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
68 #if CONFIG_MJPEG_ENCODER
70 * Encodes and outputs the entire frame in the JPEG format.
72 * @param s The MpegEncContext.
74 static void mjpeg_encode_picture_frame(MpegEncContext *s)
76 int nbits, code, table_id;
77 MJpegContext *m = s->mjpeg_ctx;
78 uint8_t *huff_size[4] = { m->huff_size_dc_luminance,
79 m->huff_size_dc_chrominance,
80 m->huff_size_ac_luminance,
81 m->huff_size_ac_chrominance };
82 uint16_t *huff_code[4] = { m->huff_code_dc_luminance,
83 m->huff_code_dc_chrominance,
84 m->huff_code_ac_luminance,
85 m->huff_code_ac_chrominance };
86 size_t total_bits = 0;
89 s->header_bits = get_bits_diff(s);
90 // Estimate the total size first
91 for (int i = 0; i < m->huff_ncode; i++) {
92 table_id = m->huff_buffer[i].table_id;
93 code = m->huff_buffer[i].code;
96 total_bits += huff_size[table_id][code] + nbits;
99 bytes_needed = (total_bits + 7) / 8;
100 ff_mpv_reallocate_putbitbuffer(s, bytes_needed, bytes_needed);
102 for (int i = 0; i < m->huff_ncode; i++) {
103 table_id = m->huff_buffer[i].table_id;
104 code = m->huff_buffer[i].code;
107 put_bits(&s->pb, huff_size[table_id][code], huff_code[table_id][code]);
109 put_sbits(&s->pb, nbits, m->huff_buffer[i].mant);
114 s->i_tex_bits = get_bits_diff(s);
118 * Builds all 4 optimal Huffman tables.
120 * Uses the data stored in the JPEG buffer to compute the tables.
121 * Stores the Huffman tables in the bits_* and val_* arrays in the MJpegContext.
123 * @param m MJpegContext containing the JPEG buffer.
125 static void mjpeg_build_optimal_huffman(MJpegContext *m)
127 MJpegEncHuffmanContext dc_luminance_ctx;
128 MJpegEncHuffmanContext dc_chrominance_ctx;
129 MJpegEncHuffmanContext ac_luminance_ctx;
130 MJpegEncHuffmanContext ac_chrominance_ctx;
131 MJpegEncHuffmanContext *ctx[4] = { &dc_luminance_ctx,
134 &ac_chrominance_ctx };
135 for (int i = 0; i < 4; i++)
136 ff_mjpeg_encode_huffman_init(ctx[i]);
138 for (int i = 0; i < m->huff_ncode; i++) {
139 int table_id = m->huff_buffer[i].table_id;
140 int code = m->huff_buffer[i].code;
142 ff_mjpeg_encode_huffman_increment(ctx[table_id], code);
145 ff_mjpeg_encode_huffman_close(&dc_luminance_ctx,
146 m->bits_dc_luminance,
147 m->val_dc_luminance, 12);
148 ff_mjpeg_encode_huffman_close(&dc_chrominance_ctx,
149 m->bits_dc_chrominance,
150 m->val_dc_chrominance, 12);
151 ff_mjpeg_encode_huffman_close(&ac_luminance_ctx,
152 m->bits_ac_luminance,
153 m->val_ac_luminance, 256);
154 ff_mjpeg_encode_huffman_close(&ac_chrominance_ctx,
155 m->bits_ac_chrominance,
156 m->val_ac_chrominance, 256);
158 ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance,
159 m->huff_code_dc_luminance,
160 m->bits_dc_luminance,
161 m->val_dc_luminance);
162 ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance,
163 m->huff_code_dc_chrominance,
164 m->bits_dc_chrominance,
165 m->val_dc_chrominance);
166 ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance,
167 m->huff_code_ac_luminance,
168 m->bits_ac_luminance,
169 m->val_ac_luminance);
170 ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance,
171 m->huff_code_ac_chrominance,
172 m->bits_ac_chrominance,
173 m->val_ac_chrominance);
178 * Writes the complete JPEG frame when optimal huffman tables are enabled,
179 * otherwise writes the stuffing.
181 * Header + values + stuffing.
183 * @param s The MpegEncContext.
184 * @return int Error code, 0 if successful.
186 int ff_mjpeg_encode_stuffing(MpegEncContext *s)
188 PutBitContext *pbc = &s->pb;
189 int mb_y = s->mb_y - !s->mb_x;
192 #if CONFIG_MJPEG_ENCODER
193 if (s->huffman == HUFFMAN_TABLE_OPTIMAL) {
194 MJpegContext *m = s->mjpeg_ctx;
196 mjpeg_build_optimal_huffman(m);
198 // Replace the VLCs with the optimal ones.
199 // The default ones may be used for trellis during quantization.
200 init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len);
201 init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len);
202 s->intra_ac_vlc_length =
203 s->intra_ac_vlc_last_length = m->uni_ac_vlc_len;
204 s->intra_chroma_ac_vlc_length =
205 s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len;
207 ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
208 s->pred, s->intra_matrix, s->chroma_intra_matrix);
209 mjpeg_encode_picture_frame(s);
213 ret = ff_mpv_reallocate_putbitbuffer(s, put_bits_count(&s->pb) / 8 + 100,
214 put_bits_count(&s->pb) / 4 + 1000);
216 av_log(s->avctx, AV_LOG_ERROR, "Buffer reallocation failed\n");
220 ff_mjpeg_escape_FF(pbc, s->esc_pos);
222 if ((s->avctx->active_thread_type & FF_THREAD_SLICE) && mb_y < s->mb_height - 1)
223 put_marker(pbc, RST0 + (mb_y&7));
224 s->esc_pos = put_bytes_count(pbc, 0);
227 for (int i = 0; i < 3; i++)
228 s->last_dc[i] = 128 << s->intra_dc_precision;
233 static int alloc_huffman(MpegEncContext *s)
235 MJpegContext *m = s->mjpeg_ctx;
236 size_t num_mbs, num_blocks, num_codes;
239 // We need to init this here as the mjpeg init is called before the common init,
240 s->mb_width = (s->width + 15) / 16;
241 s->mb_height = (s->height + 15) / 16;
243 switch (s->chroma_format) {
244 case CHROMA_420: blocks_per_mb = 6; break;
245 case CHROMA_422: blocks_per_mb = 8; break;
246 case CHROMA_444: blocks_per_mb = 12; break;
247 default: av_assert0(0);
250 // Make sure we have enough space to hold this frame.
251 num_mbs = s->mb_width * s->mb_height;
252 num_blocks = num_mbs * blocks_per_mb;
253 num_codes = num_blocks * 64;
255 m->huff_buffer = av_malloc_array(num_codes, sizeof(MJpegHuffmanCode));
257 return AVERROR(ENOMEM);
261 av_cold int ff_mjpeg_encode_init(MpegEncContext *s)
266 av_assert0(s->slice_context_count == 1);
268 /* The following check is automatically true for AMV,
269 * but it doesn't hurt either. */
270 ret = ff_mjpeg_encode_check_pix_fmt(s->avctx);
274 if (s->width > 65500 || s->height > 65500) {
275 av_log(s, AV_LOG_ERROR, "JPEG does not support resolutions above 65500x65500\n");
276 return AVERROR(EINVAL);
279 m = av_mallocz(sizeof(MJpegContext));
281 return AVERROR(ENOMEM);
286 // Build default Huffman tables.
287 // These may be overwritten later with more optimal Huffman tables, but
288 // they are needed at least right now for some processes like trellis.
289 ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance,
290 m->huff_code_dc_luminance,
291 avpriv_mjpeg_bits_dc_luminance,
292 avpriv_mjpeg_val_dc);
293 ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance,
294 m->huff_code_dc_chrominance,
295 avpriv_mjpeg_bits_dc_chrominance,
296 avpriv_mjpeg_val_dc);
297 ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance,
298 m->huff_code_ac_luminance,
299 avpriv_mjpeg_bits_ac_luminance,
300 avpriv_mjpeg_val_ac_luminance);
301 ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance,
302 m->huff_code_ac_chrominance,
303 avpriv_mjpeg_bits_ac_chrominance,
304 avpriv_mjpeg_val_ac_chrominance);
306 init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len);
307 init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len);
308 s->intra_ac_vlc_length =
309 s->intra_ac_vlc_last_length = m->uni_ac_vlc_len;
310 s->intra_chroma_ac_vlc_length =
311 s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len;
313 // Buffers start out empty.
317 if(s->huffman == HUFFMAN_TABLE_OPTIMAL)
318 return alloc_huffman(s);
323 av_cold void ff_mjpeg_encode_close(MpegEncContext *s)
326 av_freep(&s->mjpeg_ctx->huff_buffer);
327 av_freep(&s->mjpeg_ctx);
332 * Add code and table_id to the JPEG buffer.
334 * @param s The MJpegContext which contains the JPEG buffer.
335 * @param table_id Which Huffman table the code belongs to.
336 * @param code The encoded exponent of the coefficients and the run-bits.
338 static inline void ff_mjpeg_encode_code(MJpegContext *s, uint8_t table_id, int code)
340 MJpegHuffmanCode *c = &s->huff_buffer[s->huff_ncode++];
341 c->table_id = table_id;
346 * Add the coefficient's data to the JPEG buffer.
348 * @param s The MJpegContext which contains the JPEG buffer.
349 * @param table_id Which Huffman table the code belongs to.
350 * @param val The coefficient.
351 * @param run The run-bits.
353 static void ff_mjpeg_encode_coef(MJpegContext *s, uint8_t table_id, int val, int run)
358 av_assert0(run == 0);
359 ff_mjpeg_encode_code(s, table_id, 0);
367 code = (run << 4) | (av_log2_16bit(val) + 1);
369 s->huff_buffer[s->huff_ncode].mant = mant;
370 ff_mjpeg_encode_code(s, table_id, code);
375 * Add the block's data into the JPEG buffer.
377 * @param s The MJpegEncContext that contains the JPEG buffer.
378 * @param block The block.
379 * @param n The block's index or number.
381 static void record_block(MpegEncContext *s, int16_t *block, int n)
384 int component, dc, last_index, val, run;
385 MJpegContext *m = s->mjpeg_ctx;
388 component = (n <= 3 ? 0 : (n&1) + 1);
389 table_id = (n <= 3 ? 0 : 1);
390 dc = block[0]; /* overflow is impossible */
391 val = dc - s->last_dc[component];
393 ff_mjpeg_encode_coef(m, table_id, val, 0);
395 s->last_dc[component] = dc;
400 last_index = s->block_last_index[n];
403 for(i=1;i<=last_index;i++) {
404 j = s->intra_scantable.permutated[i];
411 ff_mjpeg_encode_code(m, table_id, 0xf0);
414 ff_mjpeg_encode_coef(m, table_id, val, run);
419 /* output EOB only if not already 64 values */
420 if (last_index < 63 || run != 0)
421 ff_mjpeg_encode_code(m, table_id, 0);
424 static void encode_block(MpegEncContext *s, int16_t *block, int n)
426 int mant, nbits, code, i, j;
427 int component, dc, run, last_index, val;
428 MJpegContext *m = s->mjpeg_ctx;
429 uint8_t *huff_size_ac;
430 uint16_t *huff_code_ac;
433 component = (n <= 3 ? 0 : (n&1) + 1);
434 dc = block[0]; /* overflow is impossible */
435 val = dc - s->last_dc[component];
437 ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_luminance, m->huff_code_dc_luminance);
438 huff_size_ac = m->huff_size_ac_luminance;
439 huff_code_ac = m->huff_code_ac_luminance;
441 ff_mjpeg_encode_dc(&s->pb, val, m->huff_size_dc_chrominance, m->huff_code_dc_chrominance);
442 huff_size_ac = m->huff_size_ac_chrominance;
443 huff_code_ac = m->huff_code_ac_chrominance;
445 s->last_dc[component] = dc;
450 last_index = s->block_last_index[n];
451 for(i=1;i<=last_index;i++) {
452 j = s->intra_scantable.permutated[i];
458 put_bits(&s->pb, huff_size_ac[0xf0], huff_code_ac[0xf0]);
467 nbits= av_log2_16bit(val) + 1;
468 code = (run << 4) | nbits;
470 put_bits(&s->pb, huff_size_ac[code], huff_code_ac[code]);
472 put_sbits(&s->pb, nbits, mant);
477 /* output EOB only if not already 64 values */
478 if (last_index < 63 || run != 0)
479 put_bits(&s->pb, huff_size_ac[0], huff_code_ac[0]);
482 void ff_mjpeg_encode_mb(MpegEncContext *s, int16_t block[12][64])
485 if (s->huffman == HUFFMAN_TABLE_OPTIMAL) {
486 if (s->chroma_format == CHROMA_444) {
487 record_block(s, block[0], 0);
488 record_block(s, block[2], 2);
489 record_block(s, block[4], 4);
490 record_block(s, block[8], 8);
491 record_block(s, block[5], 5);
492 record_block(s, block[9], 9);
494 if (16*s->mb_x+8 < s->width) {
495 record_block(s, block[1], 1);
496 record_block(s, block[3], 3);
497 record_block(s, block[6], 6);
498 record_block(s, block[10], 10);
499 record_block(s, block[7], 7);
500 record_block(s, block[11], 11);
504 record_block(s, block[i], i);
506 if (s->chroma_format == CHROMA_420) {
507 record_block(s, block[5], 5);
509 record_block(s, block[6], 6);
510 record_block(s, block[5], 5);
511 record_block(s, block[7], 7);
515 if (s->chroma_format == CHROMA_444) {
516 encode_block(s, block[0], 0);
517 encode_block(s, block[2], 2);
518 encode_block(s, block[4], 4);
519 encode_block(s, block[8], 8);
520 encode_block(s, block[5], 5);
521 encode_block(s, block[9], 9);
523 if (16*s->mb_x+8 < s->width) {
524 encode_block(s, block[1], 1);
525 encode_block(s, block[3], 3);
526 encode_block(s, block[6], 6);
527 encode_block(s, block[10], 10);
528 encode_block(s, block[7], 7);
529 encode_block(s, block[11], 11);
533 encode_block(s, block[i], i);
535 if (s->chroma_format == CHROMA_420) {
536 encode_block(s, block[5], 5);
538 encode_block(s, block[6], 6);
539 encode_block(s, block[5], 5);
540 encode_block(s, block[7], 7);
544 s->i_tex_bits += get_bits_diff(s);
548 #if CONFIG_AMV_ENCODER
549 // maximum over s->mjpeg_vsample[i]
551 static int amv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
552 const AVFrame *pic_arg, int *got_packet)
554 MpegEncContext *s = avctx->priv_data;
557 int chroma_h_shift, chroma_v_shift;
559 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
561 if ((avctx->height & 15) && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) {
562 av_log(avctx, AV_LOG_ERROR,
563 "Heights which are not a multiple of 16 might fail with some decoders, "
564 "use vstrict=-1 / -strict -1 to use %d anyway.\n", avctx->height);
565 av_log(avctx, AV_LOG_WARNING, "If you have a device that plays AMV videos, please test if videos "
566 "with such heights work with it and report your findings to ffmpeg-devel@ffmpeg.org\n");
567 return AVERROR_EXPERIMENTAL;
570 pic = av_frame_clone(pic_arg);
572 return AVERROR(ENOMEM);
573 //picture should be flipped upside-down
574 for(i=0; i < 3; i++) {
575 int vsample = i ? 2 >> chroma_v_shift : 2;
576 pic->data[i] += pic->linesize[i] * (vsample * s->height / V_MAX - 1);
577 pic->linesize[i] *= -1;
579 ret = ff_mpv_encode_picture(avctx, pkt, pic, got_packet);
585 #define OFFSET(x) offsetof(MpegEncContext, x)
586 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
587 static const AVOption options[] = {
589 { "pred", "Prediction method", OFFSET(pred), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 3, VE, "pred" },
590 { "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, INT_MIN, INT_MAX, VE, "pred" },
591 { "plane", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, INT_MIN, INT_MAX, VE, "pred" },
592 { "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 }, INT_MIN, INT_MAX, VE, "pred" },
593 { "huffman", "Huffman table strategy", OFFSET(huffman), AV_OPT_TYPE_INT, { .i64 = HUFFMAN_TABLE_OPTIMAL }, 0, NB_HUFFMAN_TABLE_OPTION - 1, VE, "huffman" },
594 { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_DEFAULT }, INT_MIN, INT_MAX, VE, "huffman" },
595 { "optimal", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_OPTIMAL }, INT_MIN, INT_MAX, VE, "huffman" },
596 { "force_duplicated_matrix", "Always write luma and chroma matrix for mjpeg, useful for rtp streaming.", OFFSET(force_duplicated_matrix), AV_OPT_TYPE_BOOL, {.i64 = 0 }, 0, 1, VE },
597 #if FF_API_MPEGVIDEO_OPTS
598 FF_MPV_DEPRECATED_MPEG_QUANT_OPT
599 FF_MPV_DEPRECATED_A53_CC_OPT
600 FF_MPV_DEPRECATED_BFRAME_OPTS
605 #if CONFIG_MJPEG_ENCODER
606 static const AVClass mjpeg_class = {
607 .class_name = "mjpeg encoder",
608 .item_name = av_default_item_name,
610 .version = LIBAVUTIL_VERSION_INT,
613 const AVCodec ff_mjpeg_encoder = {
615 .long_name = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"),
616 .type = AVMEDIA_TYPE_VIDEO,
617 .id = AV_CODEC_ID_MJPEG,
618 .priv_data_size = sizeof(MpegEncContext),
619 .init = ff_mpv_encode_init,
620 .encode2 = ff_mpv_encode_picture,
621 .close = ff_mpv_encode_end,
622 .capabilities = AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS,
623 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
624 .pix_fmts = (const enum AVPixelFormat[]) {
625 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P,
626 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
629 .priv_class = &mjpeg_class,
630 .profiles = NULL_IF_CONFIG_SMALL(ff_mjpeg_profiles),
634 #if CONFIG_AMV_ENCODER
635 static const AVClass amv_class = {
636 .class_name = "amv encoder",
637 .item_name = av_default_item_name,
639 .version = LIBAVUTIL_VERSION_INT,
642 const AVCodec ff_amv_encoder = {
644 .long_name = NULL_IF_CONFIG_SMALL("AMV Video"),
645 .type = AVMEDIA_TYPE_VIDEO,
646 .id = AV_CODEC_ID_AMV,
647 .priv_data_size = sizeof(MpegEncContext),
648 .init = ff_mpv_encode_init,
649 .encode2 = amv_encode_picture,
650 .close = ff_mpv_encode_end,
651 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
652 .pix_fmts = (const enum AVPixelFormat[]) {
653 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_NONE
655 .priv_class = &amv_class,