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
32 #include "huffyuvencdsp.h"
35 static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst,
36 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->hencdsp.diff_bytes(dst + 16, src + 16, src + 15, w - 16);
57 static inline void sub_left_prediction_bgr32(HYuvContext *s, uint8_t *dst,
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->hencdsp.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->hencdsp.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 assert(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);
150 ff_huffyuvencdsp_init(&s->hencdsp);
152 avctx->extradata = av_mallocz(1024*30); // 256*3+4 == 772
153 avctx->stats_out = av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
156 avctx->coded_frame = av_frame_alloc();
157 if (!avctx->coded_frame)
158 return AVERROR(ENOMEM);
160 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
161 avctx->coded_frame->key_frame = 1;
163 switch (avctx->pix_fmt) {
164 case AV_PIX_FMT_YUV420P:
165 case AV_PIX_FMT_YUV422P:
167 av_log(avctx, AV_LOG_ERROR, "Width must be even for this colorspace.\n");
170 s->bitstream_bpp = avctx->pix_fmt == AV_PIX_FMT_YUV420P ? 12 : 16;
172 case AV_PIX_FMT_RGB32:
173 s->bitstream_bpp = 32;
175 case AV_PIX_FMT_RGB24:
176 s->bitstream_bpp = 24;
179 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
182 avctx->bits_per_coded_sample = s->bitstream_bpp;
183 s->decorrelate = s->bitstream_bpp >= 24;
184 s->predictor = avctx->prediction_method;
185 s->interlaced = avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
186 if (avctx->context_model == 1) {
187 s->context = avctx->context_model;
188 if (s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)) {
189 av_log(avctx, AV_LOG_ERROR,
190 "context=1 is not compatible with "
191 "2 pass huffyuv encoding\n");
196 if (avctx->codec->id == AV_CODEC_ID_HUFFYUV) {
197 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
198 av_log(avctx, AV_LOG_ERROR,
199 "Error: YV12 is not supported by huffyuv; use "
200 "vcodec=ffvhuff or format=422p\n");
203 if (avctx->context_model) {
204 av_log(avctx, AV_LOG_ERROR,
205 "Error: per-frame huffman tables are not supported "
206 "by huffyuv; use vcodec=ffvhuff\n");
209 if (s->interlaced != ( s->height > 288 ))
210 av_log(avctx, AV_LOG_INFO,
211 "using huffyuv 2.2.0 or newer interlacing flag\n");
214 if (s->bitstream_bpp >= 24 && s->predictor == MEDIAN) {
215 av_log(avctx, AV_LOG_ERROR,
216 "Error: RGB is incompatible with median predictor\n");
220 ((uint8_t*)avctx->extradata)[0] = s->predictor | (s->decorrelate << 6);
221 ((uint8_t*)avctx->extradata)[1] = s->bitstream_bpp;
222 ((uint8_t*)avctx->extradata)[2] = s->interlaced ? 0x10 : 0x20;
224 ((uint8_t*)avctx->extradata)[2] |= 0x40;
225 ((uint8_t*)avctx->extradata)[3] = 0;
226 s->avctx->extradata_size = 4;
228 if (avctx->stats_in) {
229 char *p = avctx->stats_in;
231 for (i = 0; i < 3; i++)
232 for (j = 0; j < 256; j++)
236 for (i = 0; i < 3; i++) {
239 for (j = 0; j < 256; j++) {
240 s->stats[i][j] += strtol(p, &next, 0);
241 if (next == p) return -1;
245 if (p[0] == 0 || p[1] == 0 || p[2] == 0) break;
248 for (i = 0; i < 3; i++)
249 for (j = 0; j < 256; j++) {
250 int d = FFMIN(j, 256 - j);
252 s->stats[i][j] = 100000000 / (d + 1);
256 for (i = 0; i < 3; i++) {
257 ff_huff_gen_len_table(s->len[i], s->stats[i]);
259 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0) {
263 s->avctx->extradata_size +=
264 store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
268 for (i = 0; i < 3; i++) {
269 int pels = s->width * s->height / (i ? 40 : 10);
270 for (j = 0; j < 256; j++) {
271 int d = FFMIN(j, 256 - j);
272 s->stats[i][j] = pels/(d + 1);
276 for (i = 0; i < 3; i++)
277 for (j = 0; j < 256; j++)
281 ff_huffyuv_alloc_temp(s);
287 static int encode_422_bitstream(HYuvContext *s, int offset, int count)
290 const uint8_t *y = s->temp[0] + offset;
291 const uint8_t *u = s->temp[1] + offset / 2;
292 const uint8_t *v = s->temp[2] + offset / 2;
294 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 2 * 4 * count) {
295 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
301 int y1 = y[2 * i + 1];\
307 if (s->flags & CODEC_FLAG_PASS1) {
308 for(i = 0; i < count; i++) {
316 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
319 for (i = 0; i < count; i++) {
322 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
324 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
326 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
328 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
331 for(i = 0; i < count; i++) {
333 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
334 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
335 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
336 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
342 static int encode_gray_bitstream(HYuvContext *s, int count)
346 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 4 * count) {
347 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
352 int y0 = s->temp[0][2 * i];\
353 int y1 = s->temp[0][2 * i + 1];
358 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
359 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
363 if (s->flags & CODEC_FLAG_PASS1) {
364 for (i = 0; i < count; i++) {
369 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
373 for (i = 0; i < count; i++) {
379 for (i = 0; i < count; i++) {
387 static inline int encode_bgra_bitstream(HYuvContext *s, int count, int planes)
391 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
392 4 * planes * count) {
393 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
398 int g = s->temp[0][planes == 3 ? 3 * i + 1 : 4 * i + G]; \
399 int b = s->temp[0][planes == 3 ? 3 * i + 2 : 4 * i + B] - g & 0xFF; \
400 int r = s->temp[0][planes == 3 ? 3 * i + 0 : 4 * i + R] - g & 0xFF; \
401 int a = s->temp[0][planes * i + A];
411 put_bits(&s->pb, s->len[1][g], s->bits[1][g]); \
412 put_bits(&s->pb, s->len[0][b], s->bits[0][b]); \
413 put_bits(&s->pb, s->len[2][r], s->bits[2][r]); \
415 put_bits(&s->pb, s->len[2][a], s->bits[2][a]);
417 if ((s->flags & CODEC_FLAG_PASS1) &&
418 (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
419 for (i = 0; i < count; i++) {
423 } else if (s->context || (s->flags & CODEC_FLAG_PASS1)) {
424 for (i = 0; i < count; i++) {
430 for (i = 0; i < count; i++) {
438 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
439 const AVFrame *pict, int *got_packet)
441 HYuvContext *s = avctx->priv_data;
442 const int width = s->width;
443 const int width2 = s->width>>1;
444 const int height = s->height;
445 const int fake_ystride = s->interlaced ? pict->linesize[0]*2 : pict->linesize[0];
446 const int fake_ustride = s->interlaced ? pict->linesize[1]*2 : pict->linesize[1];
447 const int fake_vstride = s->interlaced ? pict->linesize[2]*2 : pict->linesize[2];
448 const AVFrame * const p = pict;
449 int i, j, size = 0, ret;
452 (ret = av_new_packet(pkt, width * height * 3 * 4 + FF_MIN_BUFFER_SIZE)) < 0) {
453 av_log(avctx, AV_LOG_ERROR, "Error allocating output packet.\n");
458 for (i = 0; i < 3; i++) {
459 ff_huff_gen_len_table(s->len[i], s->stats[i]);
460 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0)
462 size += store_table(s, s->len[i], &pkt->data[size]);
465 for (i = 0; i < 3; i++)
466 for (j = 0; j < 256; j++)
467 s->stats[i][j] >>= 1;
470 init_put_bits(&s->pb, pkt->data + size, pkt->size - size);
472 if (avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
473 avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
474 int lefty, leftu, leftv, y, cy;
476 put_bits(&s->pb, 8, leftv = p->data[2][0]);
477 put_bits(&s->pb, 8, lefty = p->data[0][1]);
478 put_bits(&s->pb, 8, leftu = p->data[1][0]);
479 put_bits(&s->pb, 8, p->data[0][0]);
481 lefty = sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
482 leftu = sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
483 leftv = sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
485 encode_422_bitstream(s, 2, width-2);
487 if (s->predictor==MEDIAN) {
488 int lefttopy, lefttopu, lefttopv;
491 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + p->linesize[0], width , lefty);
492 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + p->linesize[1], width2, leftu);
493 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + p->linesize[2], width2, leftv);
495 encode_422_bitstream(s, 0, width);
499 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + fake_ystride, 4, lefty);
500 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + fake_ustride, 2, leftu);
501 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + fake_vstride, 2, leftv);
503 encode_422_bitstream(s, 0, 4);
505 lefttopy = p->data[0][3];
506 lefttopu = p->data[1][1];
507 lefttopv = p->data[2][1];
508 s->hencdsp.sub_hfyu_median_pred(s->temp[0], p->data[0] + 4, p->data[0] + fake_ystride + 4, width - 4, &lefty, &lefttopy);
509 s->hencdsp.sub_hfyu_median_pred(s->temp[1], p->data[1] + 2, p->data[1] + fake_ustride + 2, width2 - 2, &leftu, &lefttopu);
510 s->hencdsp.sub_hfyu_median_pred(s->temp[2], p->data[2] + 2, p->data[2] + fake_vstride + 2, width2 - 2, &leftv, &lefttopv);
511 encode_422_bitstream(s, 0, width - 4);
514 for (; y < height; y++,cy++) {
515 uint8_t *ydst, *udst, *vdst;
517 if (s->bitstream_bpp == 12) {
519 ydst = p->data[0] + p->linesize[0] * y;
520 s->hencdsp.sub_hfyu_median_pred(s->temp[0], ydst - fake_ystride, ydst, width, &lefty, &lefttopy);
521 encode_gray_bitstream(s, width);
524 if (y >= height) break;
526 ydst = p->data[0] + p->linesize[0] * y;
527 udst = p->data[1] + p->linesize[1] * cy;
528 vdst = p->data[2] + p->linesize[2] * cy;
530 s->hencdsp.sub_hfyu_median_pred(s->temp[0], ydst - fake_ystride, ydst, width, &lefty, &lefttopy);
531 s->hencdsp.sub_hfyu_median_pred(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
532 s->hencdsp.sub_hfyu_median_pred(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
534 encode_422_bitstream(s, 0, width);
537 for (cy = y = 1; y < height; y++, cy++) {
538 uint8_t *ydst, *udst, *vdst;
540 /* encode a luma only line & y++ */
541 if (s->bitstream_bpp == 12) {
542 ydst = p->data[0] + p->linesize[0] * y;
544 if (s->predictor == PLANE && s->interlaced < y) {
545 s->hencdsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
547 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
549 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
551 encode_gray_bitstream(s, width);
553 if (y >= height) break;
556 ydst = p->data[0] + p->linesize[0] * y;
557 udst = p->data[1] + p->linesize[1] * cy;
558 vdst = p->data[2] + p->linesize[2] * cy;
560 if (s->predictor == PLANE && s->interlaced < cy) {
561 s->hencdsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
562 s->hencdsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
563 s->hencdsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
565 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
566 leftu = sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
567 leftv = sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
569 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
570 leftu = sub_left_prediction(s, s->temp[1], udst, width2, leftu);
571 leftv = sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
574 encode_422_bitstream(s, 0, width);
577 } else if(avctx->pix_fmt == AV_PIX_FMT_RGB32) {
578 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
579 const int stride = -p->linesize[0];
580 const int fake_stride = -fake_ystride;
582 int leftr, leftg, leftb, lefta;
584 put_bits(&s->pb, 8, lefta = data[A]);
585 put_bits(&s->pb, 8, leftr = data[R]);
586 put_bits(&s->pb, 8, leftg = data[G]);
587 put_bits(&s->pb, 8, leftb = data[B]);
589 sub_left_prediction_bgr32(s, s->temp[0], data + 4, width - 1,
590 &leftr, &leftg, &leftb, &lefta);
591 encode_bgra_bitstream(s, width - 1, 4);
593 for (y = 1; y < s->height; y++) {
594 uint8_t *dst = data + y*stride;
595 if (s->predictor == PLANE && s->interlaced < y) {
596 s->hencdsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width * 4);
597 sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width,
598 &leftr, &leftg, &leftb, &lefta);
600 sub_left_prediction_bgr32(s, s->temp[0], dst, width,
601 &leftr, &leftg, &leftb, &lefta);
603 encode_bgra_bitstream(s, width, 4);
605 } else if (avctx->pix_fmt == AV_PIX_FMT_RGB24) {
606 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
607 const int stride = -p->linesize[0];
608 const int fake_stride = -fake_ystride;
610 int leftr, leftg, leftb;
612 put_bits(&s->pb, 8, leftr = data[0]);
613 put_bits(&s->pb, 8, leftg = data[1]);
614 put_bits(&s->pb, 8, leftb = data[2]);
615 put_bits(&s->pb, 8, 0);
617 sub_left_prediction_rgb24(s, s->temp[0], data + 3, width - 1,
618 &leftr, &leftg, &leftb);
619 encode_bgra_bitstream(s, width-1, 3);
621 for (y = 1; y < s->height; y++) {
622 uint8_t *dst = data + y * stride;
623 if (s->predictor == PLANE && s->interlaced < y) {
624 s->hencdsp.diff_bytes(s->temp[1], dst, dst - fake_stride,
626 sub_left_prediction_rgb24(s, s->temp[0], s->temp[1], width,
627 &leftr, &leftg, &leftb);
629 sub_left_prediction_rgb24(s, s->temp[0], dst, width,
630 &leftr, &leftg, &leftb);
632 encode_bgra_bitstream(s, width, 3);
635 av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
639 size += (put_bits_count(&s->pb) + 31) / 8;
640 put_bits(&s->pb, 16, 0);
641 put_bits(&s->pb, 15, 0);
644 if ((s->flags&CODEC_FLAG_PASS1) && (s->picture_number & 31) == 0) {
646 char *p = avctx->stats_out;
647 char *end = p + 1024*30;
648 for (i = 0; i < 3; i++) {
649 for (j = 0; j < 256; j++) {
650 snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
654 snprintf(p, end-p, "\n");
658 avctx->stats_out[0] = '\0';
659 if (!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
660 flush_put_bits(&s->pb);
661 s->bdsp.bswap_buf((uint32_t *) pkt->data, (uint32_t *) pkt->data, size);
666 pkt->size = size * 4;
667 pkt->flags |= AV_PKT_FLAG_KEY;
673 static av_cold int encode_end(AVCodecContext *avctx)
675 HYuvContext *s = avctx->priv_data;
677 ff_huffyuv_common_end(s);
679 av_freep(&avctx->extradata);
680 av_freep(&avctx->stats_out);
682 av_frame_free(&avctx->coded_frame);
687 AVCodec ff_huffyuv_encoder = {
689 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
690 .type = AVMEDIA_TYPE_VIDEO,
691 .id = AV_CODEC_ID_HUFFYUV,
692 .priv_data_size = sizeof(HYuvContext),
694 .encode2 = encode_frame,
696 .pix_fmts = (const enum AVPixelFormat[]){
697 AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
698 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