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 FFmpeg.
9 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst,
36 const uint8_t *src, int w, int left)
40 for (i = 0; i < w; i++) {
41 const int temp = src[i];
47 for (i = 0; i < 16; i++) {
48 const int temp = src[i];
52 s->dsp.diff_bytes(dst + 16, src + 16, src + 15, w - 16);
57 static inline void sub_left_prediction_bgr32(HYuvContext *s, uint8_t *dst,
58 const uint8_t *src, int w,
59 int *red, int *green, int *blue, int *alpha)
67 for (i = 0; i < FFMIN(w, 4); i++) {
68 const int rt = src[i * 4 + R];
69 const int gt = src[i * 4 + G];
70 const int bt = src[i * 4 + B];
71 const int at = src[i * 4 + A];
72 dst[i * 4 + R] = rt - r;
73 dst[i * 4 + G] = gt - g;
74 dst[i * 4 + B] = bt - b;
75 dst[i * 4 + A] = at - a;
82 s->dsp.diff_bytes(dst + 16, src + 16, src + 12, w * 4 - 16);
84 *red = src[(w - 1) * 4 + R];
85 *green = src[(w - 1) * 4 + G];
86 *blue = src[(w - 1) * 4 + B];
87 *alpha = src[(w - 1) * 4 + A];
90 static inline void sub_left_prediction_rgb24(HYuvContext *s, uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue){
96 for (i = 0; i < FFMIN(w,16); i++) {
97 const int rt = src[i*3 + 0];
98 const int gt = src[i*3 + 1];
99 const int bt = src[i*3 + 2];
100 dst[i*3 + 0] = rt - r;
101 dst[i*3 + 1] = gt - g;
102 dst[i*3 + 2] = bt - b;
108 s->dsp.diff_bytes(dst + 48, src + 48, src + 48 - 3, w*3 - 48);
110 *red = src[(w - 1)*3 + 0];
111 *green = src[(w - 1)*3 + 1];
112 *blue = src[(w - 1)*3 + 2];
115 static int store_table(HYuvContext *s, const uint8_t *len, uint8_t *buf)
120 for (i = 0; i < 256;) {
124 for (; i < 256 && len[i] == val && repeat < 255; i++)
127 av_assert0(val < 32 && val >0 && repeat<256 && repeat>0);
130 buf[index++] = repeat;
132 buf[index++] = val | (repeat << 5);
139 static av_cold int encode_init(AVCodecContext *avctx)
141 HYuvContext *s = avctx->priv_data;
144 ff_huffyuv_common_init(avctx);
146 avctx->extradata = av_mallocz(1024*30); // 256*3+4 == 772
147 avctx->stats_out = av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
148 if (!avctx->extradata || !avctx->stats_out) {
149 av_freep(&avctx->stats_out);
150 return AVERROR(ENOMEM);
154 avctx->coded_frame = &s->picture;
156 switch (avctx->pix_fmt) {
157 case AV_PIX_FMT_YUV420P:
158 case AV_PIX_FMT_YUV422P:
160 av_log(avctx, AV_LOG_ERROR, "width must be even for this colorspace\n");
161 return AVERROR(EINVAL);
163 s->bitstream_bpp = avctx->pix_fmt == AV_PIX_FMT_YUV420P ? 12 : 16;
165 case AV_PIX_FMT_RGB32:
166 s->bitstream_bpp = 32;
168 case AV_PIX_FMT_RGB24:
169 s->bitstream_bpp = 24;
172 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
173 return AVERROR(EINVAL);
175 avctx->bits_per_coded_sample = s->bitstream_bpp;
176 s->decorrelate = s->bitstream_bpp >= 24;
177 s->predictor = avctx->prediction_method;
178 s->interlaced = avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
179 if (avctx->context_model == 1) {
180 s->context = avctx->context_model;
181 if (s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)) {
182 av_log(avctx, AV_LOG_ERROR,
183 "context=1 is not compatible with "
184 "2 pass huffyuv encoding\n");
185 return AVERROR(EINVAL);
189 if (avctx->codec->id == AV_CODEC_ID_HUFFYUV) {
190 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
191 av_log(avctx, AV_LOG_ERROR,
192 "Error: YV12 is not supported by huffyuv; use "
193 "vcodec=ffvhuff or format=422p\n");
194 return AVERROR(EINVAL);
196 if (avctx->context_model) {
197 av_log(avctx, AV_LOG_ERROR,
198 "Error: per-frame huffman tables are not supported "
199 "by huffyuv; use vcodec=ffvhuff\n");
200 return AVERROR(EINVAL);
202 if (s->interlaced != ( s->height > 288 ))
203 av_log(avctx, AV_LOG_INFO,
204 "using huffyuv 2.2.0 or newer interlacing flag\n");
207 if (s->bitstream_bpp >= 24 && s->predictor == MEDIAN) {
208 av_log(avctx, AV_LOG_ERROR,
209 "Error: RGB is incompatible with median predictor\n");
210 return AVERROR(EINVAL);
213 ((uint8_t*)avctx->extradata)[0] = s->predictor | (s->decorrelate << 6);
214 ((uint8_t*)avctx->extradata)[1] = s->bitstream_bpp;
215 ((uint8_t*)avctx->extradata)[2] = s->interlaced ? 0x10 : 0x20;
217 ((uint8_t*)avctx->extradata)[2] |= 0x40;
218 ((uint8_t*)avctx->extradata)[3] = 0;
219 s->avctx->extradata_size = 4;
221 if (avctx->stats_in) {
222 char *p = avctx->stats_in;
224 for (i = 0; i < 3; i++)
225 for (j = 0; j < 256; j++)
229 for (i = 0; i < 3; i++) {
232 for (j = 0; j < 256; j++) {
233 s->stats[i][j] += strtol(p, &next, 0);
234 if (next == p) return -1;
238 if (p[0] == 0 || p[1] == 0 || p[2] == 0) break;
241 for (i = 0; i < 3; i++)
242 for (j = 0; j < 256; j++) {
243 int d = FFMIN(j, 256 - j);
245 s->stats[i][j] = 100000000 / (d + 1);
249 for (i = 0; i < 3; i++) {
250 ff_huff_gen_len_table(s->len[i], s->stats[i]);
252 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0) {
256 s->avctx->extradata_size +=
257 store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
261 for (i = 0; i < 3; i++) {
262 int pels = s->width * s->height / (i ? 40 : 10);
263 for (j = 0; j < 256; j++) {
264 int d = FFMIN(j, 256 - j);
265 s->stats[i][j] = pels/(d + 1);
269 for (i = 0; i < 3; i++)
270 for (j = 0; j < 256; j++)
274 if (ff_huffyuv_alloc_temp(s)) {
275 ff_huffyuv_common_end(s);
276 return AVERROR(ENOMEM);
283 static int encode_422_bitstream(HYuvContext *s, int offset, int count)
286 const uint8_t *y = s->temp[0] + offset;
287 const uint8_t *u = s->temp[1] + offset / 2;
288 const uint8_t *v = s->temp[2] + offset / 2;
290 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 2 * 4 * count) {
291 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
297 int y1 = y[2 * i + 1];\
303 if (s->flags & CODEC_FLAG_PASS1) {
304 for(i = 0; i < count; i++) {
312 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
315 for (i = 0; i < count; i++) {
318 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
320 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
322 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
324 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
327 for(i = 0; i < count; i++) {
329 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
330 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
331 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
332 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
338 static int encode_gray_bitstream(HYuvContext *s, int count)
342 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 4 * count) {
343 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
348 int y0 = s->temp[0][2 * i];\
349 int y1 = s->temp[0][2 * i + 1];
354 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
355 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
359 if (s->flags & CODEC_FLAG_PASS1) {
360 for (i = 0; i < count; i++) {
365 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
369 for (i = 0; i < count; i++) {
375 for (i = 0; i < count; i++) {
383 static inline int encode_bgra_bitstream(HYuvContext *s, int count, int planes)
387 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 4*planes*count) {
388 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
393 int g = s->temp[0][planes==3 ? 3*i + 1 : 4*i + G];\
394 int b = (s->temp[0][planes==3 ? 3*i + 2 : 4*i + B] - g) & 0xff;\
395 int r = (s->temp[0][planes==3 ? 3*i + 0 : 4*i + R] - g) & 0xff;\
396 int a = s->temp[0][planes*i + A];
401 if(planes==4) s->stats[2][a]++;
403 put_bits(&s->pb, s->len[1][g], s->bits[1][g]);\
404 put_bits(&s->pb, s->len[0][b], s->bits[0][b]);\
405 put_bits(&s->pb, s->len[2][r], s->bits[2][r]);\
406 if(planes==4) put_bits(&s->pb, s->len[2][a], s->bits[2][a]);
408 if ((s->flags & CODEC_FLAG_PASS1) &&
409 (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
410 for (i = 0; i < count; i++) {
414 } else if (s->context || (s->flags & CODEC_FLAG_PASS1)) {
415 for (i = 0; i < count; i++) {
421 for (i = 0; i < count; i++) {
429 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
430 const AVFrame *pict, int *got_packet)
432 HYuvContext *s = avctx->priv_data;
433 const int width = s->width;
434 const int width2 = s->width>>1;
435 const int height = s->height;
436 const int fake_ystride = s->interlaced ? pict->linesize[0]*2 : pict->linesize[0];
437 const int fake_ustride = s->interlaced ? pict->linesize[1]*2 : pict->linesize[1];
438 const int fake_vstride = s->interlaced ? pict->linesize[2]*2 : pict->linesize[2];
439 AVFrame * const p = &s->picture;
440 int i, j, size = 0, ret;
442 if ((ret = ff_alloc_packet2(avctx, pkt, width * height * 3 * 4 + FF_MIN_BUFFER_SIZE)) < 0)
446 p->pict_type = AV_PICTURE_TYPE_I;
450 for (i = 0; i < 3; i++) {
451 ff_huff_gen_len_table(s->len[i], s->stats[i]);
452 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0)
454 size += store_table(s, s->len[i], &pkt->data[size]);
457 for (i = 0; i < 3; i++)
458 for (j = 0; j < 256; j++)
459 s->stats[i][j] >>= 1;
462 init_put_bits(&s->pb, pkt->data + size, pkt->size - size);
464 if (avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
465 avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
466 int lefty, leftu, leftv, y, cy;
468 put_bits(&s->pb, 8, leftv = p->data[2][0]);
469 put_bits(&s->pb, 8, lefty = p->data[0][1]);
470 put_bits(&s->pb, 8, leftu = p->data[1][0]);
471 put_bits(&s->pb, 8, p->data[0][0]);
473 lefty = sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
474 leftu = sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
475 leftv = sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
477 encode_422_bitstream(s, 2, width-2);
479 if (s->predictor==MEDIAN) {
480 int lefttopy, lefttopu, lefttopv;
483 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + p->linesize[0], width , lefty);
484 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + p->linesize[1], width2, leftu);
485 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + p->linesize[2], width2, leftv);
487 encode_422_bitstream(s, 0, width);
491 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + fake_ystride, 4, lefty);
492 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + fake_ustride, 2, leftu);
493 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + fake_vstride, 2, leftv);
495 encode_422_bitstream(s, 0, 4);
497 lefttopy = p->data[0][3];
498 lefttopu = p->data[1][1];
499 lefttopv = p->data[2][1];
500 s->dsp.sub_hfyu_median_prediction(s->temp[0], p->data[0]+4, p->data[0] + fake_ystride + 4, width - 4 , &lefty, &lefttopy);
501 s->dsp.sub_hfyu_median_prediction(s->temp[1], p->data[1]+2, p->data[1] + fake_ustride + 2, width2 - 2, &leftu, &lefttopu);
502 s->dsp.sub_hfyu_median_prediction(s->temp[2], p->data[2]+2, p->data[2] + fake_vstride + 2, width2 - 2, &leftv, &lefttopv);
503 encode_422_bitstream(s, 0, width - 4);
506 for (; y < height; y++,cy++) {
507 uint8_t *ydst, *udst, *vdst;
509 if (s->bitstream_bpp == 12) {
511 ydst = p->data[0] + p->linesize[0] * y;
512 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
513 encode_gray_bitstream(s, width);
516 if (y >= height) break;
518 ydst = p->data[0] + p->linesize[0] * y;
519 udst = p->data[1] + p->linesize[1] * cy;
520 vdst = p->data[2] + p->linesize[2] * cy;
522 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
523 s->dsp.sub_hfyu_median_prediction(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
524 s->dsp.sub_hfyu_median_prediction(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
526 encode_422_bitstream(s, 0, width);
529 for (cy = y = 1; y < height; y++, cy++) {
530 uint8_t *ydst, *udst, *vdst;
532 /* encode a luma only line & y++ */
533 if (s->bitstream_bpp == 12) {
534 ydst = p->data[0] + p->linesize[0] * y;
536 if (s->predictor == PLANE && s->interlaced < y) {
537 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
539 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
541 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
543 encode_gray_bitstream(s, width);
545 if (y >= height) break;
548 ydst = p->data[0] + p->linesize[0] * y;
549 udst = p->data[1] + p->linesize[1] * cy;
550 vdst = p->data[2] + p->linesize[2] * cy;
552 if (s->predictor == PLANE && s->interlaced < cy) {
553 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
554 s->dsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
555 s->dsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
557 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
558 leftu = sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
559 leftv = sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
561 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
562 leftu = sub_left_prediction(s, s->temp[1], udst, width2, leftu);
563 leftv = sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
566 encode_422_bitstream(s, 0, width);
569 } else if(avctx->pix_fmt == AV_PIX_FMT_RGB32) {
570 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
571 const int stride = -p->linesize[0];
572 const int fake_stride = -fake_ystride;
574 int leftr, leftg, leftb, lefta;
576 put_bits(&s->pb, 8, lefta = data[A]);
577 put_bits(&s->pb, 8, leftr = data[R]);
578 put_bits(&s->pb, 8, leftg = data[G]);
579 put_bits(&s->pb, 8, leftb = data[B]);
581 sub_left_prediction_bgr32(s, s->temp[0], data + 4, width - 1, &leftr, &leftg, &leftb, &lefta);
582 encode_bgra_bitstream(s, width - 1, 4);
584 for (y = 1; y < s->height; y++) {
585 uint8_t *dst = data + y*stride;
586 if (s->predictor == PLANE && s->interlaced < y) {
587 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width * 4);
588 sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width, &leftr, &leftg, &leftb, &lefta);
590 sub_left_prediction_bgr32(s, s->temp[0], dst, width, &leftr, &leftg, &leftb, &lefta);
592 encode_bgra_bitstream(s, width, 4);
594 }else if(avctx->pix_fmt == AV_PIX_FMT_RGB24){
595 uint8_t *data = p->data[0] + (height-1)*p->linesize[0];
596 const int stride = -p->linesize[0];
597 const int fake_stride = -fake_ystride;
599 int leftr, leftg, leftb;
601 put_bits(&s->pb, 8, leftr= data[0]);
602 put_bits(&s->pb, 8, leftg= data[1]);
603 put_bits(&s->pb, 8, leftb= data[2]);
604 put_bits(&s->pb, 8, 0);
606 sub_left_prediction_rgb24(s, s->temp[0], data+3, width-1, &leftr, &leftg, &leftb);
607 encode_bgra_bitstream(s, width-1, 3);
609 for(y=1; y<s->height; y++){
610 uint8_t *dst = data + y*stride;
611 if(s->predictor == PLANE && s->interlaced < y){
612 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width*3);
613 sub_left_prediction_rgb24(s, s->temp[0], s->temp[1], width, &leftr, &leftg, &leftb);
615 sub_left_prediction_rgb24(s, s->temp[0], dst, width, &leftr, &leftg, &leftb);
617 encode_bgra_bitstream(s, width, 3);
620 av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
624 size += (put_bits_count(&s->pb) + 31) / 8;
625 put_bits(&s->pb, 16, 0);
626 put_bits(&s->pb, 15, 0);
629 if ((s->flags&CODEC_FLAG_PASS1) && (s->picture_number & 31) == 0) {
631 char *p = avctx->stats_out;
632 char *end = p + 1024*30;
633 for (i = 0; i < 3; i++) {
634 for (j = 0; j < 256; j++) {
635 snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
639 snprintf(p, end-p, "\n");
643 avctx->stats_out[0] = '\0';
644 if (!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
645 flush_put_bits(&s->pb);
646 s->dsp.bswap_buf((uint32_t*)pkt->data, (uint32_t*)pkt->data, size);
651 pkt->size = size * 4;
652 pkt->flags |= AV_PKT_FLAG_KEY;
658 static av_cold int encode_end(AVCodecContext *avctx)
660 HYuvContext *s = avctx->priv_data;
662 ff_huffyuv_common_end(s);
664 av_freep(&avctx->extradata);
665 av_freep(&avctx->stats_out);
670 #if CONFIG_HUFFYUV_ENCODER
671 AVCodec ff_huffyuv_encoder = {
673 .type = AVMEDIA_TYPE_VIDEO,
674 .id = AV_CODEC_ID_HUFFYUV,
675 .priv_data_size = sizeof(HYuvContext),
677 .encode2 = encode_frame,
679 .pix_fmts = (const enum AVPixelFormat[]){
680 AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24, AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
682 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
686 #if CONFIG_FFVHUFF_ENCODER
687 AVCodec ff_ffvhuff_encoder = {
689 .type = AVMEDIA_TYPE_VIDEO,
690 .id = AV_CODEC_ID_FFVHUFF,
691 .priv_data_size = sizeof(HYuvContext),
693 .encode2 = encode_frame,
695 .pix_fmts = (const enum AVPixelFormat[]){
696 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24, AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
698 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),