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"
36 static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst,
37 uint8_t *src, int w, int left)
41 for (i = 0; i < w; i++) {
42 const int temp = src[i];
48 for (i = 0; i < 16; i++) {
49 const int temp = src[i];
53 s->hencdsp.diff_bytes(dst + 16, src + 16, src + 15, w - 16);
58 static inline void sub_left_prediction_bgr32(HYuvContext *s, uint8_t *dst,
60 int *red, int *green, int *blue,
70 for (i = 0; i < FFMIN(w, 4); i++) {
71 const int rt = src[i * 4 + R];
72 const int gt = src[i * 4 + G];
73 const int bt = src[i * 4 + B];
74 const int at = src[i * 4 + A];
75 dst[i * 4 + R] = rt - r;
76 dst[i * 4 + G] = gt - g;
77 dst[i * 4 + B] = bt - b;
78 dst[i * 4 + A] = at - a;
85 s->hencdsp.diff_bytes(dst + 16, src + 16, src + 12, w * 4 - 16);
87 *red = src[(w - 1) * 4 + R];
88 *green = src[(w - 1) * 4 + G];
89 *blue = src[(w - 1) * 4 + B];
90 *alpha = src[(w - 1) * 4 + A];
93 static inline void sub_left_prediction_rgb24(HYuvContext *s, uint8_t *dst,
95 int *red, int *green, int *blue)
102 for (i = 0; i < FFMIN(w, 16); i++) {
103 const int rt = src[i * 3 + 0];
104 const int gt = src[i * 3 + 1];
105 const int bt = src[i * 3 + 2];
106 dst[i * 3 + 0] = rt - r;
107 dst[i * 3 + 1] = gt - g;
108 dst[i * 3 + 2] = bt - b;
114 s->hencdsp.diff_bytes(dst + 48, src + 48, src + 48 - 3, w * 3 - 48);
116 *red = src[(w - 1) * 3 + 0];
117 *green = src[(w - 1) * 3 + 1];
118 *blue = src[(w - 1) * 3 + 2];
121 static int store_table(HYuvContext *s, const uint8_t *len, uint8_t *buf)
126 for (i = 0; i < 256;) {
130 for (; i < 256 && len[i] == val && repeat < 255; i++)
133 assert(val < 32 && val >0 && repeat<256 && repeat>0);
136 buf[index++] = repeat;
138 buf[index++] = val | (repeat << 5);
145 static av_cold int encode_init(AVCodecContext *avctx)
147 HYuvContext *s = avctx->priv_data;
150 ff_huffyuv_common_init(avctx);
151 ff_huffyuvencdsp_init(&s->hencdsp);
153 avctx->extradata = av_mallocz(1024*30); // 256*3+4 == 772
154 avctx->stats_out = av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
157 avctx->coded_frame = av_frame_alloc();
158 if (!avctx->extradata || !avctx->stats_out || !avctx->coded_frame)
159 return AVERROR(ENOMEM);
161 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
162 avctx->coded_frame->key_frame = 1;
164 switch (avctx->pix_fmt) {
165 case AV_PIX_FMT_YUV420P:
166 case AV_PIX_FMT_YUV422P:
168 av_log(avctx, AV_LOG_ERROR, "Width must be even for this colorspace.\n");
171 s->bitstream_bpp = avctx->pix_fmt == AV_PIX_FMT_YUV420P ? 12 : 16;
173 case AV_PIX_FMT_RGB32:
174 s->bitstream_bpp = 32;
176 case AV_PIX_FMT_RGB24:
177 s->bitstream_bpp = 24;
180 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
183 avctx->bits_per_coded_sample = s->bitstream_bpp;
184 s->decorrelate = s->bitstream_bpp >= 24;
185 s->predictor = avctx->prediction_method;
186 s->interlaced = avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
187 if (avctx->context_model == 1) {
188 s->context = avctx->context_model;
189 if (s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)) {
190 av_log(avctx, AV_LOG_ERROR,
191 "context=1 is not compatible with "
192 "2 pass huffyuv encoding\n");
197 if (avctx->codec->id == AV_CODEC_ID_HUFFYUV) {
198 if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
199 av_log(avctx, AV_LOG_ERROR,
200 "Error: YV12 is not supported by huffyuv; use "
201 "vcodec=ffvhuff or format=422p\n");
204 if (avctx->context_model) {
205 av_log(avctx, AV_LOG_ERROR,
206 "Error: per-frame huffman tables are not supported "
207 "by huffyuv; use vcodec=ffvhuff\n");
210 if (s->interlaced != ( s->height > 288 ))
211 av_log(avctx, AV_LOG_INFO,
212 "using huffyuv 2.2.0 or newer interlacing flag\n");
215 if (s->bitstream_bpp >= 24 && s->predictor == MEDIAN) {
216 av_log(avctx, AV_LOG_ERROR,
217 "Error: RGB is incompatible with median predictor\n");
221 ((uint8_t*)avctx->extradata)[0] = s->predictor | (s->decorrelate << 6);
222 ((uint8_t*)avctx->extradata)[1] = s->bitstream_bpp;
223 ((uint8_t*)avctx->extradata)[2] = s->interlaced ? 0x10 : 0x20;
225 ((uint8_t*)avctx->extradata)[2] |= 0x40;
226 ((uint8_t*)avctx->extradata)[3] = 0;
227 s->avctx->extradata_size = 4;
229 if (avctx->stats_in) {
230 char *p = avctx->stats_in;
232 for (i = 0; i < 3; i++)
233 for (j = 0; j < 256; j++)
237 for (i = 0; i < 3; i++) {
240 for (j = 0; j < 256; j++) {
241 s->stats[i][j] += strtol(p, &next, 0);
242 if (next == p) return -1;
246 if (p[0] == 0 || p[1] == 0 || p[2] == 0) break;
249 for (i = 0; i < 3; i++)
250 for (j = 0; j < 256; j++) {
251 int d = FFMIN(j, 256 - j);
253 s->stats[i][j] = 100000000 / (d + 1);
257 for (i = 0; i < 3; i++) {
258 ff_huff_gen_len_table(s->len[i], s->stats[i]);
260 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0) {
264 s->avctx->extradata_size +=
265 store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
269 for (i = 0; i < 3; i++) {
270 int pels = s->width * s->height / (i ? 40 : 10);
271 for (j = 0; j < 256; j++) {
272 int d = FFMIN(j, 256 - j);
273 s->stats[i][j] = pels/(d + 1);
277 for (i = 0; i < 3; i++)
278 for (j = 0; j < 256; j++)
282 ff_huffyuv_alloc_temp(s);
288 static int encode_422_bitstream(HYuvContext *s, int offset, int count)
291 const uint8_t *y = s->temp[0] + offset;
292 const uint8_t *u = s->temp[1] + offset / 2;
293 const uint8_t *v = s->temp[2] + offset / 2;
295 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 2 * 4 * count) {
296 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
302 int y1 = y[2 * i + 1];\
308 if (s->flags & CODEC_FLAG_PASS1) {
309 for(i = 0; i < count; i++) {
317 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
320 for (i = 0; i < count; i++) {
323 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
325 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
327 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
329 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
332 for(i = 0; i < count; i++) {
334 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
335 put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
336 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
337 put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
343 static int encode_gray_bitstream(HYuvContext *s, int count)
347 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < 4 * count) {
348 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
353 int y0 = s->temp[0][2 * i];\
354 int y1 = s->temp[0][2 * i + 1];
359 put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
360 put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
364 if (s->flags & CODEC_FLAG_PASS1) {
365 for (i = 0; i < count; i++) {
370 if (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)
374 for (i = 0; i < count; i++) {
380 for (i = 0; i < count; i++) {
388 static inline int encode_bgra_bitstream(HYuvContext *s, int count, int planes)
392 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
393 4 * planes * count) {
394 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
399 int g = s->temp[0][planes == 3 ? 3 * i + 1 : 4 * i + G]; \
400 int b = s->temp[0][planes == 3 ? 3 * i + 2 : 4 * i + B] - g & 0xFF; \
401 int r = s->temp[0][planes == 3 ? 3 * i + 0 : 4 * i + R] - g & 0xFF; \
402 int a = s->temp[0][planes * i + A];
412 put_bits(&s->pb, s->len[1][g], s->bits[1][g]); \
413 put_bits(&s->pb, s->len[0][b], s->bits[0][b]); \
414 put_bits(&s->pb, s->len[2][r], s->bits[2][r]); \
416 put_bits(&s->pb, s->len[2][a], s->bits[2][a]);
418 if ((s->flags & CODEC_FLAG_PASS1) &&
419 (s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
420 for (i = 0; i < count; i++) {
424 } else if (s->context || (s->flags & CODEC_FLAG_PASS1)) {
425 for (i = 0; i < count; i++) {
431 for (i = 0; i < count; i++) {
439 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
440 const AVFrame *pict, int *got_packet)
442 HYuvContext *s = avctx->priv_data;
443 const int width = s->width;
444 const int width2 = s->width>>1;
445 const int height = s->height;
446 const int fake_ystride = s->interlaced ? pict->linesize[0]*2 : pict->linesize[0];
447 const int fake_ustride = s->interlaced ? pict->linesize[1]*2 : pict->linesize[1];
448 const int fake_vstride = s->interlaced ? pict->linesize[2]*2 : pict->linesize[2];
449 const AVFrame * const p = pict;
450 int i, j, size = 0, ret;
453 (ret = av_new_packet(pkt, width * height * 3 * 4 + FF_MIN_BUFFER_SIZE)) < 0) {
454 av_log(avctx, AV_LOG_ERROR, "Error allocating output packet.\n");
459 for (i = 0; i < 3; i++) {
460 ff_huff_gen_len_table(s->len[i], s->stats[i]);
461 if (ff_huffyuv_generate_bits_table(s->bits[i], s->len[i]) < 0)
463 size += store_table(s, s->len[i], &pkt->data[size]);
466 for (i = 0; i < 3; i++)
467 for (j = 0; j < 256; j++)
468 s->stats[i][j] >>= 1;
471 init_put_bits(&s->pb, pkt->data + size, pkt->size - size);
473 if (avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
474 avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
475 int lefty, leftu, leftv, y, cy;
477 put_bits(&s->pb, 8, leftv = p->data[2][0]);
478 put_bits(&s->pb, 8, lefty = p->data[0][1]);
479 put_bits(&s->pb, 8, leftu = p->data[1][0]);
480 put_bits(&s->pb, 8, p->data[0][0]);
482 lefty = sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
483 leftu = sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
484 leftv = sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
486 encode_422_bitstream(s, 2, width-2);
488 if (s->predictor==MEDIAN) {
489 int lefttopy, lefttopu, lefttopv;
492 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + p->linesize[0], width , lefty);
493 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + p->linesize[1], width2, leftu);
494 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + p->linesize[2], width2, leftv);
496 encode_422_bitstream(s, 0, width);
500 lefty = sub_left_prediction(s, s->temp[0], p->data[0] + fake_ystride, 4, lefty);
501 leftu = sub_left_prediction(s, s->temp[1], p->data[1] + fake_ustride, 2, leftu);
502 leftv = sub_left_prediction(s, s->temp[2], p->data[2] + fake_vstride, 2, leftv);
504 encode_422_bitstream(s, 0, 4);
506 lefttopy = p->data[0][3];
507 lefttopu = p->data[1][1];
508 lefttopv = p->data[2][1];
509 s->hencdsp.sub_hfyu_median_pred(s->temp[0], p->data[0] + 4, p->data[0] + fake_ystride + 4, width - 4, &lefty, &lefttopy);
510 s->hencdsp.sub_hfyu_median_pred(s->temp[1], p->data[1] + 2, p->data[1] + fake_ustride + 2, width2 - 2, &leftu, &lefttopu);
511 s->hencdsp.sub_hfyu_median_pred(s->temp[2], p->data[2] + 2, p->data[2] + fake_vstride + 2, width2 - 2, &leftv, &lefttopv);
512 encode_422_bitstream(s, 0, width - 4);
515 for (; y < height; y++,cy++) {
516 uint8_t *ydst, *udst, *vdst;
518 if (s->bitstream_bpp == 12) {
520 ydst = p->data[0] + p->linesize[0] * y;
521 s->hencdsp.sub_hfyu_median_pred(s->temp[0], ydst - fake_ystride, ydst, width, &lefty, &lefttopy);
522 encode_gray_bitstream(s, width);
525 if (y >= height) break;
527 ydst = p->data[0] + p->linesize[0] * y;
528 udst = p->data[1] + p->linesize[1] * cy;
529 vdst = p->data[2] + p->linesize[2] * cy;
531 s->hencdsp.sub_hfyu_median_pred(s->temp[0], ydst - fake_ystride, ydst, width, &lefty, &lefttopy);
532 s->hencdsp.sub_hfyu_median_pred(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
533 s->hencdsp.sub_hfyu_median_pred(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
535 encode_422_bitstream(s, 0, width);
538 for (cy = y = 1; y < height; y++, cy++) {
539 uint8_t *ydst, *udst, *vdst;
541 /* encode a luma only line & y++ */
542 if (s->bitstream_bpp == 12) {
543 ydst = p->data[0] + p->linesize[0] * y;
545 if (s->predictor == PLANE && s->interlaced < y) {
546 s->hencdsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
548 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
550 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
552 encode_gray_bitstream(s, width);
554 if (y >= height) break;
557 ydst = p->data[0] + p->linesize[0] * y;
558 udst = p->data[1] + p->linesize[1] * cy;
559 vdst = p->data[2] + p->linesize[2] * cy;
561 if (s->predictor == PLANE && s->interlaced < cy) {
562 s->hencdsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
563 s->hencdsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
564 s->hencdsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
566 lefty = sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
567 leftu = sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
568 leftv = sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
570 lefty = sub_left_prediction(s, s->temp[0], ydst, width , lefty);
571 leftu = sub_left_prediction(s, s->temp[1], udst, width2, leftu);
572 leftv = sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
575 encode_422_bitstream(s, 0, width);
578 } else if(avctx->pix_fmt == AV_PIX_FMT_RGB32) {
579 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
580 const int stride = -p->linesize[0];
581 const int fake_stride = -fake_ystride;
583 int leftr, leftg, leftb, lefta;
585 put_bits(&s->pb, 8, lefta = data[A]);
586 put_bits(&s->pb, 8, leftr = data[R]);
587 put_bits(&s->pb, 8, leftg = data[G]);
588 put_bits(&s->pb, 8, leftb = data[B]);
590 sub_left_prediction_bgr32(s, s->temp[0], data + 4, width - 1,
591 &leftr, &leftg, &leftb, &lefta);
592 encode_bgra_bitstream(s, width - 1, 4);
594 for (y = 1; y < s->height; y++) {
595 uint8_t *dst = data + y*stride;
596 if (s->predictor == PLANE && s->interlaced < y) {
597 s->hencdsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width * 4);
598 sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width,
599 &leftr, &leftg, &leftb, &lefta);
601 sub_left_prediction_bgr32(s, s->temp[0], dst, width,
602 &leftr, &leftg, &leftb, &lefta);
604 encode_bgra_bitstream(s, width, 4);
606 } else if (avctx->pix_fmt == AV_PIX_FMT_RGB24) {
607 uint8_t *data = p->data[0] + (height - 1) * p->linesize[0];
608 const int stride = -p->linesize[0];
609 const int fake_stride = -fake_ystride;
611 int leftr, leftg, leftb;
613 put_bits(&s->pb, 8, leftr = data[0]);
614 put_bits(&s->pb, 8, leftg = data[1]);
615 put_bits(&s->pb, 8, leftb = data[2]);
616 put_bits(&s->pb, 8, 0);
618 sub_left_prediction_rgb24(s, s->temp[0], data + 3, width - 1,
619 &leftr, &leftg, &leftb);
620 encode_bgra_bitstream(s, width-1, 3);
622 for (y = 1; y < s->height; y++) {
623 uint8_t *dst = data + y * stride;
624 if (s->predictor == PLANE && s->interlaced < y) {
625 s->hencdsp.diff_bytes(s->temp[1], dst, dst - fake_stride,
627 sub_left_prediction_rgb24(s, s->temp[0], s->temp[1], width,
628 &leftr, &leftg, &leftb);
630 sub_left_prediction_rgb24(s, s->temp[0], dst, width,
631 &leftr, &leftg, &leftb);
633 encode_bgra_bitstream(s, width, 3);
636 av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
640 size += (put_bits_count(&s->pb) + 31) / 8;
641 put_bits(&s->pb, 16, 0);
642 put_bits(&s->pb, 15, 0);
645 if ((s->flags&CODEC_FLAG_PASS1) && (s->picture_number & 31) == 0) {
647 char *p = avctx->stats_out;
648 char *end = p + 1024*30;
649 for (i = 0; i < 3; i++) {
650 for (j = 0; j < 256; j++) {
651 snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
655 snprintf(p, end-p, "\n");
659 avctx->stats_out[0] = '\0';
660 if (!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)) {
661 flush_put_bits(&s->pb);
662 s->bdsp.bswap_buf((uint32_t *) pkt->data, (uint32_t *) pkt->data, size);
667 pkt->size = size * 4;
668 pkt->flags |= AV_PKT_FLAG_KEY;
674 static av_cold int encode_end(AVCodecContext *avctx)
676 HYuvContext *s = avctx->priv_data;
678 ff_huffyuv_common_end(s);
680 av_freep(&avctx->extradata);
681 av_freep(&avctx->stats_out);
683 av_frame_free(&avctx->coded_frame);
688 AVCodec ff_huffyuv_encoder = {
690 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
691 .type = AVMEDIA_TYPE_VIDEO,
692 .id = AV_CODEC_ID_HUFFYUV,
693 .priv_data_size = sizeof(HYuvContext),
695 .encode2 = encode_frame,
697 .pix_fmts = (const enum AVPixelFormat[]){
698 AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
699 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
701 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
702 FF_CODEC_CAP_INIT_CLEANUP,
705 #if CONFIG_FFVHUFF_ENCODER
706 AVCodec ff_ffvhuff_encoder = {
708 .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),
709 .type = AVMEDIA_TYPE_VIDEO,
710 .id = AV_CODEC_ID_FFVHUFF,
711 .priv_data_size = sizeof(HYuvContext),
713 .encode2 = encode_frame,
715 .pix_fmts = (const enum AVPixelFormat[]){
716 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_RGB24,
717 AV_PIX_FMT_RGB32, AV_PIX_FMT_NONE
719 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
720 FF_CODEC_CAP_INIT_CLEANUP,