3 * Copyright (c) 2012 Jan Ekström
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libavutil/imgutils.h"
28 #include "libavutil/intreadwrite.h"
31 #include "bytestream.h"
38 /* Compare huffentry symbols */
39 static int huff_cmp_sym(const void *a, const void *b)
41 const HuffEntry *aa = a, *bb = b;
42 return aa->sym - bb->sym;
45 static av_cold int utvideo_encode_close(AVCodecContext *avctx)
47 UtvideoContext *c = avctx->priv_data;
50 av_freep(&avctx->coded_frame);
51 av_freep(&c->slice_bits);
52 for (i = 0; i < 4; i++)
53 av_freep(&c->slice_buffer[i]);
58 static av_cold int utvideo_encode_init(AVCodecContext *avctx)
60 UtvideoContext *c = avctx->priv_data;
62 uint32_t original_format;
65 c->frame_info_size = 4;
66 c->slice_stride = FFALIGN(avctx->width, 32);
68 switch (avctx->pix_fmt) {
69 case AV_PIX_FMT_RGB24:
71 avctx->codec_tag = MKTAG('U', 'L', 'R', 'G');
72 original_format = UTVIDEO_RGB;
76 avctx->codec_tag = MKTAG('U', 'L', 'R', 'A');
77 original_format = UTVIDEO_RGBA;
79 case AV_PIX_FMT_YUV420P:
80 if (avctx->width & 1 || avctx->height & 1) {
81 av_log(avctx, AV_LOG_ERROR,
82 "4:2:0 video requires even width and height.\n");
83 return AVERROR_INVALIDDATA;
86 if (avctx->colorspace == AVCOL_SPC_BT709)
87 avctx->codec_tag = MKTAG('U', 'L', 'H', '0');
89 avctx->codec_tag = MKTAG('U', 'L', 'Y', '0');
90 original_format = UTVIDEO_420;
92 case AV_PIX_FMT_YUV422P:
93 if (avctx->width & 1) {
94 av_log(avctx, AV_LOG_ERROR,
95 "4:2:2 video requires even width.\n");
96 return AVERROR_INVALIDDATA;
99 if (avctx->colorspace == AVCOL_SPC_BT709)
100 avctx->codec_tag = MKTAG('U', 'L', 'H', '2');
102 avctx->codec_tag = MKTAG('U', 'L', 'Y', '2');
103 original_format = UTVIDEO_422;
106 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
108 return AVERROR_INVALIDDATA;
111 ff_dsputil_init(&c->dsp, avctx);
113 /* Check the prediction method, and error out if unsupported */
114 if (avctx->prediction_method < 0 || avctx->prediction_method > 4) {
115 av_log(avctx, AV_LOG_WARNING,
116 "Prediction method %d is not supported in Ut Video.\n",
117 avctx->prediction_method);
118 return AVERROR_OPTION_NOT_FOUND;
121 if (avctx->prediction_method == FF_PRED_PLANE) {
122 av_log(avctx, AV_LOG_ERROR,
123 "Plane prediction is not supported in Ut Video.\n");
124 return AVERROR_OPTION_NOT_FOUND;
127 /* Convert from libavcodec prediction type to Ut Video's */
128 c->frame_pred = ff_ut_pred_order[avctx->prediction_method];
130 if (c->frame_pred == PRED_GRADIENT) {
131 av_log(avctx, AV_LOG_ERROR, "Gradient prediction is not supported.\n");
132 return AVERROR_OPTION_NOT_FOUND;
135 avctx->coded_frame = av_frame_alloc();
137 if (!avctx->coded_frame) {
138 av_log(avctx, AV_LOG_ERROR, "Could not allocate frame.\n");
139 utvideo_encode_close(avctx);
140 return AVERROR(ENOMEM);
143 /* extradata size is 4 * 32bit */
144 avctx->extradata_size = 16;
146 avctx->extradata = av_mallocz(avctx->extradata_size +
147 FF_INPUT_BUFFER_PADDING_SIZE);
149 if (!avctx->extradata) {
150 av_log(avctx, AV_LOG_ERROR, "Could not allocate extradata.\n");
151 utvideo_encode_close(avctx);
152 return AVERROR(ENOMEM);
155 for (i = 0; i < c->planes; i++) {
156 c->slice_buffer[i] = av_malloc(c->slice_stride * (avctx->height + 2) +
157 FF_INPUT_BUFFER_PADDING_SIZE);
158 if (!c->slice_buffer[i]) {
159 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 1.\n");
160 utvideo_encode_close(avctx);
161 return AVERROR(ENOMEM);
166 * Set the version of the encoder.
167 * Last byte is "implementation ID", which is
168 * obtained from the creator of the format.
169 * Libavcodec has been assigned with the ID 0xF0.
171 AV_WB32(avctx->extradata, MKTAG(1, 0, 0, 0xF0));
174 * Set the "original format"
175 * Not used for anything during decoding.
177 AV_WL32(avctx->extradata + 4, original_format);
179 /* Write 4 as the 'frame info size' */
180 AV_WL32(avctx->extradata + 8, c->frame_info_size);
183 * Set how many slices are going to be used.
184 * Set one slice for now.
188 /* Set compression mode */
189 c->compression = COMP_HUFF;
192 * Set the encoding flags:
193 * - Slice count minus 1
194 * - Interlaced encoding mode flag, set to zero for now.
195 * - Compression mode (none/huff)
196 * And write the flags.
198 c->flags = (c->slices - 1) << 24;
199 c->flags |= 0 << 11; // bit field to signal interlaced encoding mode
200 c->flags |= c->compression;
202 AV_WL32(avctx->extradata + 12, c->flags);
207 static void mangle_rgb_planes(uint8_t *dst[4], int dst_stride, uint8_t *src,
208 int step, int stride, int width, int height)
211 int k = 2 * dst_stride;
214 for (j = 0; j < height; j++) {
216 for (i = 0; i < width * step; i += step) {
220 dst[1][k] = src[i + 2] - g;
221 dst[2][k] = src[i + 0] - g;
225 for (i = 0; i < width * step; i += step) {
229 dst[1][k] = src[i + 2] - g;
230 dst[2][k] = src[i + 0] - g;
231 dst[3][k] = src[i + 3];
235 k += dst_stride - width;
240 /* Write data to a plane with left prediction */
241 static void left_predict(uint8_t *src, uint8_t *dst, int stride,
242 int width, int height)
247 prev = 0x80; /* Set the initial value */
248 for (j = 0; j < height; j++) {
249 for (i = 0; i < width; i++) {
250 *dst++ = src[i] - prev;
257 /* Write data to a plane with median prediction */
258 static void median_predict(UtvideoContext *c, uint8_t *src, uint8_t *dst, int stride,
259 int width, int height)
265 /* First line uses left neighbour prediction */
266 prev = 0x80; /* Set the initial value */
267 for (i = 0; i < width; i++) {
268 *dst++ = src[i] - prev;
278 * Second line uses top prediction for the first sample,
279 * and median for the rest.
283 /* Rest of the coded part uses median prediction */
284 for (j = 1; j < height; j++) {
285 c->dsp.sub_hfyu_median_prediction(dst, src - stride, src, width, &A, &B);
291 /* Count the usage of values in a plane */
292 static void count_usage(uint8_t *src, int width,
293 int height, uint64_t *counts)
297 for (j = 0; j < height; j++) {
298 for (i = 0; i < width; i++) {
305 /* Calculate the actual huffman codes from the code lengths */
306 static void calculate_codes(HuffEntry *he)
311 qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len);
314 while (he[last].len == 255 && last)
318 for (i = last; i >= 0; i--) {
319 he[i].code = code >> (32 - he[i].len);
320 code += 0x80000000u >> (he[i].len - 1);
323 qsort(he, 256, sizeof(*he), huff_cmp_sym);
326 /* Write huffman bit codes to a memory block */
327 static int write_huff_codes(uint8_t *src, uint8_t *dst, int dst_size,
328 int width, int height, HuffEntry *he)
334 init_put_bits(&pb, dst, dst_size);
336 /* Write the codes */
337 for (j = 0; j < height; j++) {
338 for (i = 0; i < width; i++)
339 put_bits(&pb, he[src[i]].len, he[src[i]].code);
344 /* Pad output to a 32bit boundary */
345 count = put_bits_count(&pb) & 0x1F;
348 put_bits(&pb, 32 - count, 0);
350 /* Get the amount of bits written */
351 count = put_bits_count(&pb);
353 /* Flush the rest with zeroes */
359 static int encode_plane(AVCodecContext *avctx, uint8_t *src,
360 uint8_t *dst, int stride,
361 int width, int height, PutByteContext *pb)
363 UtvideoContext *c = avctx->priv_data;
364 uint8_t lengths[256];
365 uint64_t counts[256] = { 0 };
369 uint32_t offset = 0, slice_len = 0;
370 int i, sstart, send = 0;
373 /* Do prediction / make planes */
374 switch (c->frame_pred) {
376 for (i = 0; i < c->slices; i++) {
378 send = height * (i + 1) / c->slices;
379 av_image_copy_plane(dst + sstart * width, width,
380 src + sstart * stride, stride,
381 width, send - sstart);
385 for (i = 0; i < c->slices; i++) {
387 send = height * (i + 1) / c->slices;
388 left_predict(src + sstart * stride, dst + sstart * width,
389 stride, width, send - sstart);
393 for (i = 0; i < c->slices; i++) {
395 send = height * (i + 1) / c->slices;
396 median_predict(c, src + sstart * stride, dst + sstart * width,
397 stride, width, send - sstart);
401 av_log(avctx, AV_LOG_ERROR, "Unknown prediction mode: %d\n",
403 return AVERROR_OPTION_NOT_FOUND;
406 /* Count the usage of values */
407 count_usage(dst, width, height, counts);
409 /* Check for a special case where only one symbol was used */
410 for (symbol = 0; symbol < 256; symbol++) {
411 /* If non-zero count is found, see if it matches width * height */
412 if (counts[symbol]) {
413 /* Special case if only one symbol was used */
414 if (counts[symbol] == width * height) {
416 * Write a zero for the single symbol
417 * used in the plane, else 0xFF.
419 for (i = 0; i < 256; i++) {
421 bytestream2_put_byte(pb, 0);
423 bytestream2_put_byte(pb, 0xFF);
426 /* Write zeroes for lengths */
427 for (i = 0; i < c->slices; i++)
428 bytestream2_put_le32(pb, 0);
430 /* And that's all for that plane folks */
437 /* Calculate huffman lengths */
438 ff_huff_gen_len_table(lengths, counts);
441 * Write the plane's header into the output packet:
442 * - huffman code lengths (256 bytes)
443 * - slice end offsets (gotten from the slice lengths)
445 for (i = 0; i < 256; i++) {
446 bytestream2_put_byte(pb, lengths[i]);
448 he[i].len = lengths[i];
452 /* Calculate the huffman codes themselves */
456 for (i = 0; i < c->slices; i++) {
458 send = height * (i + 1) / c->slices;
461 * Write the huffman codes to a buffer,
462 * get the offset in bits and convert to bytes.
464 offset += write_huff_codes(dst + sstart * width, c->slice_bits,
465 width * (send - sstart), width,
466 send - sstart, he) >> 3;
468 slice_len = offset - slice_len;
470 /* Byteswap the written huffman codes */
471 c->dsp.bswap_buf((uint32_t *) c->slice_bits,
472 (uint32_t *) c->slice_bits,
475 /* Write the offset to the stream */
476 bytestream2_put_le32(pb, offset);
478 /* Seek to the data part of the packet */
479 bytestream2_seek_p(pb, 4 * (c->slices - i - 1) +
480 offset - slice_len, SEEK_CUR);
482 /* Write the slices' data into the output packet */
483 bytestream2_put_buffer(pb, c->slice_bits, slice_len);
485 /* Seek back to the slice offsets */
486 bytestream2_seek_p(pb, -4 * (c->slices - i - 1) - offset,
492 /* And at the end seek to the end of written slice(s) */
493 bytestream2_seek_p(pb, offset, SEEK_CUR);
498 static int utvideo_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
499 const AVFrame *pic, int *got_packet)
501 UtvideoContext *c = avctx->priv_data;
508 int width = avctx->width, height = avctx->height;
511 /* Allocate a new packet if needed, and set it to the pointer dst */
512 ret = ff_alloc_packet(pkt, (256 + 4 * c->slices + width * height) *
516 av_log(avctx, AV_LOG_ERROR,
517 "Error allocating the output packet, or the provided packet "
524 bytestream2_init_writer(&pb, dst, pkt->size);
526 av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
527 width * height + FF_INPUT_BUFFER_PADDING_SIZE);
529 if (!c->slice_bits) {
530 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 2.\n");
531 return AVERROR(ENOMEM);
534 /* In case of RGB, mangle the planes to Ut Video's format */
535 if (avctx->pix_fmt == AV_PIX_FMT_RGBA || avctx->pix_fmt == AV_PIX_FMT_RGB24)
536 mangle_rgb_planes(c->slice_buffer, c->slice_stride, pic->data[0],
537 c->planes, pic->linesize[0], width, height);
539 /* Deal with the planes */
540 switch (avctx->pix_fmt) {
541 case AV_PIX_FMT_RGB24:
542 case AV_PIX_FMT_RGBA:
543 for (i = 0; i < c->planes; i++) {
544 ret = encode_plane(avctx, c->slice_buffer[i] + 2 * c->slice_stride,
545 c->slice_buffer[i], c->slice_stride,
549 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
554 case AV_PIX_FMT_YUV422P:
555 for (i = 0; i < c->planes; i++) {
556 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
557 pic->linesize[i], width >> !!i, height, &pb);
560 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
565 case AV_PIX_FMT_YUV420P:
566 for (i = 0; i < c->planes; i++) {
567 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
568 pic->linesize[i], width >> !!i, height >> !!i,
572 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
578 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
580 return AVERROR_INVALIDDATA;
584 * Write frame information (LE 32bit unsigned)
585 * into the output packet.
586 * Contains the prediction method.
588 frame_info = c->frame_pred << 8;
589 bytestream2_put_le32(&pb, frame_info);
592 * At least currently Ut Video is IDR only.
593 * Set flags accordingly.
595 avctx->coded_frame->key_frame = 1;
596 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
598 pkt->size = bytestream2_tell_p(&pb);
599 pkt->flags |= AV_PKT_FLAG_KEY;
601 /* Packet should be done */
607 AVCodec ff_utvideo_encoder = {
609 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
610 .type = AVMEDIA_TYPE_VIDEO,
611 .id = AV_CODEC_ID_UTVIDEO,
612 .priv_data_size = sizeof(UtvideoContext),
613 .init = utvideo_encode_init,
614 .encode2 = utvideo_encode_frame,
615 .close = utvideo_encode_close,
616 .pix_fmts = (const enum AVPixelFormat[]) {
617 AV_PIX_FMT_RGB24, AV_PIX_FMT_RGBA, AV_PIX_FMT_YUV422P,
618 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE