3 * Copyright (c) 2012 Jan Ekström
5 * This file is part of FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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"
29 #include "libavutil/opt.h"
34 #include "bytestream.h"
36 #include "huffyuvencdsp.h"
41 /* Compare huffentry symbols */
42 static int huff_cmp_sym(const void *a, const void *b)
44 const HuffEntry *aa = a, *bb = b;
45 return aa->sym - bb->sym;
48 static av_cold int utvideo_encode_close(AVCodecContext *avctx)
50 UtvideoContext *c = avctx->priv_data;
53 av_freep(&c->slice_bits);
54 for (i = 0; i < 4; i++)
55 av_freep(&c->slice_buffer[i]);
60 static av_cold int utvideo_encode_init(AVCodecContext *avctx)
62 UtvideoContext *c = avctx->priv_data;
63 int i, subsampled_height;
64 uint32_t original_format;
67 c->frame_info_size = 4;
68 c->slice_stride = FFALIGN(avctx->width, 32);
70 switch (avctx->pix_fmt) {
71 case AV_PIX_FMT_RGB24:
73 avctx->codec_tag = MKTAG('U', 'L', 'R', 'G');
74 original_format = UTVIDEO_RGB;
78 avctx->codec_tag = MKTAG('U', 'L', 'R', 'A');
79 original_format = UTVIDEO_RGBA;
80 avctx->bits_per_coded_sample = 32;
82 case AV_PIX_FMT_YUV420P:
83 if (avctx->width & 1 || avctx->height & 1) {
84 av_log(avctx, AV_LOG_ERROR,
85 "4:2:0 video requires even width and height.\n");
86 return AVERROR_INVALIDDATA;
89 if (avctx->colorspace == AVCOL_SPC_BT709)
90 avctx->codec_tag = MKTAG('U', 'L', 'H', '0');
92 avctx->codec_tag = MKTAG('U', 'L', 'Y', '0');
93 original_format = UTVIDEO_420;
95 case AV_PIX_FMT_YUV422P:
96 if (avctx->width & 1) {
97 av_log(avctx, AV_LOG_ERROR,
98 "4:2:2 video requires even width.\n");
99 return AVERROR_INVALIDDATA;
102 if (avctx->colorspace == AVCOL_SPC_BT709)
103 avctx->codec_tag = MKTAG('U', 'L', 'H', '2');
105 avctx->codec_tag = MKTAG('U', 'L', 'Y', '2');
106 original_format = UTVIDEO_422;
108 case AV_PIX_FMT_YUV444P:
110 if (avctx->colorspace == AVCOL_SPC_BT709)
111 avctx->codec_tag = MKTAG('U', 'L', 'H', '4');
113 avctx->codec_tag = MKTAG('U', 'L', 'Y', '4');
114 original_format = UTVIDEO_444;
117 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
119 return AVERROR_INVALIDDATA;
122 ff_bswapdsp_init(&c->bdsp);
123 ff_huffyuvencdsp_init(&c->hdsp);
125 #if FF_API_PRIVATE_OPT
126 FF_DISABLE_DEPRECATION_WARNINGS
127 /* Check the prediction method, and error out if unsupported */
128 if (avctx->prediction_method < 0 || avctx->prediction_method > 4) {
129 av_log(avctx, AV_LOG_WARNING,
130 "Prediction method %d is not supported in Ut Video.\n",
131 avctx->prediction_method);
132 return AVERROR_OPTION_NOT_FOUND;
135 if (avctx->prediction_method == FF_PRED_PLANE) {
136 av_log(avctx, AV_LOG_ERROR,
137 "Plane prediction is not supported in Ut Video.\n");
138 return AVERROR_OPTION_NOT_FOUND;
141 /* Convert from libavcodec prediction type to Ut Video's */
142 if (avctx->prediction_method)
143 c->frame_pred = ff_ut_pred_order[avctx->prediction_method];
144 FF_ENABLE_DEPRECATION_WARNINGS
147 if (c->frame_pred == PRED_GRADIENT) {
148 av_log(avctx, AV_LOG_ERROR, "Gradient prediction is not supported.\n");
149 return AVERROR_OPTION_NOT_FOUND;
153 * Check the asked slice count for obviously invalid
154 * values (> 256 or negative).
156 if (avctx->slices > 256 || avctx->slices < 0) {
157 av_log(avctx, AV_LOG_ERROR,
158 "Slice count %d is not supported in Ut Video (theoretical range is 0-256).\n",
160 return AVERROR(EINVAL);
163 /* Check that the slice count is not larger than the subsampled height */
164 subsampled_height = avctx->height >> av_pix_fmt_desc_get(avctx->pix_fmt)->log2_chroma_h;
165 if (avctx->slices > subsampled_height) {
166 av_log(avctx, AV_LOG_ERROR,
167 "Slice count %d is larger than the subsampling-applied height %d.\n",
168 avctx->slices, subsampled_height);
169 return AVERROR(EINVAL);
172 /* extradata size is 4 * 32 bits */
173 avctx->extradata_size = 16;
175 avctx->extradata = av_mallocz(avctx->extradata_size +
176 AV_INPUT_BUFFER_PADDING_SIZE);
178 if (!avctx->extradata) {
179 av_log(avctx, AV_LOG_ERROR, "Could not allocate extradata.\n");
180 utvideo_encode_close(avctx);
181 return AVERROR(ENOMEM);
184 for (i = 0; i < c->planes; i++) {
185 c->slice_buffer[i] = av_malloc(c->slice_stride * (avctx->height + 2) +
186 AV_INPUT_BUFFER_PADDING_SIZE);
187 if (!c->slice_buffer[i]) {
188 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 1.\n");
189 utvideo_encode_close(avctx);
190 return AVERROR(ENOMEM);
195 * Set the version of the encoder.
196 * Last byte is "implementation ID", which is
197 * obtained from the creator of the format.
198 * Libavcodec has been assigned with the ID 0xF0.
200 AV_WB32(avctx->extradata, MKTAG(1, 0, 0, 0xF0));
203 * Set the "original format"
204 * Not used for anything during decoding.
206 AV_WL32(avctx->extradata + 4, original_format);
208 /* Write 4 as the 'frame info size' */
209 AV_WL32(avctx->extradata + 8, c->frame_info_size);
212 * Set how many slices are going to be used.
213 * By default uses multiple slices depending on the subsampled height.
214 * This enables multithreading in the official decoder.
216 if (!avctx->slices) {
217 c->slices = subsampled_height / 120;
221 else if (c->slices > 256)
224 c->slices = avctx->slices;
227 /* Set compression mode */
228 c->compression = COMP_HUFF;
231 * Set the encoding flags:
232 * - Slice count minus 1
233 * - Interlaced encoding mode flag, set to zero for now.
234 * - Compression mode (none/huff)
235 * And write the flags.
237 c->flags = (c->slices - 1) << 24;
238 c->flags |= 0 << 11; // bit field to signal interlaced encoding mode
239 c->flags |= c->compression;
241 AV_WL32(avctx->extradata + 12, c->flags);
246 static void mangle_rgb_planes(uint8_t *dst[4], int dst_stride, uint8_t *src,
247 int step, int stride, int width, int height)
250 int k = 2 * dst_stride;
253 for (j = 0; j < height; j++) {
255 for (i = 0; i < width * step; i += step) {
259 dst[1][k] = src[i + 2] - g;
260 dst[2][k] = src[i + 0] - g;
264 for (i = 0; i < width * step; i += step) {
268 dst[1][k] = src[i + 2] - g;
269 dst[2][k] = src[i + 0] - g;
270 dst[3][k] = src[i + 3];
274 k += dst_stride - width;
279 /* Write data to a plane with left prediction */
280 static void left_predict(uint8_t *src, uint8_t *dst, int stride,
281 int width, int height)
286 prev = 0x80; /* Set the initial value */
287 for (j = 0; j < height; j++) {
288 for (i = 0; i < width; i++) {
289 *dst++ = src[i] - prev;
296 /* Write data to a plane with median prediction */
297 static void median_predict(UtvideoContext *c, uint8_t *src, uint8_t *dst, int stride,
298 int width, int height)
304 /* First line uses left neighbour prediction */
305 prev = 0x80; /* Set the initial value */
306 for (i = 0; i < width; i++) {
307 *dst++ = src[i] - prev;
317 * Second line uses top prediction for the first sample,
318 * and median for the rest.
322 /* Rest of the coded part uses median prediction */
323 for (j = 1; j < height; j++) {
324 c->hdsp.sub_hfyu_median_pred(dst, src - stride, src, width, &A, &B);
330 /* Count the usage of values in a plane */
331 static void count_usage(uint8_t *src, int width,
332 int height, uint64_t *counts)
336 for (j = 0; j < height; j++) {
337 for (i = 0; i < width; i++) {
344 /* Calculate the actual huffman codes from the code lengths */
345 static void calculate_codes(HuffEntry *he)
350 qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len);
353 while (he[last].len == 255 && last)
357 for (i = last; i >= 0; i--) {
358 he[i].code = code >> (32 - he[i].len);
359 code += 0x80000000u >> (he[i].len - 1);
362 qsort(he, 256, sizeof(*he), huff_cmp_sym);
365 /* Write huffman bit codes to a memory block */
366 static int write_huff_codes(uint8_t *src, uint8_t *dst, int dst_size,
367 int width, int height, HuffEntry *he)
373 init_put_bits(&pb, dst, dst_size);
375 /* Write the codes */
376 for (j = 0; j < height; j++) {
377 for (i = 0; i < width; i++)
378 put_bits(&pb, he[src[i]].len, he[src[i]].code);
383 /* Pad output to a 32-bit boundary */
384 count = put_bits_count(&pb) & 0x1F;
387 put_bits(&pb, 32 - count, 0);
389 /* Get the amount of bits written */
390 count = put_bits_count(&pb);
392 /* Flush the rest with zeroes */
398 static int encode_plane(AVCodecContext *avctx, uint8_t *src,
399 uint8_t *dst, int stride, int plane_no,
400 int width, int height, PutByteContext *pb)
402 UtvideoContext *c = avctx->priv_data;
403 uint8_t lengths[256];
404 uint64_t counts[256] = { 0 };
408 uint32_t offset = 0, slice_len = 0;
409 const int cmask = ~(!plane_no && avctx->pix_fmt == AV_PIX_FMT_YUV420P);
410 int i, sstart, send = 0;
414 /* Do prediction / make planes */
415 switch (c->frame_pred) {
417 for (i = 0; i < c->slices; i++) {
419 send = height * (i + 1) / c->slices & cmask;
420 av_image_copy_plane(dst + sstart * width, width,
421 src + sstart * stride, stride,
422 width, send - sstart);
426 for (i = 0; i < c->slices; i++) {
428 send = height * (i + 1) / c->slices & cmask;
429 left_predict(src + sstart * stride, dst + sstart * width,
430 stride, width, send - sstart);
434 for (i = 0; i < c->slices; i++) {
436 send = height * (i + 1) / c->slices & cmask;
437 median_predict(c, src + sstart * stride, dst + sstart * width,
438 stride, width, send - sstart);
442 av_log(avctx, AV_LOG_ERROR, "Unknown prediction mode: %d\n",
444 return AVERROR_OPTION_NOT_FOUND;
447 /* Count the usage of values */
448 count_usage(dst, width, height, counts);
450 /* Check for a special case where only one symbol was used */
451 for (symbol = 0; symbol < 256; symbol++) {
452 /* If non-zero count is found, see if it matches width * height */
453 if (counts[symbol]) {
454 /* Special case if only one symbol was used */
455 if (counts[symbol] == width * (int64_t)height) {
457 * Write a zero for the single symbol
458 * used in the plane, else 0xFF.
460 for (i = 0; i < 256; i++) {
462 bytestream2_put_byte(pb, 0);
464 bytestream2_put_byte(pb, 0xFF);
467 /* Write zeroes for lengths */
468 for (i = 0; i < c->slices; i++)
469 bytestream2_put_le32(pb, 0);
471 /* And that's all for that plane folks */
478 /* Calculate huffman lengths */
479 if ((ret = ff_huff_gen_len_table(lengths, counts, 256, 1)) < 0)
483 * Write the plane's header into the output packet:
484 * - huffman code lengths (256 bytes)
485 * - slice end offsets (gotten from the slice lengths)
487 for (i = 0; i < 256; i++) {
488 bytestream2_put_byte(pb, lengths[i]);
490 he[i].len = lengths[i];
494 /* Calculate the huffman codes themselves */
498 for (i = 0; i < c->slices; i++) {
500 send = height * (i + 1) / c->slices & cmask;
503 * Write the huffman codes to a buffer,
504 * get the offset in bits and convert to bytes.
506 offset += write_huff_codes(dst + sstart * width, c->slice_bits,
507 width * height + 4, width,
508 send - sstart, he) >> 3;
510 slice_len = offset - slice_len;
512 /* Byteswap the written huffman codes */
513 c->bdsp.bswap_buf((uint32_t *) c->slice_bits,
514 (uint32_t *) c->slice_bits,
517 /* Write the offset to the stream */
518 bytestream2_put_le32(pb, offset);
520 /* Seek to the data part of the packet */
521 bytestream2_seek_p(pb, 4 * (c->slices - i - 1) +
522 offset - slice_len, SEEK_CUR);
524 /* Write the slices' data into the output packet */
525 bytestream2_put_buffer(pb, c->slice_bits, slice_len);
527 /* Seek back to the slice offsets */
528 bytestream2_seek_p(pb, -4 * (c->slices - i - 1) - offset,
534 /* And at the end seek to the end of written slice(s) */
535 bytestream2_seek_p(pb, offset, SEEK_CUR);
540 static int utvideo_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
541 const AVFrame *pic, int *got_packet)
543 UtvideoContext *c = avctx->priv_data;
550 int width = avctx->width, height = avctx->height;
553 /* Allocate a new packet if needed, and set it to the pointer dst */
554 ret = ff_alloc_packet2(avctx, pkt, (256 + 4 * c->slices + width * height) *
562 bytestream2_init_writer(&pb, dst, pkt->size);
564 av_fast_padded_malloc(&c->slice_bits, &c->slice_bits_size, width * height + 4);
566 if (!c->slice_bits) {
567 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 2.\n");
568 return AVERROR(ENOMEM);
571 /* In case of RGB, mangle the planes to Ut Video's format */
572 if (avctx->pix_fmt == AV_PIX_FMT_RGBA || avctx->pix_fmt == AV_PIX_FMT_RGB24)
573 mangle_rgb_planes(c->slice_buffer, c->slice_stride, pic->data[0],
574 c->planes, pic->linesize[0], width, height);
576 /* Deal with the planes */
577 switch (avctx->pix_fmt) {
578 case AV_PIX_FMT_RGB24:
579 case AV_PIX_FMT_RGBA:
580 for (i = 0; i < c->planes; i++) {
581 ret = encode_plane(avctx, c->slice_buffer[i] + 2 * c->slice_stride,
582 c->slice_buffer[i], c->slice_stride, i,
586 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
591 case AV_PIX_FMT_YUV444P:
592 for (i = 0; i < c->planes; i++) {
593 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
594 pic->linesize[i], i, width, height, &pb);
597 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
602 case AV_PIX_FMT_YUV422P:
603 for (i = 0; i < c->planes; i++) {
604 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
605 pic->linesize[i], i, width >> !!i, height, &pb);
608 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
613 case AV_PIX_FMT_YUV420P:
614 for (i = 0; i < c->planes; i++) {
615 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
616 pic->linesize[i], i, width >> !!i, height >> !!i,
620 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
626 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
628 return AVERROR_INVALIDDATA;
632 * Write frame information (LE 32-bit unsigned)
633 * into the output packet.
634 * Contains the prediction method.
636 frame_info = c->frame_pred << 8;
637 bytestream2_put_le32(&pb, frame_info);
640 * At least currently Ut Video is IDR only.
641 * Set flags accordingly.
643 #if FF_API_CODED_FRAME
644 FF_DISABLE_DEPRECATION_WARNINGS
645 avctx->coded_frame->key_frame = 1;
646 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
647 FF_ENABLE_DEPRECATION_WARNINGS
650 pkt->size = bytestream2_tell_p(&pb);
651 pkt->flags |= AV_PKT_FLAG_KEY;
653 /* Packet should be done */
659 #define OFFSET(x) offsetof(UtvideoContext, x)
660 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
661 static const AVOption options[] = {
662 { "pred", "Prediction method", OFFSET(frame_pred), AV_OPT_TYPE_INT, { .i64 = PRED_LEFT }, PRED_NONE, PRED_MEDIAN, VE, "pred" },
663 { "none", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_NONE }, INT_MIN, INT_MAX, VE, "pred" },
664 { "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_LEFT }, INT_MIN, INT_MAX, VE, "pred" },
665 { "gradient", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_GRADIENT }, INT_MIN, INT_MAX, VE, "pred" },
666 { "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_MEDIAN }, INT_MIN, INT_MAX, VE, "pred" },
671 static const AVClass utvideo_class = {
672 .class_name = "utvideo",
673 .item_name = av_default_item_name,
675 .version = LIBAVUTIL_VERSION_INT,
678 AVCodec ff_utvideo_encoder = {
680 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
681 .type = AVMEDIA_TYPE_VIDEO,
682 .id = AV_CODEC_ID_UTVIDEO,
683 .priv_data_size = sizeof(UtvideoContext),
684 .priv_class = &utvideo_class,
685 .init = utvideo_encode_init,
686 .encode2 = utvideo_encode_frame,
687 .close = utvideo_encode_close,
688 .capabilities = AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_INTRA_ONLY,
689 .pix_fmts = (const enum AVPixelFormat[]) {
690 AV_PIX_FMT_RGB24, AV_PIX_FMT_RGBA, AV_PIX_FMT_YUV422P,
691 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE