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,
69 for (i = 0; i < FFMIN(w, 4); i++) {
70 const int rt = src[i * 4 + R];
71 const int gt = src[i * 4 + G];
72 const int bt = src[i * 4 + B];
73 const int at = src[i * 4 + A];
74 dst[i * 4 + R] = rt - r;
75 dst[i * 4 + G] = gt - g;
76 dst[i * 4 + B] = bt - b;
77 dst[i * 4 + A] = at - a;
84 s->dsp.diff_bytes(dst + 16, src + 16, src + 12, w * 4 - 16);
86 *red = src[(w - 1) * 4 + R];
87 *green = src[(w - 1) * 4 + G];
88 *blue = src[(w - 1) * 4 + B];
89 *alpha = src[(w - 1) * 4 + A];
92 static inline void sub_left_prediction_rgb24(HYuvContext *s, uint8_t *dst,
94 int *red, int *green, int *blue)
101 for (i = 0; i < FFMIN(w, 16); i++) {
102 const int rt = src[i * 3 + 0];
103 const int gt = src[i * 3 + 1];
104 const int bt = src[i * 3 + 2];
105 dst[i * 3 + 0] = rt - r;
106 dst[i * 3 + 1] = gt - g;
107 dst[i * 3 + 2] = bt - b;
113 s->dsp.diff_bytes(dst + 48, src + 48, src + 48 - 3, w * 3 - 48);
115 *red = src[(w - 1) * 3 + 0];
116 *green = src[(w - 1) * 3 + 1];
117 *blue = src[(w - 1) * 3 + 2];
120 static int store_table(HYuvContext *s, const uint8_t *len, uint8_t *buf)
125 for (i = 0; i < 256;) {
129 for (; i < 256 && len[i] == val && repeat < 255; i++)
132 av_assert0(val < 32 && val >0 && repeat<256 && repeat>0);
135 buf[index++] = repeat;
137 buf[index++] = val | (repeat << 5);
144 static av_cold int encode_init(AVCodecContext *avctx)
146 HYuvContext *s = avctx->priv_data;
149 ff_huffyuv_common_init(avctx);
151 avctx->extradata = av_mallocz(1024*30); // 256*3+4 == 772
152 avctx->stats_out = av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
153 if (!avctx->extradata || !avctx->stats_out) {
154 av_freep(&avctx->stats_out);
155 return AVERROR(ENOMEM);
159 avctx->coded_frame = av_frame_alloc();
160 if (!avctx->coded_frame)
161 return AVERROR(ENOMEM);
163 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
164 avctx->coded_frame->key_frame = 1;
166 switch (avctx->pix_fmt) {
167 case AV_PIX_FMT_YUV420P:
168 case AV_PIX_FMT_YUV422P:
170 av_log(avctx, AV_LOG_ERROR, "Width must be even for this colorspace.\n");
171 return AVERROR(EINVAL);
173 s->bitstream_bpp = avctx->pix_fmt == AV_PIX_FMT_YUV420P ? 12 : 16;
175 case AV_PIX_FMT_RGB32:
176 s->bitstream_bpp = 32;
178 case AV_PIX_FMT_RGB24:
179 s->bitstream_bpp = 24;
182 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
183 return AVERROR(EINVAL);
185 avctx->bits_per_coded_sample = s->bitstream_bpp;
186 s->decorrelate = s->bitstream_bpp >= 24;
187 s->predictor = avctx->prediction_method;
188 s->interlaced = avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
189 if (avctx->context_model == 1) {
190 s->context = avctx->context_model;
191 if (s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)) {
192 av_log(avctx, AV_LOG_ERROR,
193 "context=1 is not compatible with "
194 "2 pass huffyuv encoding\n");
195 return AVERROR(EINVAL);
199 if (avctx->codec->id == AV_CODEC_ID_HUFFYUV) {
200 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
201 av_log(avctx, AV_LOG_ERROR,
202 "Error: YV12 is not supported by huffyuv; use "
203 "vcodec=ffvhuff or format=422p\n");
204 return AVERROR(EINVAL);
206 if (avctx->context_model) {
207 av_log(avctx, AV_LOG_ERROR,
208 "Error: per-frame huffman tables are not supported "
209 "by huffyuv; use vcodec=ffvhuff\n");
210 return AVERROR(EINVAL);
212 if (s->interlaced != ( s->height > 288 ))
213 av_log(avctx, AV_LOG_INFO,
214 "using huffyuv 2.2.0 or newer interlacing flag\n");
217 if (s->bitstream_bpp >= 24 && s->predictor == MEDIAN) {
218 av_log(avctx, AV_LOG_ERROR,
219 "Error: RGB is incompatible with median predictor\n");
220 return AVERROR(EINVAL);
223 ((uint8_t*)avctx->extradata)[0] = s->predictor | (s->decorrelate << 6);
224 ((uint8_t*)avctx->extradata)[1] = s->bitstream_bpp;
225 ((uint8_t*)avctx->extradata)[2] = s->interlaced ? 0x10 : 0x20;
227 ((uint8_t*)avctx->extradata)[2] |= 0x40;
228 ((uint8_t*)avctx->extradata)[3] = 0;
229 s->avctx->extradata_size = 4;
231 if (avctx->stats_in) {
232 char *p = avctx->stats_in;
234 for (i = 0; i < 3; i++)
235 for (j = 0; j < 256; j++)
239 for (i = 0; i < 3; i++) {
242 for (j = 0; j < 256; j++) {
243 s->stats[i][j] += strtol(p, &next, 0);
244 if (next == p) return -1;
248 if (p[0] == 0 || p[1] == 0 || p[2] == 0) break;
251 for (i = 0; i < 3; i++)
252 for (j = 0; j < 256; j++) {
253 int d = FFMIN(j, 256 - j);
255 s->stats[i][j] = 100000000 / (d + 1);
259 for (i = 0; i < 3; i++) {
260 ff_huff_gen_len_table(s->len[i], s->stats[i]);
262 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0) {
266 s->avctx->extradata_size +=
267 store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
271 for (i = 0; i < 3; i++) {
272 int pels = s->width * s->height / (i ? 40 : 10);
273 for (j = 0; j < 256; j++) {
274 int d = FFMIN(j, 256 - j);
275 s->stats[i][j] = pels/(d + 1);
279 for (i = 0; i < 3; i++)
280 for (j = 0; j < 256; j++)
284 if (ff_huffyuv_alloc_temp(s)) {
285 ff_huffyuv_common_end(s);
286 return AVERROR(ENOMEM);
293 static int encode_422_bitstream(HYuvContext *s, int offset, int count)
296 const uint8_t *y = s->temp[0] + offset;
297 const uint8_t *u = s->temp[1] + offset / 2;
298 const uint8_t *v = s->temp[2] + offset / 2;
300 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 2 * 4 * count) {
301 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
307 int y1 = y[2 * i + 1];\
313 if (s->flags & CODEC_FLAG_PASS1) {
314 for(i = 0; i < count; i++) {
322 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
325 for (i = 0; i < count; i++) {
328 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]);
332 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]);
337 for(i = 0; i < count; i++) {
339 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
340 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
341 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
342 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
348 static int encode_gray_bitstream(HYuvContext *s, int count)
352 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 4 * count) {
353 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
358 int y0 = s->temp[0][2 * i];\
359 int y1 = s->temp[0][2 * i + 1];
364 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
365 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
369 if (s->flags & CODEC_FLAG_PASS1) {
370 for (i = 0; i < count; i++) {
375 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
379 for (i = 0; i < count; i++) {
385 for (i = 0; i < count; i++) {
393 static inline int encode_bgra_bitstream(HYuvContext *s, int count, int planes)
397 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
398 4 * planes * count) {
399 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
404 int g = s->temp[0][planes == 3 ? 3 * i + 1 : 4 * i + G]; \
405 int b =(s->temp[0][planes == 3 ? 3 * i + 2 : 4 * i + B] - g) & 0xFF;\
406 int r =(s->temp[0][planes == 3 ? 3 * i + 0 : 4 * i + R] - g) & 0xFF;\
407 int a = s->temp[0][planes * i + A];
417 put_bits(&s->pb, s->len[1][g], s->bits[1][g]); \
418 put_bits(&s->pb, s->len[0][b], s->bits[0][b]); \
419 put_bits(&s->pb, s->len[2][r], s->bits[2][r]); \
421 put_bits(&s->pb, s->len[2][a], s->bits[2][a]);
423 if ((s->flags & CODEC_FLAG_PASS1) &&
424 (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
425 for (i = 0; i < count; i++) {
429 } else if (s->context || (s->flags & CODEC_FLAG_PASS1)) {
430 for (i = 0; i < count; i++) {
436 for (i = 0; i < count; i++) {
444 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
445 const AVFrame *pict, int *got_packet)
447 HYuvContext *s = avctx->priv_data;
448 const int width = s->width;
449 const int width2 = s->width>>1;
450 const int height = s->height;
451 const int fake_ystride = s->interlaced ? pict->linesize[0]*2 : pict->linesize[0];
452 const int fake_ustride = s->interlaced ? pict->linesize[1]*2 : pict->linesize[1];
453 const int fake_vstride = s->interlaced ? pict->linesize[2]*2 : pict->linesize[2];
454 const AVFrame * const p = pict;
455 int i, j, size = 0, ret;
457 if ((ret = ff_alloc_packet2(avctx, pkt, width * height * 3 * 4 + FF_MIN_BUFFER_SIZE)) < 0)
461 for (i = 0; i < 3; i++) {
462 ff_huff_gen_len_table(s->len[i], s->stats[i]);
463 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0)
465 size += store_table(s, s->len[i], &pkt->data[size]);
468 for (i = 0; i < 3; i++)
469 for (j = 0; j < 256; j++)
470 s->stats[i][j] >>= 1;
473 init_put_bits(&s->pb, pkt->data + size, pkt->size - size);
475 if (avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
476 avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
477 int lefty, leftu, leftv, y, cy;
479 put_bits(&s->pb, 8, leftv = p->data[2][0]);
480 put_bits(&s->pb, 8, lefty = p->data[0][1]);
481 put_bits(&s->pb, 8, leftu = p->data[1][0]);
482 put_bits(&s->pb, 8, p->data[0][0]);
484 lefty = sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
485 leftu = sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
486 leftv = sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
488 encode_422_bitstream(s, 2, width-2);
490 if (s->predictor==MEDIAN) {
491 int lefttopy, lefttopu, lefttopv;
494 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + p->linesize[0], width , lefty);
495 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + p->linesize[1], width2, leftu);
496 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + p->linesize[2], width2, leftv);
498 encode_422_bitstream(s, 0, width);
502 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + fake_ystride, 4, lefty);
503 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + fake_ustride, 2, leftu);
504 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + fake_vstride, 2, leftv);
506 encode_422_bitstream(s, 0, 4);
508 lefttopy = p->data[0][3];
509 lefttopu = p->data[1][1];
510 lefttopv = p->data[2][1];
511 s->dsp.sub_hfyu_median_prediction(s->temp[0], p->data[0]+4, p->data[0] + fake_ystride + 4, width - 4 , &lefty, &lefttopy);
512 s->dsp.sub_hfyu_median_prediction(s->temp[1], p->data[1]+2, p->data[1] + fake_ustride + 2, width2 - 2, &leftu, &lefttopu);
513 s->dsp.sub_hfyu_median_prediction(s->temp[2], p->data[2]+2, p->data[2] + fake_vstride + 2, width2 - 2, &leftv, &lefttopv);
514 encode_422_bitstream(s, 0, width - 4);
517 for (; y < height; y++,cy++) {
518 uint8_t *ydst, *udst, *vdst;
520 if (s->bitstream_bpp == 12) {
522 ydst = p->data[0] + p->linesize[0] * y;
523 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
524 encode_gray_bitstream(s, width);
527 if (y >= height) break;
529 ydst = p->data[0] + p->linesize[0] * y;
530 udst = p->data[1] + p->linesize[1] * cy;
531 vdst = p->data[2] + p->linesize[2] * cy;
533 s->dsp.sub_hfyu_median_prediction(s->temp[0], ydst - fake_ystride, ydst, width , &lefty, &lefttopy);
534 s->dsp.sub_hfyu_median_prediction(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
535 s->dsp.sub_hfyu_median_prediction(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
537 encode_422_bitstream(s, 0, width);
540 for (cy = y = 1; y < height; y++, cy++) {
541 uint8_t *ydst, *udst, *vdst;
543 /* encode a luma only line & y++ */
544 if (s->bitstream_bpp == 12) {
545 ydst = p->data[0] + p->linesize[0] * y;
547 if (s->predictor == PLANE && s->interlaced < y) {
548 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
550 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
552 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
554 encode_gray_bitstream(s, width);
556 if (y >= height) break;
559 ydst = p->data[0] + p->linesize[0] * y;
560 udst = p->data[1] + p->linesize[1] * cy;
561 vdst = p->data[2] + p->linesize[2] * cy;
563 if (s->predictor == PLANE && s->interlaced < cy) {
564 s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
565 s->dsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
566 s->dsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
568 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
569 leftu = sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
570 leftv = sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
572 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
573 leftu = sub_left_prediction(s, s->temp[1], udst, width2, leftu);
574 leftv = sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
577 encode_422_bitstream(s, 0, width);
580 } else if(avctx->pix_fmt == AV_PIX_FMT_RGB32) {
581 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
582 const int stride = -p->linesize[0];
583 const int fake_stride = -fake_ystride;
585 int leftr, leftg, leftb, lefta;
587 put_bits(&s->pb, 8, lefta = data[A]);
588 put_bits(&s->pb, 8, leftr = data[R]);
589 put_bits(&s->pb, 8, leftg = data[G]);
590 put_bits(&s->pb, 8, leftb = data[B]);
592 sub_left_prediction_bgr32(s, s->temp[0], data + 4, width - 1,
593 &leftr, &leftg, &leftb, &lefta);
594 encode_bgra_bitstream(s, width - 1, 4);
596 for (y = 1; y < s->height; y++) {
597 uint8_t *dst = data + y*stride;
598 if (s->predictor == PLANE && s->interlaced < y) {
599 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width * 4);
600 sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width,
601 &leftr, &leftg, &leftb, &lefta);
603 sub_left_prediction_bgr32(s, s->temp[0], dst, width,
604 &leftr, &leftg, &leftb, &lefta);
606 encode_bgra_bitstream(s, width, 4);
608 } else if (avctx->pix_fmt == AV_PIX_FMT_RGB24) {
609 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
610 const int stride = -p->linesize[0];
611 const int fake_stride = -fake_ystride;
613 int leftr, leftg, leftb;
615 put_bits(&s->pb, 8, leftr = data[0]);
616 put_bits(&s->pb, 8, leftg = data[1]);
617 put_bits(&s->pb, 8, leftb = data[2]);
618 put_bits(&s->pb, 8, 0);
620 sub_left_prediction_rgb24(s, s->temp[0], data + 3, width - 1,
621 &leftr, &leftg, &leftb);
622 encode_bgra_bitstream(s, width-1, 3);
624 for (y = 1; y < s->height; y++) {
625 uint8_t *dst = data + y * stride;
626 if (s->predictor == PLANE && s->interlaced < y) {
627 s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride,
629 sub_left_prediction_rgb24(s, s->temp[0], s->temp[1], width,
630 &leftr, &leftg, &leftb);
632 sub_left_prediction_rgb24(s, s->temp[0], dst, width,
633 &leftr, &leftg, &leftb);
635 encode_bgra_bitstream(s, width, 3);
638 av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
642 size += (put_bits_count(&s->pb) + 31) / 8;
643 put_bits(&s->pb, 16, 0);
644 put_bits(&s->pb, 15, 0);
647 if ((s->flags&CODEC_FLAG_PASS1) && (s->picture_number & 31) == 0) {
649 char *p = avctx->stats_out;
650 char *end = p + 1024*30;
651 for (i = 0; i < 3; i++) {
652 for (j = 0; j < 256; j++) {
653 snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
657 snprintf(p, end-p, "\n");
661 avctx->stats_out[0] = '\0';
662 if (!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
663 flush_put_bits(&s->pb);
664 s->dsp.bswap_buf((uint32_t*)pkt->data, (uint32_t*)pkt->data, size);
669 pkt->size = size * 4;
670 pkt->flags |= AV_PKT_FLAG_KEY;
676 static av_cold int encode_end(AVCodecContext *avctx)
678 HYuvContext *s = avctx->priv_data;
680 ff_huffyuv_common_end(s);
682 av_freep(&avctx->extradata);
683 av_freep(&avctx->stats_out);
685 av_frame_free(&avctx->coded_frame);
690 #if CONFIG_HUFFYUV_ENCODER
691 AVCodec ff_huffyuv_encoder = {
693 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
694 .type = AVMEDIA_TYPE_VIDEO,
695 .id = AV_CODEC_ID_HUFFYUV,
696 .priv_data_size = sizeof(HYuvContext),
698 .encode2 = encode_frame,
700 .pix_fmts = (const enum AVPixelFormat[]){
701 AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
702 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
707 #if CONFIG_FFVHUFF_ENCODER
708 AVCodec ff_ffvhuff_encoder = {
710 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),
711 .type = AVMEDIA_TYPE_VIDEO,
712 .id = AV_CODEC_ID_FFVHUFF,
713 .priv_data_size = sizeof(HYuvContext),
715 .encode2 = encode_frame,
717 .pix_fmts = (const enum AVPixelFormat[]){
718 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
719 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE