1/* 2 * Ut Video encoder 3 * Copyright (c) 2012 Jan Ekström 4 * 5 * This file is part of FFmpeg. 6 * 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. 11 * 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. 16 * 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 20 */ 21 22/** 23 * @file 24 * Ut Video encoder 25 */ 26 27#include "libavutil/imgutils.h" 28#include "libavutil/intreadwrite.h" 29#include "libavutil/opt.h" 30 31#include "avcodec.h" 32#include "codec_internal.h" 33#include "encode.h" 34#include "bswapdsp.h" 35#include "bytestream.h" 36#include "put_bits.h" 37#include "mathops.h" 38#include "utvideo.h" 39#include "huffman.h" 40 41typedef struct HuffEntry { 42 uint16_t sym; 43 uint8_t len; 44 uint32_t code; 45} HuffEntry; 46 47/* Compare huffman tree nodes */ 48static int ut_huff_cmp_len(const void *a, const void *b) 49{ 50 const HuffEntry *aa = a, *bb = b; 51 return (aa->len - bb->len)*256 + aa->sym - bb->sym; 52} 53 54/* Compare huffentry symbols */ 55static int huff_cmp_sym(const void *a, const void *b) 56{ 57 const HuffEntry *aa = a, *bb = b; 58 return aa->sym - bb->sym; 59} 60 61static av_cold int utvideo_encode_close(AVCodecContext *avctx) 62{ 63 UtvideoContext *c = avctx->priv_data; 64 int i; 65 66 av_freep(&c->slice_bits); 67 for (i = 0; i < 4; i++) 68 av_freep(&c->slice_buffer[i]); 69 70 return 0; 71} 72 73static av_cold int utvideo_encode_init(AVCodecContext *avctx) 74{ 75 UtvideoContext *c = avctx->priv_data; 76 int i, subsampled_height; 77 uint32_t original_format; 78 79 c->avctx = avctx; 80 c->frame_info_size = 4; 81 c->slice_stride = FFALIGN(avctx->width, 32); 82 83 switch (avctx->pix_fmt) { 84 case AV_PIX_FMT_GBRP: 85 c->planes = 3; 86 avctx->codec_tag = MKTAG('U', 'L', 'R', 'G'); 87 original_format = UTVIDEO_RGB; 88 break; 89 case AV_PIX_FMT_GBRAP: 90 c->planes = 4; 91 avctx->codec_tag = MKTAG('U', 'L', 'R', 'A'); 92 original_format = UTVIDEO_RGBA; 93 avctx->bits_per_coded_sample = 32; 94 break; 95 case AV_PIX_FMT_YUV420P: 96 if (avctx->width & 1 || avctx->height & 1) { 97 av_log(avctx, AV_LOG_ERROR, 98 "4:2:0 video requires even width and height.\n"); 99 return AVERROR_INVALIDDATA; 100 } 101 c->planes = 3; 102 if (avctx->colorspace == AVCOL_SPC_BT709) 103 avctx->codec_tag = MKTAG('U', 'L', 'H', '0'); 104 else 105 avctx->codec_tag = MKTAG('U', 'L', 'Y', '0'); 106 original_format = UTVIDEO_420; 107 break; 108 case AV_PIX_FMT_YUV422P: 109 if (avctx->width & 1) { 110 av_log(avctx, AV_LOG_ERROR, 111 "4:2:2 video requires even width.\n"); 112 return AVERROR_INVALIDDATA; 113 } 114 c->planes = 3; 115 if (avctx->colorspace == AVCOL_SPC_BT709) 116 avctx->codec_tag = MKTAG('U', 'L', 'H', '2'); 117 else 118 avctx->codec_tag = MKTAG('U', 'L', 'Y', '2'); 119 original_format = UTVIDEO_422; 120 break; 121 case AV_PIX_FMT_YUV444P: 122 c->planes = 3; 123 if (avctx->colorspace == AVCOL_SPC_BT709) 124 avctx->codec_tag = MKTAG('U', 'L', 'H', '4'); 125 else 126 avctx->codec_tag = MKTAG('U', 'L', 'Y', '4'); 127 original_format = UTVIDEO_444; 128 break; 129 default: 130 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n", 131 avctx->pix_fmt); 132 return AVERROR_INVALIDDATA; 133 } 134 135 ff_bswapdsp_init(&c->bdsp); 136 ff_llvidencdsp_init(&c->llvidencdsp); 137 138 if (c->frame_pred == PRED_GRADIENT) { 139 av_log(avctx, AV_LOG_ERROR, "Gradient prediction is not supported.\n"); 140 return AVERROR_OPTION_NOT_FOUND; 141 } 142 143 /* 144 * Check the asked slice count for obviously invalid 145 * values (> 256 or negative). 146 */ 147 if (avctx->slices > 256 || avctx->slices < 0) { 148 av_log(avctx, AV_LOG_ERROR, 149 "Slice count %d is not supported in Ut Video (theoretical range is 0-256).\n", 150 avctx->slices); 151 return AVERROR(EINVAL); 152 } 153 154 /* Check that the slice count is not larger than the subsampled height */ 155 subsampled_height = avctx->height >> av_pix_fmt_desc_get(avctx->pix_fmt)->log2_chroma_h; 156 if (avctx->slices > subsampled_height) { 157 av_log(avctx, AV_LOG_ERROR, 158 "Slice count %d is larger than the subsampling-applied height %d.\n", 159 avctx->slices, subsampled_height); 160 return AVERROR(EINVAL); 161 } 162 163 /* extradata size is 4 * 32 bits */ 164 avctx->extradata_size = 16; 165 166 avctx->extradata = av_mallocz(avctx->extradata_size + 167 AV_INPUT_BUFFER_PADDING_SIZE); 168 169 if (!avctx->extradata) { 170 av_log(avctx, AV_LOG_ERROR, "Could not allocate extradata.\n"); 171 return AVERROR(ENOMEM); 172 } 173 174 for (i = 0; i < c->planes; i++) { 175 c->slice_buffer[i] = av_malloc(c->slice_stride * (avctx->height + 2) + 176 AV_INPUT_BUFFER_PADDING_SIZE); 177 if (!c->slice_buffer[i]) { 178 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 1.\n"); 179 return AVERROR(ENOMEM); 180 } 181 } 182 183 /* 184 * Set the version of the encoder. 185 * Last byte is "implementation ID", which is 186 * obtained from the creator of the format. 187 * Libavcodec has been assigned with the ID 0xF0. 188 */ 189 AV_WB32(avctx->extradata, MKTAG(1, 0, 0, 0xF0)); 190 191 /* 192 * Set the "original format" 193 * Not used for anything during decoding. 194 */ 195 AV_WL32(avctx->extradata + 4, original_format); 196 197 /* Write 4 as the 'frame info size' */ 198 AV_WL32(avctx->extradata + 8, c->frame_info_size); 199 200 /* 201 * Set how many slices are going to be used. 202 * By default uses multiple slices depending on the subsampled height. 203 * This enables multithreading in the official decoder. 204 */ 205 if (!avctx->slices) { 206 c->slices = subsampled_height / 120; 207 208 if (!c->slices) 209 c->slices = 1; 210 else if (c->slices > 256) 211 c->slices = 256; 212 } else { 213 c->slices = avctx->slices; 214 } 215 216 /* Set compression mode */ 217 c->compression = COMP_HUFF; 218 219 /* 220 * Set the encoding flags: 221 * - Slice count minus 1 222 * - Interlaced encoding mode flag, set to zero for now. 223 * - Compression mode (none/huff) 224 * And write the flags. 225 */ 226 c->flags = (c->slices - 1) << 24; 227 c->flags |= 0 << 11; // bit field to signal interlaced encoding mode 228 c->flags |= c->compression; 229 230 AV_WL32(avctx->extradata + 12, c->flags); 231 232 return 0; 233} 234 235static void mangle_rgb_planes(uint8_t *dst[4], ptrdiff_t dst_stride, 236 uint8_t *const src[4], int planes, const int stride[4], 237 int width, int height) 238{ 239 int i, j; 240 int k = 2 * dst_stride; 241 const uint8_t *sg = src[0]; 242 const uint8_t *sb = src[1]; 243 const uint8_t *sr = src[2]; 244 const uint8_t *sa = src[3]; 245 unsigned int g; 246 247 for (j = 0; j < height; j++) { 248 if (planes == 3) { 249 for (i = 0; i < width; i++) { 250 g = sg[i]; 251 dst[0][k] = g; 252 g += 0x80; 253 dst[1][k] = sb[i] - g; 254 dst[2][k] = sr[i] - g; 255 k++; 256 } 257 } else { 258 for (i = 0; i < width; i++) { 259 g = sg[i]; 260 dst[0][k] = g; 261 g += 0x80; 262 dst[1][k] = sb[i] - g; 263 dst[2][k] = sr[i] - g; 264 dst[3][k] = sa[i]; 265 k++; 266 } 267 sa += stride[3]; 268 } 269 k += dst_stride - width; 270 sg += stride[0]; 271 sb += stride[1]; 272 sr += stride[2]; 273 } 274} 275 276#undef A 277#undef B 278 279/* Write data to a plane with median prediction */ 280static void median_predict(UtvideoContext *c, uint8_t *src, uint8_t *dst, 281 ptrdiff_t stride, int width, int height) 282{ 283 int i, j; 284 int A, B; 285 uint8_t prev; 286 287 /* First line uses left neighbour prediction */ 288 prev = 0x80; /* Set the initial value */ 289 for (i = 0; i < width; i++) { 290 *dst++ = src[i] - prev; 291 prev = src[i]; 292 } 293 294 if (height == 1) 295 return; 296 297 src += stride; 298 299 /* 300 * Second line uses top prediction for the first sample, 301 * and median for the rest. 302 */ 303 A = B = 0; 304 305 /* Rest of the coded part uses median prediction */ 306 for (j = 1; j < height; j++) { 307 c->llvidencdsp.sub_median_pred(dst, src - stride, src, width, &A, &B); 308 dst += width; 309 src += stride; 310 } 311} 312 313/* Count the usage of values in a plane */ 314static void count_usage(uint8_t *src, int width, 315 int height, uint64_t *counts) 316{ 317 int i, j; 318 319 for (j = 0; j < height; j++) { 320 for (i = 0; i < width; i++) { 321 counts[src[i]]++; 322 } 323 src += width; 324 } 325} 326 327/* Calculate the actual huffman codes from the code lengths */ 328static void calculate_codes(HuffEntry *he) 329{ 330 int last, i; 331 uint32_t code; 332 333 qsort(he, 256, sizeof(*he), ut_huff_cmp_len); 334 335 last = 255; 336 while (he[last].len == 255 && last) 337 last--; 338 339 code = 0; 340 for (i = last; i >= 0; i--) { 341 he[i].code = code >> (32 - he[i].len); 342 code += 0x80000000u >> (he[i].len - 1); 343 } 344 345 qsort(he, 256, sizeof(*he), huff_cmp_sym); 346} 347 348/* Write huffman bit codes to a memory block */ 349static int write_huff_codes(uint8_t *src, uint8_t *dst, int dst_size, 350 int width, int height, HuffEntry *he) 351{ 352 PutBitContext pb; 353 int i, j; 354 int count; 355 356 init_put_bits(&pb, dst, dst_size); 357 358 /* Write the codes */ 359 for (j = 0; j < height; j++) { 360 for (i = 0; i < width; i++) 361 put_bits(&pb, he[src[i]].len, he[src[i]].code); 362 363 src += width; 364 } 365 366 /* Pad output to a 32-bit boundary */ 367 count = put_bits_count(&pb) & 0x1F; 368 369 if (count) 370 put_bits(&pb, 32 - count, 0); 371 372 /* Flush the rest with zeroes */ 373 flush_put_bits(&pb); 374 375 /* Return the amount of bytes written */ 376 return put_bytes_output(&pb); 377} 378 379static int encode_plane(AVCodecContext *avctx, uint8_t *src, 380 uint8_t *dst, ptrdiff_t stride, int plane_no, 381 int width, int height, PutByteContext *pb) 382{ 383 UtvideoContext *c = avctx->priv_data; 384 uint8_t lengths[256]; 385 uint64_t counts[256] = { 0 }; 386 387 HuffEntry he[256]; 388 389 uint32_t offset = 0, slice_len = 0; 390 const int cmask = ~(!plane_no && avctx->pix_fmt == AV_PIX_FMT_YUV420P); 391 int i, sstart, send = 0; 392 int symbol; 393 int ret; 394 395 /* Do prediction / make planes */ 396 switch (c->frame_pred) { 397 case PRED_NONE: 398 for (i = 0; i < c->slices; i++) { 399 sstart = send; 400 send = height * (i + 1) / c->slices & cmask; 401 av_image_copy_plane(dst + sstart * width, width, 402 src + sstart * stride, stride, 403 width, send - sstart); 404 } 405 break; 406 case PRED_LEFT: 407 for (i = 0; i < c->slices; i++) { 408 sstart = send; 409 send = height * (i + 1) / c->slices & cmask; 410 c->llvidencdsp.sub_left_predict(dst + sstart * width, src + sstart * stride, stride, width, send - sstart); 411 } 412 break; 413 case PRED_MEDIAN: 414 for (i = 0; i < c->slices; i++) { 415 sstart = send; 416 send = height * (i + 1) / c->slices & cmask; 417 median_predict(c, src + sstart * stride, dst + sstart * width, 418 stride, width, send - sstart); 419 } 420 break; 421 default: 422 av_log(avctx, AV_LOG_ERROR, "Unknown prediction mode: %d\n", 423 c->frame_pred); 424 return AVERROR_OPTION_NOT_FOUND; 425 } 426 427 /* Count the usage of values */ 428 count_usage(dst, width, height, counts); 429 430 /* Check for a special case where only one symbol was used */ 431 for (symbol = 0; symbol < 256; symbol++) { 432 /* If non-zero count is found, see if it matches width * height */ 433 if (counts[symbol]) { 434 /* Special case if only one symbol was used */ 435 if (counts[symbol] == width * (int64_t)height) { 436 /* 437 * Write a zero for the single symbol 438 * used in the plane, else 0xFF. 439 */ 440 for (i = 0; i < 256; i++) { 441 if (i == symbol) 442 bytestream2_put_byte(pb, 0); 443 else 444 bytestream2_put_byte(pb, 0xFF); 445 } 446 447 /* Write zeroes for lengths */ 448 for (i = 0; i < c->slices; i++) 449 bytestream2_put_le32(pb, 0); 450 451 /* And that's all for that plane folks */ 452 return 0; 453 } 454 break; 455 } 456 } 457 458 /* Calculate huffman lengths */ 459 if ((ret = ff_huff_gen_len_table(lengths, counts, 256, 1)) < 0) 460 return ret; 461 462 /* 463 * Write the plane's header into the output packet: 464 * - huffman code lengths (256 bytes) 465 * - slice end offsets (gotten from the slice lengths) 466 */ 467 for (i = 0; i < 256; i++) { 468 bytestream2_put_byte(pb, lengths[i]); 469 470 he[i].len = lengths[i]; 471 he[i].sym = i; 472 } 473 474 /* Calculate the huffman codes themselves */ 475 calculate_codes(he); 476 477 send = 0; 478 for (i = 0; i < c->slices; i++) { 479 sstart = send; 480 send = height * (i + 1) / c->slices & cmask; 481 482 /* 483 * Write the huffman codes to a buffer, 484 * get the offset in bytes. 485 */ 486 offset += write_huff_codes(dst + sstart * width, c->slice_bits, 487 width * height + 4, width, 488 send - sstart, he); 489 490 slice_len = offset - slice_len; 491 492 /* Byteswap the written huffman codes */ 493 c->bdsp.bswap_buf((uint32_t *) c->slice_bits, 494 (uint32_t *) c->slice_bits, 495 slice_len >> 2); 496 497 /* Write the offset to the stream */ 498 bytestream2_put_le32(pb, offset); 499 500 /* Seek to the data part of the packet */ 501 bytestream2_seek_p(pb, 4 * (c->slices - i - 1) + 502 offset - slice_len, SEEK_CUR); 503 504 /* Write the slices' data into the output packet */ 505 bytestream2_put_buffer(pb, c->slice_bits, slice_len); 506 507 /* Seek back to the slice offsets */ 508 bytestream2_seek_p(pb, -4 * (c->slices - i - 1) - offset, 509 SEEK_CUR); 510 511 slice_len = offset; 512 } 513 514 /* And at the end seek to the end of written slice(s) */ 515 bytestream2_seek_p(pb, offset, SEEK_CUR); 516 517 return 0; 518} 519 520static int utvideo_encode_frame(AVCodecContext *avctx, AVPacket *pkt, 521 const AVFrame *pic, int *got_packet) 522{ 523 UtvideoContext *c = avctx->priv_data; 524 PutByteContext pb; 525 526 uint32_t frame_info; 527 528 uint8_t *dst; 529 530 int width = avctx->width, height = avctx->height; 531 int i, ret = 0; 532 533 /* Allocate a new packet if needed, and set it to the pointer dst */ 534 ret = ff_alloc_packet(avctx, pkt, (256 + 4 * c->slices + width * height) 535 * c->planes + 4); 536 537 if (ret < 0) 538 return ret; 539 540 dst = pkt->data; 541 542 bytestream2_init_writer(&pb, dst, pkt->size); 543 544 av_fast_padded_malloc(&c->slice_bits, &c->slice_bits_size, width * height + 4); 545 546 if (!c->slice_bits) { 547 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 2.\n"); 548 return AVERROR(ENOMEM); 549 } 550 551 /* In case of RGB, mangle the planes to Ut Video's format */ 552 if (avctx->pix_fmt == AV_PIX_FMT_GBRAP || avctx->pix_fmt == AV_PIX_FMT_GBRP) 553 mangle_rgb_planes(c->slice_buffer, c->slice_stride, pic->data, 554 c->planes, pic->linesize, width, height); 555 556 /* Deal with the planes */ 557 switch (avctx->pix_fmt) { 558 case AV_PIX_FMT_GBRP: 559 case AV_PIX_FMT_GBRAP: 560 for (i = 0; i < c->planes; i++) { 561 ret = encode_plane(avctx, c->slice_buffer[i] + 2 * c->slice_stride, 562 c->slice_buffer[i], c->slice_stride, i, 563 width, height, &pb); 564 565 if (ret) { 566 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 567 return ret; 568 } 569 } 570 break; 571 case AV_PIX_FMT_YUV444P: 572 for (i = 0; i < c->planes; i++) { 573 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0], 574 pic->linesize[i], i, width, height, &pb); 575 576 if (ret) { 577 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 578 return ret; 579 } 580 } 581 break; 582 case AV_PIX_FMT_YUV422P: 583 for (i = 0; i < c->planes; i++) { 584 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0], 585 pic->linesize[i], i, width >> !!i, height, &pb); 586 587 if (ret) { 588 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 589 return ret; 590 } 591 } 592 break; 593 case AV_PIX_FMT_YUV420P: 594 for (i = 0; i < c->planes; i++) { 595 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0], 596 pic->linesize[i], i, width >> !!i, height >> !!i, 597 &pb); 598 599 if (ret) { 600 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 601 return ret; 602 } 603 } 604 break; 605 default: 606 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n", 607 avctx->pix_fmt); 608 return AVERROR_INVALIDDATA; 609 } 610 611 /* 612 * Write frame information (LE 32-bit unsigned) 613 * into the output packet. 614 * Contains the prediction method. 615 */ 616 frame_info = c->frame_pred << 8; 617 bytestream2_put_le32(&pb, frame_info); 618 619 pkt->size = bytestream2_tell_p(&pb); 620 621 /* Packet should be done */ 622 *got_packet = 1; 623 624 return 0; 625} 626 627#define OFFSET(x) offsetof(UtvideoContext, x) 628#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM 629static const AVOption options[] = { 630{ "pred", "Prediction method", OFFSET(frame_pred), AV_OPT_TYPE_INT, { .i64 = PRED_LEFT }, PRED_NONE, PRED_MEDIAN, VE, "pred" }, 631 { "none", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_NONE }, INT_MIN, INT_MAX, VE, "pred" }, 632 { "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_LEFT }, INT_MIN, INT_MAX, VE, "pred" }, 633 { "gradient", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_GRADIENT }, INT_MIN, INT_MAX, VE, "pred" }, 634 { "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRED_MEDIAN }, INT_MIN, INT_MAX, VE, "pred" }, 635 636 { NULL}, 637}; 638 639static const AVClass utvideo_class = { 640 .class_name = "utvideo", 641 .item_name = av_default_item_name, 642 .option = options, 643 .version = LIBAVUTIL_VERSION_INT, 644}; 645 646const FFCodec ff_utvideo_encoder = { 647 .p.name = "utvideo", 648 .p.long_name = NULL_IF_CONFIG_SMALL("Ut Video"), 649 .p.type = AVMEDIA_TYPE_VIDEO, 650 .p.id = AV_CODEC_ID_UTVIDEO, 651 .priv_data_size = sizeof(UtvideoContext), 652 .p.priv_class = &utvideo_class, 653 .init = utvideo_encode_init, 654 FF_CODEC_ENCODE_CB(utvideo_encode_frame), 655 .close = utvideo_encode_close, 656 .p.capabilities = AV_CODEC_CAP_FRAME_THREADS, 657 .p.pix_fmts = (const enum AVPixelFormat[]) { 658 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP, AV_PIX_FMT_YUV422P, 659 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE 660 }, 661 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP, 662}; 663