, keyframe); |
716 | 9.57k | if (res) |
717 | 26 | return res; |
718 | | |
719 | 9.54k | res = write_header(s, buf, buf_size); |
720 | 9.54k | if (res < 0) { |
721 | 0 | return res; |
722 | 9.54k | } else { |
723 | 9.54k | buf_pos = res; |
724 | 9.54k | } |
725 | 9.54k | res = write_all_blocks(s, buf + buf_pos, buf_size - buf_pos); |
726 | 9.54k | if (res < 0) |
727 | 0 | return res; |
728 | 9.54k | buf_pos += res; |
729 | | #ifndef FLASHSV2_DUMB |
730 | | s->total_bits += ((double) buf_pos) * 8.0; |
731 | | #endif |
732 | | |
733 | 9.54k | return buf_pos; |
734 | 9.54k | } |
735 | | |
736 | | static void recommend_keyframe(FlashSV2Context * s, int *keyframe) |
737 | 3.66k | { |
738 | | #ifndef FLASHSV2_DUMB |
739 | | double block_ratio, line_ratio, enc_ratio, comp_ratio, data_ratio; |
740 | | if (s->avctx->gop_size > 0) { |
741 | | block_ratio = s->diff_blocks / s->tot_blocks; |
742 | | line_ratio = s->diff_lines / s->tot_lines; |
743 | | enc_ratio = s->uncomp_size / s->raw_size; |
744 | | comp_ratio = s->comp_size / s->uncomp_size; |
745 | | data_ratio = s->comp_size / s->raw_size; |
746 | | |
747 | | if ((block_ratio >= 0.5 && line_ratio / block_ratio <= 0.5) || line_ratio >= 0.95) { |
748 | | *keyframe = 1; |
749 | | return; |
750 | | } |
751 | | } |
752 | | #else |
753 | 3.66k | return; |
754 | 3.66k | #endif |
755 | 3.66k | } |
756 | | |
757 | | #ifndef FLASHSV2_DUMB |
758 | | static const double block_size_fraction = 1.0 / 300; |
759 | | static const double use15_7_threshold = 8192; |
760 | | static const double color15_7_factor = 100; |
761 | | #endif |
762 | | static int optimum_block_width(FlashSV2Context * s) |
763 | 1.37k | { |
764 | | #ifndef FLASHSV2_DUMB |
765 | | double save = (1-pow(s->diff_lines/s->diff_blocks/s->block_height, 0.5)) * s->comp_size/s->tot_blocks; |
766 | | double width = block_size_fraction * sqrt(0.5 * save * s->rows * s->cols) * s->image_width; |
767 | | int pwidth = ((int) width); |
768 | | return FFCLIP(pwidth & ~15, 256, 16); |
769 | | #else |
770 | 1.37k | return 64; |
771 | 1.37k | #endif |
772 | 1.37k | } |
773 | | |
774 | | static int optimum_block_height(FlashSV2Context * s) |
775 | 1.37k | { |
776 | | #ifndef FLASHSV2_DUMB |
777 | | double save = (1-pow(s->diff_lines/s->diff_blocks/s->block_height, 0.5)) * s->comp_size/s->tot_blocks; |
778 | | double height = block_size_fraction * sqrt(0.5 * save * s->rows * s->cols) * s->image_height; |
779 | | int pheight = ((int) height); |
780 | | return FFCLIP(pheight & ~15, 256, 16); |
781 | | #else |
782 | 1.37k | return 64; |
783 | 1.37k | #endif |
784 | 1.37k | } |
785 | | |
786 | | static int optimum_use15_7(FlashSV2Context * s) |
787 | 1.37k | { |
788 | | #ifndef FLASHSV2_DUMB |
789 | | double ideal = ((double)(s->avctx->bit_rate * s->avctx->time_base.den * s->avctx->ticks_per_frame)) / |
790 | | ((double) s->avctx->time_base.num) * s->avctx->frame_num; |
791 | | if (ideal + use15_7_threshold < s->total_bits) { |
792 | | return 1; |
793 | | } else { |
794 | | return 0; |
795 | | } |
796 | | #else |
797 | 1.37k | return s->avctx->global_quality == 0; |
798 | 1.37k | #endif |
799 | 1.37k | } |
800 | | |
801 | | static int optimum_dist(FlashSV2Context * s) |
802 | 9.57k | { |
803 | | #ifndef FLASHSV2_DUMB |
804 | | double ideal = |
805 | | s->avctx->bit_rate * s->avctx->time_base.den * |
806 | | s->avctx->ticks_per_frame; |
807 | | int dist = pow((s->total_bits / ideal) * color15_7_factor, 3); |
808 | | av_log(s->avctx, AV_LOG_DEBUG, "dist: %d\n", dist); |
809 | | return dist; |
810 | | #else |
811 | 9.57k | return 15; |
812 | 9.57k | #endif |
813 | 9.57k | } |
814 | | |
815 | | |
816 | | static int reconfigure_at_keyframe(FlashSV2Context * s, const uint8_t * image, |
817 | | int stride) |
818 | 1.37k | { |
819 | 1.37k | int update_palette = 0; |
820 | 1.37k | int res; |
821 | 1.37k | int block_width = optimum_block_width (s); |
822 | 1.37k | int block_height = optimum_block_height(s); |
823 | | |
824 | 1.37k | if (block_width != s->block_width || block_height != s->block_height) { |
825 | 0 | res = update_block_dimensions(s, block_width, block_height); |
826 | 0 | if (res < 0) |
827 | 0 | return res; |
828 | 0 | } |
829 | | |
830 | 1.37k | s->use15_7 = optimum_use15_7(s); |
831 | 1.37k | if (s->use15_7) { |
832 | 1.37k | if ((s->use_custom_palette && s->palette_type != 1) || update_palette) { |
833 | 0 | res = generate_optimum_palette(&s->palette, image, s->image_width, s->image_height, stride); |
834 | 0 | if (res) |
835 | 0 | return res; |
836 | 0 | s->palette_type = 1; |
837 | 0 | av_log(s->avctx, AV_LOG_DEBUG, "Generated optimum palette\n"); |
838 | 1.37k | } else if (!s->use_custom_palette && s->palette_type != 0) { |
839 | 1.16k | res = generate_default_palette(&s->palette); |
840 | 1.16k | if (res) |
841 | 0 | return res; |
842 | 1.16k | s->palette_type = 0; |
843 | 1.16k | av_log(s->avctx, AV_LOG_DEBUG, "Generated default palette\n"); |
844 | 1.16k | } |
845 | 1.37k | } |
846 | | |
847 | | |
848 | 1.37k | reset_stats(s); |
849 | | |
850 | 1.37k | return 0; |
851 | 1.37k | } |
852 | | |
853 | | static int flashsv2_encode_frame(AVCodecContext *avctx, AVPacket *pkt, |
854 | | const AVFrame *p, int *got_packet) |
855 | 9.57k | { |
856 | 9.57k | FlashSV2Context *const s = avctx->priv_data; |
857 | 9.57k | int res; |
858 | 9.57k | int keyframe = 0; |
859 | | |
860 | 9.57k | if ((res = ff_alloc_packet(avctx, pkt, s->frame_size + FF_INPUT_BUFFER_MIN_SIZE)) < 0) |
861 | 0 | return res; |
862 | | |
863 | | /* First frame needs to be a keyframe */ |
864 | 9.57k | if (avctx->frame_num == 0) |
865 | 1.16k | keyframe = 1; |
866 | | |
867 | | /* Check the placement of keyframes */ |
868 | 9.57k | if (avctx->gop_size > 0) { |
869 | 8.42k | if (avctx->frame_num >= s->last_key_frame + avctx->gop_size) |
870 | 211 | keyframe = 1; |
871 | 8.42k | } |
872 | | |
873 | 9.57k | if (!keyframe |
874 | 9.57k | && avctx->frame_num > s->last_key_frame + avctx->keyint_min) { |
875 | 3.66k | recommend_keyframe(s, &keyframe); |
876 | 3.66k | if (keyframe) |
877 | 0 | av_log(avctx, AV_LOG_DEBUG, "Recommending key frame at frame %"PRId64"\n", avctx->frame_num); |
878 | 3.66k | } |
879 | | |
880 | 9.57k | if (keyframe) { |
881 | 1.37k | res = reconfigure_at_keyframe(s, p->data[0], p->linesize[0]); |
882 | 1.37k | if (res) |
883 | 0 | return res; |
884 | 1.37k | } |
885 | | |
886 | 9.57k | if (s->use15_7) |
887 | 9.57k | s->dist = optimum_dist(s); |
888 | | |
889 | 9.57k | res = write_bitstream(s, p->data[0], p->linesize[0], pkt->data, pkt->size, keyframe); |
890 | | |
891 | 9.57k | if (keyframe) { |
892 | 1.37k | new_key_frame(s); |
893 | 1.37k | s->last_key_frame = avctx->frame_num; |
894 | 1.37k | pkt->flags |= AV_PKT_FLAG_KEY; |
895 | 1.37k | av_log(avctx, AV_LOG_DEBUG, "Inserting key frame at frame %"PRId64"\n", avctx->frame_num); |
896 | 1.37k | } |
897 | | |
898 | 9.57k | pkt->size = res; |
899 | 9.57k | *got_packet = 1; |
900 | | |
901 | 9.57k | return 0; |
902 | 9.57k | } |
903 | | |
904 | | static av_cold int flashsv2_encode_end(AVCodecContext * avctx) |
905 | 1.21k | { |
906 | 1.21k | FlashSV2Context *s = avctx->priv_data; |
907 | | |
908 | 1.21k | cleanup(s); |
909 | | |
910 | 1.21k | return 0; |
911 | 1.21k | } |
912 | | |
913 | | const FFCodec ff_flashsv2_encoder = { |
914 | | .p.name = "flashsv2", |
915 | | CODEC_LONG_NAME("Flash Screen Video Version 2"), |
916 | | .p.type = AVMEDIA_TYPE_VIDEO, |
917 | | .p.id = AV_CODEC_ID_FLASHSV2, |
918 | | .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE, |
919 | | .priv_data_size = sizeof(FlashSV2Context), |
920 | | .init = flashsv2_encode_init, |
921 | | FF_CODEC_ENCODE_CB(flashsv2_encode_frame), |
922 | | .close = flashsv2_encode_end, |
923 | | .p.pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_BGR24, AV_PIX_FMT_NONE }, |
924 | | .caps_internal = FF_CODEC_CAP_INIT_CLEANUP, |
925 | | }; |