blob: cc917edb7fe47ad65cce6bc6676f007890bd5c44
1 | /* |
2 | * MJPEG encoder |
3 | * Copyright (c) 2000, 2001 Fabrice Bellard |
4 | * Copyright (c) 2003 Alex Beregszaszi |
5 | * Copyright (c) 2003-2004 Michael Niedermayer |
6 | * |
7 | * Support for external huffman table, various fixes (AVID workaround), |
8 | * aspecting, new decode_frame mechanism and apple mjpeg-b support |
9 | * by Alex Beregszaszi |
10 | * |
11 | * This file is part of FFmpeg. |
12 | * |
13 | * FFmpeg is free software; you can redistribute it and/or |
14 | * modify it under the terms of the GNU Lesser General Public |
15 | * License as published by the Free Software Foundation; either |
16 | * version 2.1 of the License, or (at your option) any later version. |
17 | * |
18 | * FFmpeg is distributed in the hope that it will be useful, |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
21 | * Lesser General Public License for more details. |
22 | * |
23 | * You should have received a copy of the GNU Lesser General Public |
24 | * License along with FFmpeg; if not, write to the Free Software |
25 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
26 | */ |
27 | |
28 | /** |
29 | * @file |
30 | * MJPEG encoder. |
31 | */ |
32 | |
33 | #include "libavutil/pixdesc.h" |
34 | |
35 | #include "avcodec.h" |
36 | #include "jpegtables.h" |
37 | #include "mjpegenc_common.h" |
38 | #include "mpegvideo.h" |
39 | #include "mjpeg.h" |
40 | #include "mjpegenc.h" |
41 | |
42 | |
43 | static int alloc_huffman(MpegEncContext *s) |
44 | { |
45 | MJpegContext *m = s->mjpeg_ctx; |
46 | size_t num_mbs, num_blocks, num_codes; |
47 | int blocks_per_mb; |
48 | |
49 | // We need to init this here as the mjpeg init is called before the common init, |
50 | s->mb_width = (s->width + 15) / 16; |
51 | s->mb_height = (s->height + 15) / 16; |
52 | |
53 | switch (s->chroma_format) { |
54 | case CHROMA_420: blocks_per_mb = 6; break; |
55 | case CHROMA_422: blocks_per_mb = 8; break; |
56 | case CHROMA_444: blocks_per_mb = 12; break; |
57 | default: av_assert0(0); |
58 | }; |
59 | |
60 | // Make sure we have enough space to hold this frame. |
61 | num_mbs = s->mb_width * s->mb_height; |
62 | num_blocks = num_mbs * blocks_per_mb; |
63 | num_codes = num_blocks * 64; |
64 | |
65 | m->huff_buffer = av_malloc_array(num_codes, sizeof(MJpegHuffmanCode)); |
66 | if (!m->huff_buffer) |
67 | return AVERROR(ENOMEM); |
68 | return 0; |
69 | } |
70 | |
71 | av_cold int ff_mjpeg_encode_init(MpegEncContext *s) |
72 | { |
73 | MJpegContext *m; |
74 | |
75 | av_assert0(s->slice_context_count == 1); |
76 | |
77 | if (s->width > 65500 || s->height > 65500) { |
78 | av_log(s, AV_LOG_ERROR, "JPEG does not support resolutions above 65500x65500\n"); |
79 | return AVERROR(EINVAL); |
80 | } |
81 | |
82 | m = av_mallocz(sizeof(MJpegContext)); |
83 | if (!m) |
84 | return AVERROR(ENOMEM); |
85 | |
86 | s->min_qcoeff=-1023; |
87 | s->max_qcoeff= 1023; |
88 | |
89 | // Build default Huffman tables. |
90 | // These may be overwritten later with more optimal Huffman tables, but |
91 | // they are needed at least right now for some processes like trellis. |
92 | ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance, |
93 | m->huff_code_dc_luminance, |
94 | avpriv_mjpeg_bits_dc_luminance, |
95 | avpriv_mjpeg_val_dc); |
96 | ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance, |
97 | m->huff_code_dc_chrominance, |
98 | avpriv_mjpeg_bits_dc_chrominance, |
99 | avpriv_mjpeg_val_dc); |
100 | ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance, |
101 | m->huff_code_ac_luminance, |
102 | avpriv_mjpeg_bits_ac_luminance, |
103 | avpriv_mjpeg_val_ac_luminance); |
104 | ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance, |
105 | m->huff_code_ac_chrominance, |
106 | avpriv_mjpeg_bits_ac_chrominance, |
107 | avpriv_mjpeg_val_ac_chrominance); |
108 | |
109 | ff_init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len); |
110 | ff_init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len); |
111 | s->intra_ac_vlc_length = |
112 | s->intra_ac_vlc_last_length = m->uni_ac_vlc_len; |
113 | s->intra_chroma_ac_vlc_length = |
114 | s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len; |
115 | |
116 | // Buffers start out empty. |
117 | m->huff_ncode = 0; |
118 | s->mjpeg_ctx = m; |
119 | |
120 | if(s->huffman == HUFFMAN_TABLE_OPTIMAL) |
121 | return alloc_huffman(s); |
122 | |
123 | return 0; |
124 | } |
125 | |
126 | av_cold void ff_mjpeg_encode_close(MpegEncContext *s) |
127 | { |
128 | av_freep(&s->mjpeg_ctx->huff_buffer); |
129 | av_freep(&s->mjpeg_ctx); |
130 | } |
131 | |
132 | /** |
133 | * Encodes and outputs the entire frame in the JPEG format. |
134 | * |
135 | * @param s The MpegEncContext. |
136 | */ |
137 | void ff_mjpeg_encode_picture_frame(MpegEncContext *s) |
138 | { |
139 | int i, nbits, code, table_id; |
140 | MJpegContext *m = s->mjpeg_ctx; |
141 | uint8_t *huff_size[4] = {m->huff_size_dc_luminance, |
142 | m->huff_size_dc_chrominance, |
143 | m->huff_size_ac_luminance, |
144 | m->huff_size_ac_chrominance}; |
145 | uint16_t *huff_code[4] = {m->huff_code_dc_luminance, |
146 | m->huff_code_dc_chrominance, |
147 | m->huff_code_ac_luminance, |
148 | m->huff_code_ac_chrominance}; |
149 | size_t total_bits = 0; |
150 | size_t bytes_needed; |
151 | |
152 | s->header_bits = get_bits_diff(s); |
153 | // Estimate the total size first |
154 | for (i = 0; i < m->huff_ncode; i++) { |
155 | table_id = m->huff_buffer[i].table_id; |
156 | code = m->huff_buffer[i].code; |
157 | nbits = code & 0xf; |
158 | |
159 | total_bits += huff_size[table_id][code] + nbits; |
160 | } |
161 | |
162 | bytes_needed = (total_bits + 7) / 8; |
163 | ff_mpv_reallocate_putbitbuffer(s, bytes_needed, bytes_needed); |
164 | |
165 | for (i = 0; i < m->huff_ncode; i++) { |
166 | table_id = m->huff_buffer[i].table_id; |
167 | code = m->huff_buffer[i].code; |
168 | nbits = code & 0xf; |
169 | |
170 | put_bits(&s->pb, huff_size[table_id][code], huff_code[table_id][code]); |
171 | if (nbits != 0) { |
172 | put_sbits(&s->pb, nbits, m->huff_buffer[i].mant); |
173 | } |
174 | } |
175 | |
176 | m->huff_ncode = 0; |
177 | s->i_tex_bits = get_bits_diff(s); |
178 | } |
179 | |
180 | /** |
181 | * Add code and table_id to the JPEG buffer. |
182 | * |
183 | * @param s The MJpegContext which contains the JPEG buffer. |
184 | * @param table_id Which Huffman table the code belongs to. |
185 | * @param code The encoded exponent of the coefficients and the run-bits. |
186 | */ |
187 | static inline void ff_mjpeg_encode_code(MJpegContext *s, uint8_t table_id, int code) |
188 | { |
189 | MJpegHuffmanCode *c = &s->huff_buffer[s->huff_ncode++]; |
190 | c->table_id = table_id; |
191 | c->code = code; |
192 | } |
193 | |
194 | /** |
195 | * Add the coefficient's data to the JPEG buffer. |
196 | * |
197 | * @param s The MJpegContext which contains the JPEG buffer. |
198 | * @param table_id Which Huffman table the code belongs to. |
199 | * @param val The coefficient. |
200 | * @param run The run-bits. |
201 | */ |
202 | static void ff_mjpeg_encode_coef(MJpegContext *s, uint8_t table_id, int val, int run) |
203 | { |
204 | int mant, code; |
205 | |
206 | if (val == 0) { |
207 | av_assert0(run == 0); |
208 | ff_mjpeg_encode_code(s, table_id, 0); |
209 | } else { |
210 | mant = val; |
211 | if (val < 0) { |
212 | val = -val; |
213 | mant--; |
214 | } |
215 | |
216 | code = (run << 4) | (av_log2_16bit(val) + 1); |
217 | |
218 | s->huff_buffer[s->huff_ncode].mant = mant; |
219 | ff_mjpeg_encode_code(s, table_id, code); |
220 | } |
221 | } |
222 | |
223 | /** |
224 | * Add the block's data into the JPEG buffer. |
225 | * |
226 | * @param s The MJpegEncContext that contains the JPEG buffer. |
227 | * @param block The block. |
228 | * @param n The block's index or number. |
229 | */ |
230 | static void record_block(MpegEncContext *s, int16_t *block, int n) |
231 | { |
232 | int i, j, table_id; |
233 | int component, dc, last_index, val, run; |
234 | MJpegContext *m = s->mjpeg_ctx; |
235 | |
236 | /* DC coef */ |
237 | component = (n <= 3 ? 0 : (n&1) + 1); |
238 | table_id = (n <= 3 ? 0 : 1); |
239 | dc = block[0]; /* overflow is impossible */ |
240 | val = dc - s->last_dc[component]; |
241 | |
242 | ff_mjpeg_encode_coef(m, table_id, val, 0); |
243 | |
244 | s->last_dc[component] = dc; |
245 | |
246 | /* AC coefs */ |
247 | |
248 | run = 0; |
249 | last_index = s->block_last_index[n]; |
250 | table_id |= 2; |
251 | |
252 | for(i=1;i<=last_index;i++) { |
253 | j = s->intra_scantable.permutated[i]; |
254 | val = block[j]; |
255 | |
256 | if (val == 0) { |
257 | run++; |
258 | } else { |
259 | while (run >= 16) { |
260 | ff_mjpeg_encode_code(m, table_id, 0xf0); |
261 | run -= 16; |
262 | } |
263 | ff_mjpeg_encode_coef(m, table_id, val, run); |
264 | run = 0; |
265 | } |
266 | } |
267 | |
268 | /* output EOB only if not already 64 values */ |
269 | if (last_index < 63 || run != 0) |
270 | ff_mjpeg_encode_code(m, table_id, 0); |
271 | } |
272 | |
273 | static void encode_block(MpegEncContext *s, int16_t *block, int n) |
274 | { |
275 | int mant, nbits, code, i, j; |
276 | int component, dc, run, last_index, val; |
277 | MJpegContext *m = s->mjpeg_ctx; |
278 | uint8_t *huff_size_ac; |
279 | uint16_t *huff_code_ac; |
280 | |
281 | /* DC coef */ |
282 | component = (n <= 3 ? 0 : (n&1) + 1); |
283 | dc = block[0]; /* overflow is impossible */ |
284 | val = dc - s->last_dc[component]; |
285 | if (n < 4) { |
286 | ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_luminance, m->huff_code_dc_luminance); |
287 | huff_size_ac = m->huff_size_ac_luminance; |
288 | huff_code_ac = m->huff_code_ac_luminance; |
289 | } else { |
290 | ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_chrominance, m->huff_code_dc_chrominance); |
291 | huff_size_ac = m->huff_size_ac_chrominance; |
292 | huff_code_ac = m->huff_code_ac_chrominance; |
293 | } |
294 | s->last_dc[component] = dc; |
295 | |
296 | /* AC coefs */ |
297 | |
298 | run = 0; |
299 | last_index = s->block_last_index[n]; |
300 | for(i=1;i<=last_index;i++) { |
301 | j = s->intra_scantable.permutated[i]; |
302 | val = block[j]; |
303 | if (val == 0) { |
304 | run++; |
305 | } else { |
306 | while (run >= 16) { |
307 | put_bits(&s->pb, huff_size_ac[0xf0], huff_code_ac[0xf0]); |
308 | run -= 16; |
309 | } |
310 | mant = val; |
311 | if (val < 0) { |
312 | val = -val; |
313 | mant--; |
314 | } |
315 | |
316 | nbits= av_log2_16bit(val) + 1; |
317 | code = (run << 4) | nbits; |
318 | |
319 | put_bits(&s->pb, huff_size_ac[code], huff_code_ac[code]); |
320 | |
321 | put_sbits(&s->pb, nbits, mant); |
322 | run = 0; |
323 | } |
324 | } |
325 | |
326 | /* output EOB only if not already 64 values */ |
327 | if (last_index < 63 || run != 0) |
328 | put_bits(&s->pb, huff_size_ac[0], huff_code_ac[0]); |
329 | } |
330 | |
331 | void ff_mjpeg_encode_mb(MpegEncContext *s, int16_t block[12][64]) |
332 | { |
333 | int i; |
334 | if (s->huffman == HUFFMAN_TABLE_OPTIMAL) { |
335 | if (s->chroma_format == CHROMA_444) { |
336 | record_block(s, block[0], 0); |
337 | record_block(s, block[2], 2); |
338 | record_block(s, block[4], 4); |
339 | record_block(s, block[8], 8); |
340 | record_block(s, block[5], 5); |
341 | record_block(s, block[9], 9); |
342 | |
343 | if (16*s->mb_x+8 < s->width) { |
344 | record_block(s, block[1], 1); |
345 | record_block(s, block[3], 3); |
346 | record_block(s, block[6], 6); |
347 | record_block(s, block[10], 10); |
348 | record_block(s, block[7], 7); |
349 | record_block(s, block[11], 11); |
350 | } |
351 | } else { |
352 | for(i=0;i<5;i++) { |
353 | record_block(s, block[i], i); |
354 | } |
355 | if (s->chroma_format == CHROMA_420) { |
356 | record_block(s, block[5], 5); |
357 | } else { |
358 | record_block(s, block[6], 6); |
359 | record_block(s, block[5], 5); |
360 | record_block(s, block[7], 7); |
361 | } |
362 | } |
363 | } else { |
364 | if (s->chroma_format == CHROMA_444) { |
365 | encode_block(s, block[0], 0); |
366 | encode_block(s, block[2], 2); |
367 | encode_block(s, block[4], 4); |
368 | encode_block(s, block[8], 8); |
369 | encode_block(s, block[5], 5); |
370 | encode_block(s, block[9], 9); |
371 | |
372 | if (16*s->mb_x+8 < s->width) { |
373 | encode_block(s, block[1], 1); |
374 | encode_block(s, block[3], 3); |
375 | encode_block(s, block[6], 6); |
376 | encode_block(s, block[10], 10); |
377 | encode_block(s, block[7], 7); |
378 | encode_block(s, block[11], 11); |
379 | } |
380 | } else { |
381 | for(i=0;i<5;i++) { |
382 | encode_block(s, block[i], i); |
383 | } |
384 | if (s->chroma_format == CHROMA_420) { |
385 | encode_block(s, block[5], 5); |
386 | } else { |
387 | encode_block(s, block[6], 6); |
388 | encode_block(s, block[5], 5); |
389 | encode_block(s, block[7], 7); |
390 | } |
391 | } |
392 | |
393 | s->i_tex_bits += get_bits_diff(s); |
394 | } |
395 | } |
396 | |
397 | // maximum over s->mjpeg_vsample[i] |
398 | #define V_MAX 2 |
399 | static int amv_encode_picture(AVCodecContext *avctx, AVPacket *pkt, |
400 | const AVFrame *pic_arg, int *got_packet) |
401 | |
402 | { |
403 | MpegEncContext *s = avctx->priv_data; |
404 | AVFrame *pic; |
405 | int i, ret; |
406 | int chroma_h_shift, chroma_v_shift; |
407 | |
408 | av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift); |
409 | |
410 | #if FF_API_EMU_EDGE |
411 | //CODEC_FLAG_EMU_EDGE have to be cleared |
412 | if(s->avctx->flags & CODEC_FLAG_EMU_EDGE) |
413 | return AVERROR(EINVAL); |
414 | #endif |
415 | |
416 | if ((avctx->height & 15) && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) { |
417 | av_log(avctx, AV_LOG_ERROR, |
418 | "Heights which are not a multiple of 16 might fail with some decoders, " |
419 | "use vstrict=-1 / -strict -1 to use %d anyway.\n", avctx->height); |
420 | av_log(avctx, AV_LOG_WARNING, "If you have a device that plays AMV videos, please test if videos " |
421 | "with such heights work with it and report your findings to ffmpeg-devel@ffmpeg.org\n"); |
422 | return AVERROR_EXPERIMENTAL; |
423 | } |
424 | |
425 | pic = av_frame_clone(pic_arg); |
426 | if (!pic) |
427 | return AVERROR(ENOMEM); |
428 | //picture should be flipped upside-down |
429 | for(i=0; i < 3; i++) { |
430 | int vsample = i ? 2 >> chroma_v_shift : 2; |
431 | pic->data[i] += pic->linesize[i] * (vsample * s->height / V_MAX - 1); |
432 | pic->linesize[i] *= -1; |
433 | } |
434 | ret = ff_mpv_encode_picture(avctx, pkt, pic, got_packet); |
435 | av_frame_free(&pic); |
436 | return ret; |
437 | } |
438 | |
439 | #define OFFSET(x) offsetof(MpegEncContext, x) |
440 | #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
441 | static const AVOption options[] = { |
442 | FF_MPV_COMMON_OPTS |
443 | { "pred", "Prediction method", OFFSET(pred), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 3, VE, "pred" }, |
444 | { "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, INT_MIN, INT_MAX, VE, "pred" }, |
445 | { "plane", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, INT_MIN, INT_MAX, VE, "pred" }, |
446 | { "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 }, INT_MIN, INT_MAX, VE, "pred" }, |
447 | { "huffman", "Huffman table strategy", OFFSET(huffman), AV_OPT_TYPE_INT, { .i64 = HUFFMAN_TABLE_DEFAULT }, 0, NB_HUFFMAN_TABLE_OPTION - 1, VE, "huffman" }, |
448 | { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_DEFAULT }, INT_MIN, INT_MAX, VE, "huffman" }, |
449 | { "optimal", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_OPTIMAL }, INT_MIN, INT_MAX, VE, "huffman" }, |
450 | { NULL}, |
451 | }; |
452 | |
453 | #if CONFIG_MJPEG_ENCODER |
454 | |
455 | static const AVClass mjpeg_class = { |
456 | .class_name = "mjpeg encoder", |
457 | .item_name = av_default_item_name, |
458 | .option = options, |
459 | .version = LIBAVUTIL_VERSION_INT, |
460 | }; |
461 | |
462 | AVCodec ff_mjpeg_encoder = { |
463 | .name = "mjpeg", |
464 | .long_name = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"), |
465 | .type = AVMEDIA_TYPE_VIDEO, |
466 | .id = AV_CODEC_ID_MJPEG, |
467 | .priv_data_size = sizeof(MpegEncContext), |
468 | .init = ff_mpv_encode_init, |
469 | .encode2 = ff_mpv_encode_picture, |
470 | .close = ff_mpv_encode_end, |
471 | .capabilities = AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_INTRA_ONLY, |
472 | .pix_fmts = (const enum AVPixelFormat[]){ |
473 | AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_NONE |
474 | }, |
475 | .priv_class = &mjpeg_class, |
476 | }; |
477 | #endif |
478 | #if CONFIG_AMV_ENCODER |
479 | static const AVClass amv_class = { |
480 | .class_name = "amv encoder", |
481 | .item_name = av_default_item_name, |
482 | .option = options, |
483 | .version = LIBAVUTIL_VERSION_INT, |
484 | }; |
485 | |
486 | AVCodec ff_amv_encoder = { |
487 | .name = "amv", |
488 | .long_name = NULL_IF_CONFIG_SMALL("AMV Video"), |
489 | .type = AVMEDIA_TYPE_VIDEO, |
490 | .id = AV_CODEC_ID_AMV, |
491 | .priv_data_size = sizeof(MpegEncContext), |
492 | .init = ff_mpv_encode_init, |
493 | .encode2 = amv_encode_picture, |
494 | .close = ff_mpv_encode_end, |
495 | .pix_fmts = (const enum AVPixelFormat[]){ |
496 | AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_NONE |
497 | }, |
498 | .priv_class = &amv_class, |
499 | }; |
500 | #endif |
501 |