2 * Copyright (c) 2002-2003 Michael Niedermayer <michaelni@gmx.at>
4 * see http://www.pcisys.net/~melanson/codecs/huffyuv.txt for a description of
7 * This file is part of Libav.
9 * Libav is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * Libav is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
34 static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst,
35 uint8_t *src, int w, int left)
39 for (i = 0; i < w; i++) {
40 const int temp = src[i];
46 for (i = 0; i < 16; i++) {
47 const int temp = src[i];
51 s->dsp.diff_bytes(dst + 16, src + 16, src + 15, w - 16);
56 static inline void sub_left_prediction_bgr32(HYuvContext *s, uint8_t *dst,
58 int *red, int *green, int *blue,
68 for (i = 0; i < FFMIN(w, 4); i++) {
69 const int rt = src[i * 4 + R];
70 const int gt = src[i * 4 + G];
71 const int bt = src[i * 4 + B];
72 const int at = src[i * 4 + A];
73 dst[i * 4 + R] = rt - r;
74 dst[i * 4 + G] = gt - g;
75 dst[i * 4 + B] = bt - b;
76 dst[i * 4 + A] = at - a;
83 s->dsp.diff_bytes(dst + 16, src + 16, src + 12, w * 4 - 16);
85 *red = src[(w - 1) * 4 + R];
86 *green = src[(w - 1) * 4 + G];
87 *blue = src[(w - 1) * 4 + B];
88 *alpha = src[(w - 1) * 4 + A];
91 static inline void sub_left_prediction_rgb24(HYuvContext *s, uint8_t *dst,
93 int *red, int *green, int *blue)
100 for (i = 0; i < FFMIN(w, 16); i++) {
101 const int rt = src[i * 3 + 0];
102 const int gt = src[i * 3 + 1];
103 const int bt = src[i * 3 + 2];
104 dst[i * 3 + 0] = rt - r;
105 dst[i * 3 + 1] = gt - g;
106 dst[i * 3 + 2] = bt - b;
112 s->dsp.diff_bytes(dst + 48, src + 48, src + 48 - 3, w * 3 - 48);
114 *red = src[(w - 1) * 3 + 0];
115 *green = src[(w - 1) * 3 + 1];
116 *blue = src[(w - 1) * 3 + 2];
119 static int store_table(HYuvContext *s, const uint8_t *len, uint8_t *buf)
124 for (i = 0; i < 256;) {
128 for (; i < 256 && len[i] == val && repeat < 255; i++)
131 assert(val < 32 && val >0 && repeat<256 && repeat>0);
134 buf[index++] = repeat;
136 buf[index++] = val | (repeat << 5);
143 static av_cold int encode_init(AVCodecContext *avctx)
145 HYuvContext *s = avctx->priv_data;
148 ff_huffyuv_common_init(avctx);
150 avctx->extradata = av_mallocz(1024*30); // 256*3+4 == 772
151 avctx->stats_out = av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
154 avctx->coded_frame = av_frame_alloc();
155 if (!avctx->coded_frame)
156 return AVERROR(ENOMEM);
158 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
159 avctx->coded_frame->key_frame = 1;
161 switch (avctx->pix_fmt) {
162 case AV_PIX_FMT_YUV420P:
163 case AV_PIX_FMT_YUV422P:
165 av_log(avctx, AV_LOG_ERROR, "Width must be even for this colorspace.\n");
168 s->bitstream_bpp = avctx->pix_fmt == AV_PIX_FMT_YUV420P ? 12 : 16;
170 case AV_PIX_FMT_RGB32:
171 s->bitstream_bpp = 32;
173 case AV_PIX_FMT_RGB24:
174 s->bitstream_bpp = 24;
177 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
180 avctx->bits_per_coded_sample = s->bitstream_bpp;
181 s->decorrelate = s->bitstream_bpp >= 24;
182 s->predictor = avctx->prediction_method;
183 s->interlaced = avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
184 if (avctx->context_model == 1) {
185 s->context = avctx->context_model;
186 if (s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)) {
187 av_log(avctx, AV_LOG_ERROR,
188 "context=1 is not compatible with "
189 "2 pass huffyuv encoding\n");
194 if (avctx->codec->id == AV_CODEC_ID_HUFFYUV) {
195 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
196 av_log(avctx, AV_LOG_ERROR,
197 "Error: YV12 is not supported by huffyuv; use "
198 "vcodec=ffvhuff or format=422p\n");
201 if (avctx->context_model) {
202 av_log(avctx, AV_LOG_ERROR,
203 "Error: per-frame huffman tables are not supported "
204 "by huffyuv; use vcodec=ffvhuff\n");
207 if (s->interlaced != ( s->height > 288 ))
208 av_log(avctx, AV_LOG_INFO,
209 "using huffyuv 2.2.0 or newer interlacing flag\n");
212 if (s->bitstream_bpp >= 24 && s->predictor == MEDIAN) {
213 av_log(avctx, AV_LOG_ERROR,
214 "Error: RGB is incompatible with median predictor\n");
218 ((uint8_t*)avctx->extradata)[0] = s->predictor | (s->decorrelate << 6);
219 ((uint8_t*)avctx->extradata)[1] = s->bitstream_bpp;
220 ((uint8_t*)avctx->extradata)[2] = s->interlaced ? 0x10 : 0x20;
222 ((uint8_t*)avctx->extradata)[2] |= 0x40;
223 ((uint8_t*)avctx->extradata)[3] = 0;
224 s->avctx->extradata_size = 4;
226 if (avctx->stats_in) {
227 char *p = avctx->stats_in;
229 for (i = 0; i < 3; i++)
230 for (j = 0; j < 256; j++)
234 for (i = 0; i < 3; i++) {
237 for (j = 0; j < 256; j++) {
238 s->stats[i][j] += strtol(p, &next, 0);
239 if (next == p) return -1;
243 if (p[0] == 0 || p[1] == 0 || p[2] == 0) break;
246 for (i = 0; i < 3; i++)
247 for (j = 0; j < 256; j++) {
248 int d = FFMIN(j, 256 - j);
250 s->stats[i][j] = 100000000 / (d + 1);
254 for (i = 0; i < 3; i++) {
255 ff_huff_gen_len_table(s->len[i], s->stats[i]);
257 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0) {
261 s->avctx->extradata_size +=
262 store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
266 for (i = 0; i < 3; i++) {
267 int pels = s->width * s->height / (i ? 40 : 10);
268 for (j = 0; j < 256; j++) {
269 int d = FFMIN(j, 256 - j);
270 s->stats[i][j] = pels/(d + 1);
274 for (i = 0; i < 3; i++)
275 for (j = 0; j < 256; j++)
279 ff_huffyuv_alloc_temp(s);
285 static int encode_422_bitstream(HYuvContext *s, int offset, int count)
288 const uint8_t *y = s->temp[0] + offset;
289 const uint8_t *u = s->temp[1] + offset / 2;
290 const uint8_t *v = s->temp[2] + offset / 2;
292 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 2 * 4 * count) {
293 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
299 int y1 = y[2 * i + 1];\
305 if (s->flags & CODEC_FLAG_PASS1) {
306 for(i = 0; i < count; i++) {
314 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
317 for (i = 0; i < count; i++) {
320 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
322 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
324 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
326 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
329 for(i = 0; i < count; i++) {
331 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
332 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
333 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
334 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
340 static int encode_gray_bitstream(HYuvContext *s, int count)
344 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 4 * count) {
345 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
350 int y0 = s->temp[0][2 * i];\
351 int y1 = s->temp[0][2 * i + 1];
356 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
357 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
361 if (s->flags & CODEC_FLAG_PASS1) {
362 for (i = 0; i < count; i++) {
367 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
371 for (i = 0; i < count; i++) {
377 for (i = 0; i < count; i++) {
385 static inline int encode_bgra_bitstream(HYuvContext *s, int count, int planes)
389 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
390 4 * planes * count) {
391 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
396 int g = s->temp[0][planes == 3 ? 3 * i + 1 : 4 * i + G]; \
397 int b = s->temp[0][planes == 3 ? 3 * i + 2 : 4 * i + B] - g & 0xFF; \
398 int r = s->temp[0][planes == 3 ? 3 * i + 0 : 4 * i + R] - g & 0xFF; \
399 int a = s->temp[0][planes * i + A];
409 put_bits(&s->pb, s->len[1][g], s->bits[1][g]); \
410 put_bits(&s->pb, s->len[0][b], s->bits[0][b]); \
411 put_bits(&s->pb, s->len[2][r], s->bits[2][r]); \
413 put_bits(&s->pb, s->len[2][a], s->bits[2][a]);
415 if ((s->flags & CODEC_FLAG_PASS1) &&
416 (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
417 for (i = 0; i < count; i++) {
421 } else if (s->context || (s->flags & CODEC_FLAG_PASS1)) {
422 for (i = 0; i < count; i++) {
428 for (i = 0; i < count; i++) {
436 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
437 const AVFrame *pict, int *got_packet)
439 HYuvContext *s = avctx->priv_data;
440 const int width = s->width;
441 const int width2 = s->width>>1;
442 const int height = s->height;
443 const int fake_ystride = s->interlaced ? pict->linesize[0]*2 : pict->linesize[0];
444 const int fake_ustride = s->interlaced ? pict->linesize[1]*2 : pict->linesize[1];
445 const int fake_vstride = s->interlaced ? pict->linesize[2]*2 : pict->linesize[2];
446 const AVFrame * const p = pict;
447 int i, j, size = 0, ret;
450 (ret = av_new_packet(pkt, width * height * 3 * 4 + FF_MIN_BUFFER_SIZE)) < 0) {
451 av_log(avctx, AV_LOG_ERROR, "Error allocating output packet.\n");
456 for (i = 0; i < 3; i++) {
457 ff_huff_gen_len_table(s->len[i], s->stats[i]);
458 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0)
460 size += store_table(s, s->len[i], &pkt->data[size]);
463 for (i = 0; i < 3; i++)
464 for (j = 0; j < 256; j++)
465 s->stats[i][j] >>= 1;
468 init_put_bits(&s->pb, pkt->data + size, pkt->size - size);
470 if (avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
471 avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
472 int lefty, leftu, leftv, y, cy;
474 put_bits(&s->pb, 8, leftv = p->data[2][0]);
475 put_bits(&s->pb, 8, lefty = p->data[0][1]);
476 put_bits(&s->pb, 8, leftu = p->data[1][0]);
477 put_bits(&s->pb, 8, p->data[0][0]);
479 lefty = sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
480 leftu = sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
481 leftv = sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
483 encode_422_bitstream(s, 2, width-2);
485 if (s->predictor==MEDIAN) {
486 int lefttopy, lefttopu, lefttopv;
489 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + p->linesize[0], width , lefty);
490 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + p->linesize[1], width2, leftu);
491 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + p->linesize[2], width2, leftv);
493 encode_422_bitstream(s, 0, width);
497 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + fake_ystride, 4, lefty);
498 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + fake_ustride, 2, leftu);
499 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + fake_vstride, 2, leftv);
501 encode_422_bitstream(s, 0, 4);
503 lefttopy = p->data[0][3];
504 lefttopu = p->data[1][1];
505 lefttopv = p->data[2][1];
506 s->dsp.sub_hfyu_median_prediction(s->temp[0], p->data[0]+4, p->data[0] + fake_ystride + 4, width - 4 , &lefty, &lefttopy);
507 s->dsp.sub_hfyu_median_prediction(s->temp[1], p->data[1]+2, p->data[1] + fake_ustride + 2, width2 - 2, &leftu, &lefttopu);
508 s->dsp.sub_hfyu_median_prediction(s->temp[2], p->data[2]+2, p->data[2] + fake_vstride + 2, width2 - 2, &leftv, &lefttopv);
509 encode_422_bitstream(s, 0, width - 4);
512 for (; y < height; y++,cy++) {
513 uint8_t *ydst, *udst, *vdst;
515 if (s->bitstream_bpp == 12) {
517 ydst = p->data[0] + p->linesize[0] * y;
518 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
519 encode_gray_bitstream(s, width);
522 if (y >= height) break;
524 ydst = p->data[0] + p->linesize[0] * y;
525 udst = p->data[1] + p->linesize[1] * cy;
526 vdst = p->data[2] + p->linesize[2] * cy;
528 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
529 s->dsp.sub_hfyu_median_prediction(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
530 s->dsp.sub_hfyu_median_prediction(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
532 encode_422_bitstream(s, 0, width);
535 for (cy = y = 1; y < height; y++, cy++) {
536 uint8_t *ydst, *udst, *vdst;
538 /* encode a luma only line & y++ */
539 if (s->bitstream_bpp == 12) {
540 ydst = p->data[0] + p->linesize[0] * y;
542 if (s->predictor == PLANE && s->interlaced < y) {
543 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
545 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
547 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
549 encode_gray_bitstream(s, width);
551 if (y >= height) break;
554 ydst = p->data[0] + p->linesize[0] * y;
555 udst = p->data[1] + p->linesize[1] * cy;
556 vdst = p->data[2] + p->linesize[2] * cy;
558 if (s->predictor == PLANE && s->interlaced < cy) {
559 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
560 s->dsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
561 s->dsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
563 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
564 leftu = sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
565 leftv = sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
567 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
568 leftu = sub_left_prediction(s, s->temp[1], udst, width2, leftu);
569 leftv = sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
572 encode_422_bitstream(s, 0, width);
575 } else if(avctx->pix_fmt == AV_PIX_FMT_RGB32) {
576 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
577 const int stride = -p->linesize[0];
578 const int fake_stride = -fake_ystride;
580 int leftr, leftg, leftb, lefta;
582 put_bits(&s->pb, 8, lefta = data[A]);
583 put_bits(&s->pb, 8, leftr = data[R]);
584 put_bits(&s->pb, 8, leftg = data[G]);
585 put_bits(&s->pb, 8, leftb = data[B]);
587 sub_left_prediction_bgr32(s, s->temp[0], data + 4, width - 1,
588 &leftr, &leftg, &leftb, &lefta);
589 encode_bgra_bitstream(s, width - 1, 4);
591 for (y = 1; y < s->height; y++) {
592 uint8_t *dst = data + y*stride;
593 if (s->predictor == PLANE && s->interlaced < y) {
594 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width * 4);
595 sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width,
596 &leftr, &leftg, &leftb, &lefta);
598 sub_left_prediction_bgr32(s, s->temp[0], dst, width,
599 &leftr, &leftg, &leftb, &lefta);
601 encode_bgra_bitstream(s, width, 4);
603 } else if (avctx->pix_fmt == AV_PIX_FMT_RGB24) {
604 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
605 const int stride = -p->linesize[0];
606 const int fake_stride = -fake_ystride;
608 int leftr, leftg, leftb;
610 put_bits(&s->pb, 8, leftr = data[0]);
611 put_bits(&s->pb, 8, leftg = data[1]);
612 put_bits(&s->pb, 8, leftb = data[2]);
613 put_bits(&s->pb, 8, 0);
615 sub_left_prediction_rgb24(s, s->temp[0], data + 3, width - 1,
616 &leftr, &leftg, &leftb);
617 encode_bgra_bitstream(s, width-1, 3);
619 for (y = 1; y < s->height; y++) {
620 uint8_t *dst = data + y * stride;
621 if (s->predictor == PLANE && s->interlaced < y) {
622 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride,
624 sub_left_prediction_rgb24(s, s->temp[0], s->temp[1], width,
625 &leftr, &leftg, &leftb);
627 sub_left_prediction_rgb24(s, s->temp[0], dst, width,
628 &leftr, &leftg, &leftb);
630 encode_bgra_bitstream(s, width, 3);
633 av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
637 size += (put_bits_count(&s->pb) + 31) / 8;
638 put_bits(&s->pb, 16, 0);
639 put_bits(&s->pb, 15, 0);
642 if ((s->flags&CODEC_FLAG_PASS1) && (s->picture_number & 31) == 0) {
644 char *p = avctx->stats_out;
645 char *end = p + 1024*30;
646 for (i = 0; i < 3; i++) {
647 for (j = 0; j < 256; j++) {
648 snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
652 snprintf(p, end-p, "\n");
656 avctx->stats_out[0] = '\0';
657 if (!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
658 flush_put_bits(&s->pb);
659 s->dsp.bswap_buf((uint32_t*)pkt->data, (uint32_t*)pkt->data, size);
664 pkt->size = size * 4;
665 pkt->flags |= AV_PKT_FLAG_KEY;
671 static av_cold int encode_end(AVCodecContext *avctx)
673 HYuvContext *s = avctx->priv_data;
675 ff_huffyuv_common_end(s);
677 av_freep(&avctx->extradata);
678 av_freep(&avctx->stats_out);
680 av_frame_free(&avctx->coded_frame);
685 #if CONFIG_HUFFYUV_ENCODER
686 AVCodec ff_huffyuv_encoder = {
688 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
689 .type = AVMEDIA_TYPE_VIDEO,
690 .id = AV_CODEC_ID_HUFFYUV,
691 .priv_data_size = sizeof(HYuvContext),
693 .encode2 = encode_frame,
695 .pix_fmts = (const enum AVPixelFormat[]){
696 AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
697 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
702 #if CONFIG_FFVHUFF_ENCODER
703 AVCodec ff_ffvhuff_encoder = {
705 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),
706 .type = AVMEDIA_TYPE_VIDEO,
707 .id = AV_CODEC_ID_FFVHUFF,
708 .priv_data_size = sizeof(HYuvContext),
710 .encode2 = encode_frame,
712 .pix_fmts = (const enum AVPixelFormat[]){
713 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
714 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE