3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2003 Alex Beregszaszi
5 * Copyright (c) 2003-2004 Michael Niedermayer
7 * Support for external huffman table, various fixes (AVID workaround),
8 * aspecting, new decode_frame mechanism and apple mjpeg-b support
11 * This file is part of FFmpeg.
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.
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.
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
33 #include "libavutil/pixdesc.h"
36 #include "jpegtables.h"
37 #include "mjpegenc_common.h"
38 #include "mjpegenc_huffman.h"
39 #include "mpegvideo.h"
44 static av_cold void init_uni_ac_vlc(const uint8_t huff_size_ac[256],
45 uint8_t *uni_ac_vlc_len)
47 for (int i = 0; i < 128; i++) {
51 for (int run = 0; run < 64; run++) {
53 int alevel = FFABS(level);
55 len = (run >> 4) * huff_size_ac[0xf0];
57 nbits= av_log2_16bit(alevel) + 1;
58 code = ((15&run) << 4) | nbits;
60 len += huff_size_ac[code] + nbits;
62 uni_ac_vlc_len[UNI_AC_ENC_INDEX(run, i)] = len;
63 // We ignore EOB as its just a constant which does not change generally
69 * Encodes and outputs the entire frame in the JPEG format.
71 * @param s The MpegEncContext.
73 static void mjpeg_encode_picture_frame(MpegEncContext *s)
75 int nbits, code, table_id;
76 MJpegContext *m = s->mjpeg_ctx;
77 uint8_t *huff_size[4] = { m->huff_size_dc_luminance,
78 m->huff_size_dc_chrominance,
79 m->huff_size_ac_luminance,
80 m->huff_size_ac_chrominance };
81 uint16_t *huff_code[4] = { m->huff_code_dc_luminance,
82 m->huff_code_dc_chrominance,
83 m->huff_code_ac_luminance,
84 m->huff_code_ac_chrominance };
85 size_t total_bits = 0;
88 s->header_bits = get_bits_diff(s);
89 // Estimate the total size first
90 for (int i = 0; i < m->huff_ncode; i++) {
91 table_id = m->huff_buffer[i].table_id;
92 code = m->huff_buffer[i].code;
95 total_bits += huff_size[table_id][code] + nbits;
98 bytes_needed = (total_bits + 7) / 8;
99 ff_mpv_reallocate_putbitbuffer(s, bytes_needed, bytes_needed);
101 for (int i = 0; i < m->huff_ncode; i++) {
102 table_id = m->huff_buffer[i].table_id;
103 code = m->huff_buffer[i].code;
106 put_bits(&s->pb, huff_size[table_id][code], huff_code[table_id][code]);
108 put_sbits(&s->pb, nbits, m->huff_buffer[i].mant);
113 s->i_tex_bits = get_bits_diff(s);
117 * Builds all 4 optimal Huffman tables.
119 * Uses the data stored in the JPEG buffer to compute the tables.
120 * Stores the Huffman tables in the bits_* and val_* arrays in the MJpegContext.
122 * @param m MJpegContext containing the JPEG buffer.
124 static void mjpeg_build_optimal_huffman(MJpegContext *m)
126 MJpegEncHuffmanContext dc_luminance_ctx;
127 MJpegEncHuffmanContext dc_chrominance_ctx;
128 MJpegEncHuffmanContext ac_luminance_ctx;
129 MJpegEncHuffmanContext ac_chrominance_ctx;
130 MJpegEncHuffmanContext *ctx[4] = { &dc_luminance_ctx,
133 &ac_chrominance_ctx };
134 for (int i = 0; i < 4; i++)
135 ff_mjpeg_encode_huffman_init(ctx[i]);
137 for (int i = 0; i < m->huff_ncode; i++) {
138 int table_id = m->huff_buffer[i].table_id;
139 int code = m->huff_buffer[i].code;
141 ff_mjpeg_encode_huffman_increment(ctx[table_id], code);
144 ff_mjpeg_encode_huffman_close(&dc_luminance_ctx,
145 m->bits_dc_luminance,
146 m->val_dc_luminance, 12);
147 ff_mjpeg_encode_huffman_close(&dc_chrominance_ctx,
148 m->bits_dc_chrominance,
149 m->val_dc_chrominance, 12);
150 ff_mjpeg_encode_huffman_close(&ac_luminance_ctx,
151 m->bits_ac_luminance,
152 m->val_ac_luminance, 256);
153 ff_mjpeg_encode_huffman_close(&ac_chrominance_ctx,
154 m->bits_ac_chrominance,
155 m->val_ac_chrominance, 256);
157 ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance,
158 m->huff_code_dc_luminance,
159 m->bits_dc_luminance,
160 m->val_dc_luminance);
161 ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance,
162 m->huff_code_dc_chrominance,
163 m->bits_dc_chrominance,
164 m->val_dc_chrominance);
165 ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance,
166 m->huff_code_ac_luminance,
167 m->bits_ac_luminance,
168 m->val_ac_luminance);
169 ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance,
170 m->huff_code_ac_chrominance,
171 m->bits_ac_chrominance,
172 m->val_ac_chrominance);
176 * Writes the complete JPEG frame when optimal huffman tables are enabled,
177 * otherwise writes the stuffing.
179 * Header + values + stuffing.
181 * @param s The MpegEncContext.
182 * @return int Error code, 0 if successful.
184 int ff_mjpeg_encode_stuffing(MpegEncContext *s)
186 PutBitContext *pbc = &s->pb;
187 int mb_y = s->mb_y - !s->mb_x;
193 if (s->huffman == HUFFMAN_TABLE_OPTIMAL) {
194 mjpeg_build_optimal_huffman(m);
196 // Replace the VLCs with the optimal ones.
197 // The default ones may be used for trellis during quantization.
198 init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len);
199 init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len);
200 s->intra_ac_vlc_length =
201 s->intra_ac_vlc_last_length = m->uni_ac_vlc_len;
202 s->intra_chroma_ac_vlc_length =
203 s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len;
205 ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
206 s->pred, s->intra_matrix, s->chroma_intra_matrix);
207 mjpeg_encode_picture_frame(s);
210 ret = ff_mpv_reallocate_putbitbuffer(s, put_bits_count(&s->pb) / 8 + 100,
211 put_bits_count(&s->pb) / 4 + 1000);
213 av_log(s->avctx, AV_LOG_ERROR, "Buffer reallocation failed\n");
217 ff_mjpeg_escape_FF(pbc, s->esc_pos);
219 if ((s->avctx->active_thread_type & FF_THREAD_SLICE) && mb_y < s->mb_height - 1)
220 put_marker(pbc, RST0 + (mb_y&7));
221 s->esc_pos = put_bytes_count(pbc, 0);
224 for (int i = 0; i < 3; i++)
225 s->last_dc[i] = 128 << s->intra_dc_precision;
230 static int alloc_huffman(MpegEncContext *s)
232 MJpegContext *m = s->mjpeg_ctx;
233 size_t num_mbs, num_blocks, num_codes;
236 // We need to init this here as the mjpeg init is called before the common init,
237 s->mb_width = (s->width + 15) / 16;
238 s->mb_height = (s->height + 15) / 16;
240 switch (s->chroma_format) {
241 case CHROMA_420: blocks_per_mb = 6; break;
242 case CHROMA_422: blocks_per_mb = 8; break;
243 case CHROMA_444: blocks_per_mb = 12; break;
244 default: av_assert0(0);
247 // Make sure we have enough space to hold this frame.
248 num_mbs = s->mb_width * s->mb_height;
249 num_blocks = num_mbs * blocks_per_mb;
250 num_codes = num_blocks * 64;
252 m->huff_buffer = av_malloc_array(num_codes, sizeof(MJpegHuffmanCode));
254 return AVERROR(ENOMEM);
258 av_cold int ff_mjpeg_encode_init(MpegEncContext *s)
262 av_assert0(s->slice_context_count == 1);
264 if (s->width > 65500 || s->height > 65500) {
265 av_log(s, AV_LOG_ERROR, "JPEG does not support resolutions above 65500x65500\n");
266 return AVERROR(EINVAL);
269 m = av_mallocz(sizeof(MJpegContext));
271 return AVERROR(ENOMEM);
276 // Build default Huffman tables.
277 // These may be overwritten later with more optimal Huffman tables, but
278 // they are needed at least right now for some processes like trellis.
279 ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance,
280 m->huff_code_dc_luminance,
281 avpriv_mjpeg_bits_dc_luminance,
282 avpriv_mjpeg_val_dc);
283 ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance,
284 m->huff_code_dc_chrominance,
285 avpriv_mjpeg_bits_dc_chrominance,
286 avpriv_mjpeg_val_dc);
287 ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance,
288 m->huff_code_ac_luminance,
289 avpriv_mjpeg_bits_ac_luminance,
290 avpriv_mjpeg_val_ac_luminance);
291 ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance,
292 m->huff_code_ac_chrominance,
293 avpriv_mjpeg_bits_ac_chrominance,
294 avpriv_mjpeg_val_ac_chrominance);
296 init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len);
297 init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len);
298 s->intra_ac_vlc_length =
299 s->intra_ac_vlc_last_length = m->uni_ac_vlc_len;
300 s->intra_chroma_ac_vlc_length =
301 s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len;
303 // Buffers start out empty.
307 if(s->huffman == HUFFMAN_TABLE_OPTIMAL)
308 return alloc_huffman(s);
313 av_cold void ff_mjpeg_encode_close(MpegEncContext *s)
316 av_freep(&s->mjpeg_ctx->huff_buffer);
317 av_freep(&s->mjpeg_ctx);
322 * Add code and table_id to the JPEG buffer.
324 * @param s The MJpegContext which contains the JPEG buffer.
325 * @param table_id Which Huffman table the code belongs to.
326 * @param code The encoded exponent of the coefficients and the run-bits.
328 static inline void ff_mjpeg_encode_code(MJpegContext *s, uint8_t table_id, int code)
330 MJpegHuffmanCode *c = &s->huff_buffer[s->huff_ncode++];
331 c->table_id = table_id;
336 * Add the coefficient's data to the JPEG buffer.
338 * @param s The MJpegContext which contains the JPEG buffer.
339 * @param table_id Which Huffman table the code belongs to.
340 * @param val The coefficient.
341 * @param run The run-bits.
343 static void ff_mjpeg_encode_coef(MJpegContext *s, uint8_t table_id, int val, int run)
348 av_assert0(run == 0);
349 ff_mjpeg_encode_code(s, table_id, 0);
357 code = (run << 4) | (av_log2_16bit(val) + 1);
359 s->huff_buffer[s->huff_ncode].mant = mant;
360 ff_mjpeg_encode_code(s, table_id, code);
365 * Add the block's data into the JPEG buffer.
367 * @param s The MJpegEncContext that contains the JPEG buffer.
368 * @param block The block.
369 * @param n The block's index or number.
371 static void record_block(MpegEncContext *s, int16_t *block, int n)
374 int component, dc, last_index, val, run;
375 MJpegContext *m = s->mjpeg_ctx;
378 component = (n <= 3 ? 0 : (n&1) + 1);
379 table_id = (n <= 3 ? 0 : 1);
380 dc = block[0]; /* overflow is impossible */
381 val = dc - s->last_dc[component];
383 ff_mjpeg_encode_coef(m, table_id, val, 0);
385 s->last_dc[component] = dc;
390 last_index = s->block_last_index[n];
393 for(i=1;i<=last_index;i++) {
394 j = s->intra_scantable.permutated[i];
401 ff_mjpeg_encode_code(m, table_id, 0xf0);
404 ff_mjpeg_encode_coef(m, table_id, val, run);
409 /* output EOB only if not already 64 values */
410 if (last_index < 63 || run != 0)
411 ff_mjpeg_encode_code(m, table_id, 0);
414 static void encode_block(MpegEncContext *s, int16_t *block, int n)
416 int mant, nbits, code, i, j;
417 int component, dc, run, last_index, val;
418 MJpegContext *m = s->mjpeg_ctx;
419 uint8_t *huff_size_ac;
420 uint16_t *huff_code_ac;
423 component = (n <= 3 ? 0 : (n&1) + 1);
424 dc = block[0]; /* overflow is impossible */
425 val = dc - s->last_dc[component];
427 ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_luminance, m->huff_code_dc_luminance);
428 huff_size_ac = m->huff_size_ac_luminance;
429 huff_code_ac = m->huff_code_ac_luminance;
431 ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_chrominance, m->huff_code_dc_chrominance);
432 huff_size_ac = m->huff_size_ac_chrominance;
433 huff_code_ac = m->huff_code_ac_chrominance;
435 s->last_dc[component] = dc;
440 last_index = s->block_last_index[n];
441 for(i=1;i<=last_index;i++) {
442 j = s->intra_scantable.permutated[i];
448 put_bits(&s->pb, huff_size_ac[0xf0], huff_code_ac[0xf0]);
457 nbits= av_log2_16bit(val) + 1;
458 code = (run << 4) | nbits;
460 put_bits(&s->pb, huff_size_ac[code], huff_code_ac[code]);
462 put_sbits(&s->pb, nbits, mant);
467 /* output EOB only if not already 64 values */
468 if (last_index < 63 || run != 0)
469 put_bits(&s->pb, huff_size_ac[0], huff_code_ac[0]);
472 void ff_mjpeg_encode_mb(MpegEncContext *s, int16_t block[12][64])
475 if (s->huffman == HUFFMAN_TABLE_OPTIMAL) {
476 if (s->chroma_format == CHROMA_444) {
477 record_block(s, block[0], 0);
478 record_block(s, block[2], 2);
479 record_block(s, block[4], 4);
480 record_block(s, block[8], 8);
481 record_block(s, block[5], 5);
482 record_block(s, block[9], 9);
484 if (16*s->mb_x+8 < s->width) {
485 record_block(s, block[1], 1);
486 record_block(s, block[3], 3);
487 record_block(s, block[6], 6);
488 record_block(s, block[10], 10);
489 record_block(s, block[7], 7);
490 record_block(s, block[11], 11);
494 record_block(s, block[i], i);
496 if (s->chroma_format == CHROMA_420) {
497 record_block(s, block[5], 5);
499 record_block(s, block[6], 6);
500 record_block(s, block[5], 5);
501 record_block(s, block[7], 7);
505 if (s->chroma_format == CHROMA_444) {
506 encode_block(s, block[0], 0);
507 encode_block(s, block[2], 2);
508 encode_block(s, block[4], 4);
509 encode_block(s, block[8], 8);
510 encode_block(s, block[5], 5);
511 encode_block(s, block[9], 9);
513 if (16*s->mb_x+8 < s->width) {
514 encode_block(s, block[1], 1);
515 encode_block(s, block[3], 3);
516 encode_block(s, block[6], 6);
517 encode_block(s, block[10], 10);
518 encode_block(s, block[7], 7);
519 encode_block(s, block[11], 11);
523 encode_block(s, block[i], i);
525 if (s->chroma_format == CHROMA_420) {
526 encode_block(s, block[5], 5);
528 encode_block(s, block[6], 6);
529 encode_block(s, block[5], 5);
530 encode_block(s, block[7], 7);
534 s->i_tex_bits += get_bits_diff(s);
538 #if CONFIG_AMV_ENCODER
539 // maximum over s->mjpeg_vsample[i]
541 static int amv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
542 const AVFrame *pic_arg, int *got_packet)
544 MpegEncContext *s = avctx->priv_data;
547 int chroma_h_shift, chroma_v_shift;
549 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
551 if ((avctx->height & 15) && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) {
552 av_log(avctx, AV_LOG_ERROR,
553 "Heights which are not a multiple of 16 might fail with some decoders, "
554 "use vstrict=-1 / -strict -1 to use %d anyway.\n", avctx->height);
555 av_log(avctx, AV_LOG_WARNING, "If you have a device that plays AMV videos, please test if videos "
556 "with such heights work with it and report your findings to ffmpeg-devel@ffmpeg.org\n");
557 return AVERROR_EXPERIMENTAL;
560 pic = av_frame_clone(pic_arg);
562 return AVERROR(ENOMEM);
563 //picture should be flipped upside-down
564 for(i=0; i < 3; i++) {
565 int vsample = i ? 2 >> chroma_v_shift : 2;
566 pic->data[i] += pic->linesize[i] * (vsample * s->height / V_MAX - 1);
567 pic->linesize[i] *= -1;
569 ret = ff_mpv_encode_picture(avctx, pkt, pic, got_packet);
575 #define OFFSET(x) offsetof(MpegEncContext, x)
576 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
577 static const AVOption options[] = {
579 { "pred", "Prediction method", OFFSET(pred), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 3, VE, "pred" },
580 { "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, INT_MIN, INT_MAX, VE, "pred" },
581 { "plane", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, INT_MIN, INT_MAX, VE, "pred" },
582 { "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 }, INT_MIN, INT_MAX, VE, "pred" },
583 { "huffman", "Huffman table strategy", OFFSET(huffman), AV_OPT_TYPE_INT, { .i64 = HUFFMAN_TABLE_OPTIMAL }, 0, NB_HUFFMAN_TABLE_OPTION - 1, VE, "huffman" },
584 { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_DEFAULT }, INT_MIN, INT_MAX, VE, "huffman" },
585 { "optimal", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_OPTIMAL }, INT_MIN, INT_MAX, VE, "huffman" },
589 #if CONFIG_MJPEG_ENCODER
590 static const AVClass mjpeg_class = {
591 .class_name = "mjpeg encoder",
592 .item_name = av_default_item_name,
594 .version = LIBAVUTIL_VERSION_INT,
597 AVCodec ff_mjpeg_encoder = {
599 .long_name = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"),
600 .type = AVMEDIA_TYPE_VIDEO,
601 .id = AV_CODEC_ID_MJPEG,
602 .priv_data_size = sizeof(MpegEncContext),
603 .init = ff_mpv_encode_init,
604 .encode2 = ff_mpv_encode_picture,
605 .close = ff_mpv_encode_end,
606 .capabilities = AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS,
607 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
608 .pix_fmts = (const enum AVPixelFormat[]) {
609 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_NONE
611 .priv_class = &mjpeg_class,
612 .profiles = NULL_IF_CONFIG_SMALL(ff_mjpeg_profiles),
616 #if CONFIG_AMV_ENCODER
617 static const AVClass amv_class = {
618 .class_name = "amv encoder",
619 .item_name = av_default_item_name,
621 .version = LIBAVUTIL_VERSION_INT,
624 AVCodec ff_amv_encoder = {
626 .long_name = NULL_IF_CONFIG_SMALL("AMV Video"),
627 .type = AVMEDIA_TYPE_VIDEO,
628 .id = AV_CODEC_ID_AMV,
629 .priv_data_size = sizeof(MpegEncContext),
630 .init = ff_mpv_encode_init,
631 .encode2 = amv_encode_picture,
632 .close = ff_mpv_encode_end,
633 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
634 .pix_fmts = (const enum AVPixelFormat[]) {
635 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_NONE
637 .priv_class = &amv_class,