1/* 2 * Ut Video encoder 3 * Copyright (c) 2012 Jan Ekstr��m 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 "avcodec.h" 30#include "internal.h" 31#include "bswapdsp.h" 32#include "bytestream.h" 33#include "put_bits.h" 34#include "huffyuvencdsp.h" 35#include "mathops.h" 36#include "utvideo.h" 37#include "huffman.h" 38 39/* Compare huffentry symbols */ 40static int huff_cmp_sym(const void *a, const void *b) 41{ 42 const HuffEntry *aa = a, *bb = b; 43 return aa->sym - bb->sym; 44} 45 46static av_cold int utvideo_encode_close(AVCodecContext *avctx) 47{ 48 UtvideoContext *c = avctx->priv_data; 49 int i; 50 51 av_freep(&avctx->coded_frame); 52 av_freep(&c->slice_bits); 53 for (i = 0; i < 4; i++) 54 av_freep(&c->slice_buffer[i]); 55 56 return 0; 57} 58 59static av_cold int utvideo_encode_init(AVCodecContext *avctx) 60{ 61 UtvideoContext *c = avctx->priv_data; 62 int i, subsampled_height; 63 uint32_t original_format; 64 65 c->avctx = avctx; 66 c->frame_info_size = 4; 67 c->slice_stride = FFALIGN(avctx->width, 32); 68 69 switch (avctx->pix_fmt) { 70 case AV_PIX_FMT_RGB24: 71 c->planes = 3; 72 avctx->codec_tag = MKTAG('U', 'L', 'R', 'G'); 73 original_format = UTVIDEO_RGB; 74 break; 75 case AV_PIX_FMT_RGBA: 76 c->planes = 4; 77 avctx->codec_tag = MKTAG('U', 'L', 'R', 'A'); 78 original_format = UTVIDEO_RGBA; 79 break; 80 case AV_PIX_FMT_YUV420P: 81 if (avctx->width & 1 || avctx->height & 1) { 82 av_log(avctx, AV_LOG_ERROR, 83 "4:2:0 video requires even width and height.\n"); 84 return AVERROR_INVALIDDATA; 85 } 86 c->planes = 3; 87 if (avctx->colorspace == AVCOL_SPC_BT709) 88 avctx->codec_tag = MKTAG('U', 'L', 'H', '0'); 89 else 90 avctx->codec_tag = MKTAG('U', 'L', 'Y', '0'); 91 original_format = UTVIDEO_420; 92 break; 93 case AV_PIX_FMT_YUV422P: 94 if (avctx->width & 1) { 95 av_log(avctx, AV_LOG_ERROR, 96 "4:2:2 video requires even width.\n"); 97 return AVERROR_INVALIDDATA; 98 } 99 c->planes = 3; 100 if (avctx->colorspace == AVCOL_SPC_BT709) 101 avctx->codec_tag = MKTAG('U', 'L', 'H', '2'); 102 else 103 avctx->codec_tag = MKTAG('U', 'L', 'Y', '2'); 104 original_format = UTVIDEO_422; 105 break; 106 default: 107 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n", 108 avctx->pix_fmt); 109 return AVERROR_INVALIDDATA; 110 } 111 112 ff_bswapdsp_init(&c->bdsp); 113 ff_huffyuvencdsp_init(&c->hdsp); 114 115 /* Check the prediction method, and error out if unsupported */ 116 if (avctx->prediction_method < 0 || avctx->prediction_method > 4) { 117 av_log(avctx, AV_LOG_WARNING, 118 "Prediction method %d is not supported in Ut Video.\n", 119 avctx->prediction_method); 120 return AVERROR_OPTION_NOT_FOUND; 121 } 122 123 if (avctx->prediction_method == FF_PRED_PLANE) { 124 av_log(avctx, AV_LOG_ERROR, 125 "Plane prediction is not supported in Ut Video.\n"); 126 return AVERROR_OPTION_NOT_FOUND; 127 } 128 129 /* Convert from libavcodec prediction type to Ut Video's */ 130 c->frame_pred = ff_ut_pred_order[avctx->prediction_method]; 131 132 if (c->frame_pred == PRED_GRADIENT) { 133 av_log(avctx, AV_LOG_ERROR, "Gradient prediction is not supported.\n"); 134 return AVERROR_OPTION_NOT_FOUND; 135 } 136 137 /* 138 * Check the asked slice count for obviously invalid 139 * values (> 256 or negative). 140 */ 141 if (avctx->slices > 256 || avctx->slices < 0) { 142 av_log(avctx, AV_LOG_ERROR, 143 "Slice count %d is not supported in Ut Video (theoretical range is 0-256).\n", 144 avctx->slices); 145 return AVERROR(EINVAL); 146 } 147 148 /* Check that the slice count is not larger than the subsampled height */ 149 subsampled_height = avctx->height >> av_pix_fmt_desc_get(avctx->pix_fmt)->log2_chroma_h; 150 if (avctx->slices > subsampled_height) { 151 av_log(avctx, AV_LOG_ERROR, 152 "Slice count %d is larger than the subsampling-applied height %d.\n", 153 avctx->slices, subsampled_height); 154 return AVERROR(EINVAL); 155 } 156 157 avctx->coded_frame = av_frame_alloc(); 158 159 if (!avctx->coded_frame) { 160 av_log(avctx, AV_LOG_ERROR, "Could not allocate frame.\n"); 161 utvideo_encode_close(avctx); 162 return AVERROR(ENOMEM); 163 } 164 165 /* extradata size is 4 * 32bit */ 166 avctx->extradata_size = 16; 167 168 avctx->extradata = av_mallocz(avctx->extradata_size + 169 FF_INPUT_BUFFER_PADDING_SIZE); 170 171 if (!avctx->extradata) { 172 av_log(avctx, AV_LOG_ERROR, "Could not allocate extradata.\n"); 173 utvideo_encode_close(avctx); 174 return AVERROR(ENOMEM); 175 } 176 177 for (i = 0; i < c->planes; i++) { 178 c->slice_buffer[i] = av_malloc(c->slice_stride * (avctx->height + 2) + 179 FF_INPUT_BUFFER_PADDING_SIZE); 180 if (!c->slice_buffer[i]) { 181 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 1.\n"); 182 utvideo_encode_close(avctx); 183 return AVERROR(ENOMEM); 184 } 185 } 186 187 /* 188 * Set the version of the encoder. 189 * Last byte is "implementation ID", which is 190 * obtained from the creator of the format. 191 * Libavcodec has been assigned with the ID 0xF0. 192 */ 193 AV_WB32(avctx->extradata, MKTAG(1, 0, 0, 0xF0)); 194 195 /* 196 * Set the "original format" 197 * Not used for anything during decoding. 198 */ 199 AV_WL32(avctx->extradata + 4, original_format); 200 201 /* Write 4 as the 'frame info size' */ 202 AV_WL32(avctx->extradata + 8, c->frame_info_size); 203 204 /* 205 * Set how many slices are going to be used. 206 * By default uses multiple slices depending on the subsampled height. 207 * This enables multithreading in the official decoder. 208 */ 209 if (!avctx->slices) { 210 c->slices = subsampled_height / 120; 211 212 if (!c->slices) 213 c->slices = 1; 214 else if (c->slices > 256) 215 c->slices = 256; 216 } else { 217 c->slices = avctx->slices; 218 } 219 220 /* Set compression mode */ 221 c->compression = COMP_HUFF; 222 223 /* 224 * Set the encoding flags: 225 * - Slice count minus 1 226 * - Interlaced encoding mode flag, set to zero for now. 227 * - Compression mode (none/huff) 228 * And write the flags. 229 */ 230 c->flags = (c->slices - 1) << 24; 231 c->flags |= 0 << 11; // bit field to signal interlaced encoding mode 232 c->flags |= c->compression; 233 234 AV_WL32(avctx->extradata + 12, c->flags); 235 236 return 0; 237} 238 239static void mangle_rgb_planes(uint8_t *dst[4], int dst_stride, uint8_t *src, 240 int step, int stride, int width, int height) 241{ 242 int i, j; 243 int k = 2 * dst_stride; 244 unsigned int g; 245 246 for (j = 0; j < height; j++) { 247 if (step == 3) { 248 for (i = 0; i < width * step; i += step) { 249 g = src[i + 1]; 250 dst[0][k] = g; 251 g += 0x80; 252 dst[1][k] = src[i + 2] - g; 253 dst[2][k] = src[i + 0] - g; 254 k++; 255 } 256 } else { 257 for (i = 0; i < width * step; i += step) { 258 g = src[i + 1]; 259 dst[0][k] = g; 260 g += 0x80; 261 dst[1][k] = src[i + 2] - g; 262 dst[2][k] = src[i + 0] - g; 263 dst[3][k] = src[i + 3]; 264 k++; 265 } 266 } 267 k += dst_stride - width; 268 src += stride; 269 } 270} 271 272/* Write data to a plane with left prediction */ 273static void left_predict(uint8_t *src, uint8_t *dst, int stride, 274 int width, int height) 275{ 276 int i, j; 277 uint8_t prev; 278 279 prev = 0x80; /* Set the initial value */ 280 for (j = 0; j < height; j++) { 281 for (i = 0; i < width; i++) { 282 *dst++ = src[i] - prev; 283 prev = src[i]; 284 } 285 src += stride; 286 } 287} 288 289/* Write data to a plane with median prediction */ 290static void median_predict(UtvideoContext *c, uint8_t *src, uint8_t *dst, int stride, 291 int width, int height) 292{ 293 int i, j; 294 int A, B; 295 uint8_t prev; 296 297 /* First line uses left neighbour prediction */ 298 prev = 0x80; /* Set the initial value */ 299 for (i = 0; i < width; i++) { 300 *dst++ = src[i] - prev; 301 prev = src[i]; 302 } 303 304 if (height == 1) 305 return; 306 307 src += stride; 308 309 /* 310 * Second line uses top prediction for the first sample, 311 * and median for the rest. 312 */ 313 A = B = 0; 314 315 /* Rest of the coded part uses median prediction */ 316 for (j = 1; j < height; j++) { 317 c->hdsp.sub_hfyu_median_pred(dst, src - stride, src, width, &A, &B); 318 dst += width; 319 src += stride; 320 } 321} 322 323/* Count the usage of values in a plane */ 324static void count_usage(uint8_t *src, int width, 325 int height, uint64_t *counts) 326{ 327 int i, j; 328 329 for (j = 0; j < height; j++) { 330 for (i = 0; i < width; i++) { 331 counts[src[i]]++; 332 } 333 src += width; 334 } 335} 336 337/* Calculate the actual huffman codes from the code lengths */ 338static void calculate_codes(HuffEntry *he) 339{ 340 int last, i; 341 uint32_t code; 342 343 qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len); 344 345 last = 255; 346 while (he[last].len == 255 && last) 347 last--; 348 349 code = 1; 350 for (i = last; i >= 0; i--) { 351 he[i].code = code >> (32 - he[i].len); 352 code += 0x80000000u >> (he[i].len - 1); 353 } 354 355 qsort(he, 256, sizeof(*he), huff_cmp_sym); 356} 357 358/* Write huffman bit codes to a memory block */ 359static int write_huff_codes(uint8_t *src, uint8_t *dst, int dst_size, 360 int width, int height, HuffEntry *he) 361{ 362 PutBitContext pb; 363 int i, j; 364 int count; 365 366 init_put_bits(&pb, dst, dst_size); 367 368 /* Write the codes */ 369 for (j = 0; j < height; j++) { 370 for (i = 0; i < width; i++) 371 put_bits(&pb, he[src[i]].len, he[src[i]].code); 372 373 src += width; 374 } 375 376 /* Pad output to a 32bit boundary */ 377 count = put_bits_count(&pb) & 0x1F; 378 379 if (count) 380 put_bits(&pb, 32 - count, 0); 381 382 /* Get the amount of bits written */ 383 count = put_bits_count(&pb); 384 385 /* Flush the rest with zeroes */ 386 flush_put_bits(&pb); 387 388 return count; 389} 390 391static int encode_plane(AVCodecContext *avctx, uint8_t *src, 392 uint8_t *dst, int stride, 393 int width, int height, PutByteContext *pb) 394{ 395 UtvideoContext *c = avctx->priv_data; 396 uint8_t lengths[256]; 397 uint64_t counts[256] = { 0 }; 398 399 HuffEntry he[256]; 400 401 uint32_t offset = 0, slice_len = 0; 402 int i, sstart, send = 0; 403 int symbol; 404 int ret; 405 406 /* Do prediction / make planes */ 407 switch (c->frame_pred) { 408 case PRED_NONE: 409 for (i = 0; i < c->slices; i++) { 410 sstart = send; 411 send = height * (i + 1) / c->slices; 412 av_image_copy_plane(dst + sstart * width, width, 413 src + sstart * stride, stride, 414 width, send - sstart); 415 } 416 break; 417 case PRED_LEFT: 418 for (i = 0; i < c->slices; i++) { 419 sstart = send; 420 send = height * (i + 1) / c->slices; 421 left_predict(src + sstart * stride, dst + sstart * width, 422 stride, width, send - sstart); 423 } 424 break; 425 case PRED_MEDIAN: 426 for (i = 0; i < c->slices; i++) { 427 sstart = send; 428 send = height * (i + 1) / c->slices; 429 median_predict(c, src + sstart * stride, dst + sstart * width, 430 stride, width, send - sstart); 431 } 432 break; 433 default: 434 av_log(avctx, AV_LOG_ERROR, "Unknown prediction mode: %d\n", 435 c->frame_pred); 436 return AVERROR_OPTION_NOT_FOUND; 437 } 438 439 /* Count the usage of values */ 440 count_usage(dst, width, height, counts); 441 442 /* Check for a special case where only one symbol was used */ 443 for (symbol = 0; symbol < 256; symbol++) { 444 /* If non-zero count is found, see if it matches width * height */ 445 if (counts[symbol]) { 446 /* Special case if only one symbol was used */ 447 if (counts[symbol] == width * (int64_t)height) { 448 /* 449 * Write a zero for the single symbol 450 * used in the plane, else 0xFF. 451 */ 452 for (i = 0; i < 256; i++) { 453 if (i == symbol) 454 bytestream2_put_byte(pb, 0); 455 else 456 bytestream2_put_byte(pb, 0xFF); 457 } 458 459 /* Write zeroes for lengths */ 460 for (i = 0; i < c->slices; i++) 461 bytestream2_put_le32(pb, 0); 462 463 /* And that's all for that plane folks */ 464 return 0; 465 } 466 break; 467 } 468 } 469 470 /* Calculate huffman lengths */ 471 if ((ret = ff_huff_gen_len_table(lengths, counts, 256, 1)) < 0) 472 return ret; 473 474 /* 475 * Write the plane's header into the output packet: 476 * - huffman code lengths (256 bytes) 477 * - slice end offsets (gotten from the slice lengths) 478 */ 479 for (i = 0; i < 256; i++) { 480 bytestream2_put_byte(pb, lengths[i]); 481 482 he[i].len = lengths[i]; 483 he[i].sym = i; 484 } 485 486 /* Calculate the huffman codes themselves */ 487 calculate_codes(he); 488 489 send = 0; 490 for (i = 0; i < c->slices; i++) { 491 sstart = send; 492 send = height * (i + 1) / c->slices; 493 494 /* 495 * Write the huffman codes to a buffer, 496 * get the offset in bits and convert to bytes. 497 */ 498 offset += write_huff_codes(dst + sstart * width, c->slice_bits, 499 width * height + 4, width, 500 send - sstart, he) >> 3; 501 502 slice_len = offset - slice_len; 503 504 /* Byteswap the written huffman codes */ 505 c->bdsp.bswap_buf((uint32_t *) c->slice_bits, 506 (uint32_t *) c->slice_bits, 507 slice_len >> 2); 508 509 /* Write the offset to the stream */ 510 bytestream2_put_le32(pb, offset); 511 512 /* Seek to the data part of the packet */ 513 bytestream2_seek_p(pb, 4 * (c->slices - i - 1) + 514 offset - slice_len, SEEK_CUR); 515 516 /* Write the slices' data into the output packet */ 517 bytestream2_put_buffer(pb, c->slice_bits, slice_len); 518 519 /* Seek back to the slice offsets */ 520 bytestream2_seek_p(pb, -4 * (c->slices - i - 1) - offset, 521 SEEK_CUR); 522 523 slice_len = offset; 524 } 525 526 /* And at the end seek to the end of written slice(s) */ 527 bytestream2_seek_p(pb, offset, SEEK_CUR); 528 529 return 0; 530} 531 532static int utvideo_encode_frame(AVCodecContext *avctx, AVPacket *pkt, 533 const AVFrame *pic, int *got_packet) 534{ 535 UtvideoContext *c = avctx->priv_data; 536 PutByteContext pb; 537 538 uint32_t frame_info; 539 540 uint8_t *dst; 541 542 int width = avctx->width, height = avctx->height; 543 int i, ret = 0; 544 545 /* Allocate a new packet if needed, and set it to the pointer dst */ 546 ret = ff_alloc_packet2(avctx, pkt, (256 + 4 * c->slices + width * height) * 547 c->planes + 4); 548 549 if (ret < 0) 550 return ret; 551 552 dst = pkt->data; 553 554 bytestream2_init_writer(&pb, dst, pkt->size); 555 556 av_fast_padded_malloc(&c->slice_bits, &c->slice_bits_size, width * height + 4); 557 558 if (!c->slice_bits) { 559 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 2.\n"); 560 return AVERROR(ENOMEM); 561 } 562 563 /* In case of RGB, mangle the planes to Ut Video's format */ 564 if (avctx->pix_fmt == AV_PIX_FMT_RGBA || avctx->pix_fmt == AV_PIX_FMT_RGB24) 565 mangle_rgb_planes(c->slice_buffer, c->slice_stride, pic->data[0], 566 c->planes, pic->linesize[0], width, height); 567 568 /* Deal with the planes */ 569 switch (avctx->pix_fmt) { 570 case AV_PIX_FMT_RGB24: 571 case AV_PIX_FMT_RGBA: 572 for (i = 0; i < c->planes; i++) { 573 ret = encode_plane(avctx, c->slice_buffer[i] + 2 * c->slice_stride, 574 c->slice_buffer[i], c->slice_stride, 575 width, height, &pb); 576 577 if (ret) { 578 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 579 return ret; 580 } 581 } 582 break; 583 case AV_PIX_FMT_YUV422P: 584 for (i = 0; i < c->planes; i++) { 585 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0], 586 pic->linesize[i], width >> !!i, height, &pb); 587 588 if (ret) { 589 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 590 return ret; 591 } 592 } 593 break; 594 case AV_PIX_FMT_YUV420P: 595 for (i = 0; i < c->planes; i++) { 596 ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0], 597 pic->linesize[i], width >> !!i, height >> !!i, 598 &pb); 599 600 if (ret) { 601 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i); 602 return ret; 603 } 604 } 605 break; 606 default: 607 av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n", 608 avctx->pix_fmt); 609 return AVERROR_INVALIDDATA; 610 } 611 612 /* 613 * Write frame information (LE 32bit unsigned) 614 * into the output packet. 615 * Contains the prediction method. 616 */ 617 frame_info = c->frame_pred << 8; 618 bytestream2_put_le32(&pb, frame_info); 619 620 /* 621 * At least currently Ut Video is IDR only. 622 * Set flags accordingly. 623 */ 624 avctx->coded_frame->key_frame = 1; 625 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; 626 627 pkt->size = bytestream2_tell_p(&pb); 628 pkt->flags |= AV_PKT_FLAG_KEY; 629 630 /* Packet should be done */ 631 *got_packet = 1; 632 633 return 0; 634} 635 636AVCodec ff_utvideo_encoder = { 637 .name = "utvideo", 638 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"), 639 .type = AVMEDIA_TYPE_VIDEO, 640 .id = AV_CODEC_ID_UTVIDEO, 641 .priv_data_size = sizeof(UtvideoContext), 642 .init = utvideo_encode_init, 643 .encode2 = utvideo_encode_frame, 644 .close = utvideo_encode_close, 645 .capabilities = CODEC_CAP_FRAME_THREADS | CODEC_CAP_INTRA_ONLY, 646 .pix_fmts = (const enum AVPixelFormat[]) { 647 AV_PIX_FMT_RGB24, AV_PIX_FMT_RGBA, AV_PIX_FMT_YUV422P, 648 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE 649 }, 650}; 651