libavcodec/utvideoenc.c

   1 /*
   2  * Ut Video encoder
   3  * Copyright (c) 2012 Jan Ekström
   4  *
   5  * This file is part of Libav.
   6  *
   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.
  11  *
  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.
  16  *
  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
  20  */
  21
  22 /**
  23  * @file
  24  * Ut Video encoder
  25  */
  26
  27 #include "libavutil/intreadwrite.h"
  28 #include "avcodec.h"
  29 #include "internal.h"
  30 #include "bytestream.h"
  31 #include "put_bits.h"
  32 #include "dsputil.h"
  33 #include "mathops.h"
  34 #include "utvideo.h"
  35 #include "huffman.h"
  36
  37 /* Compare huffentry symbols */
  38 static int huff_cmp_sym(const void *a, const void *b)
  39 {
  40     const HuffEntry *aa = a, *bb = b;
  41     return aa->sym - bb->sym;
  42 }
  43
  44 static av_cold int utvideo_encode_close(AVCodecContext *avctx)
  45 {
  46     UtvideoContext *c = avctx->priv_data;
  47     int i;
  48
  49     av_freep(&avctx->coded_frame);
  50     av_freep(&c->slice_bits);
  51     for (i = 0; i < 4; i++)
  52         av_freep(&c->slice_buffer[i]);
  53
  54     return 0;
  55 }
  56
  57 static av_cold int utvideo_encode_init(AVCodecContext *avctx)
  58 {
  59     UtvideoContext *c = avctx->priv_data;
  60     int i;
  61     uint32_t original_format;
  62
  63     c->avctx           = avctx;
  64     c->frame_info_size = 4;
  65     c->slice_stride    = FFALIGN(avctx->width, 32);
  66
  67     switch (avctx->pix_fmt) {
  68     case AV_PIX_FMT_RGB24:
  69         c->planes        = 3;
  70         avctx->codec_tag = MKTAG('U', 'L', 'R', 'G');
  71         original_format  = UTVIDEO_RGB;
  72         break;
  73     case AV_PIX_FMT_RGBA:
  74         c->planes        = 4;
  75         avctx->codec_tag = MKTAG('U', 'L', 'R', 'A');
  76         original_format  = UTVIDEO_RGBA;
  77         break;
  78     case AV_PIX_FMT_YUV420P:
  79         if (avctx->width & 1 || avctx->height & 1) {
  80             av_log(avctx, AV_LOG_ERROR,
  81                    "4:2:0 video requires even width and height.\n");
  82             return AVERROR_INVALIDDATA;
  83         }
  84         c->planes        = 3;
  85         avctx->codec_tag = MKTAG('U', 'L', 'Y', '0');
  86         original_format  = UTVIDEO_420;
  87         break;
  88     case AV_PIX_FMT_YUV422P:
  89         if (avctx->width & 1) {
  90             av_log(avctx, AV_LOG_ERROR,
  91                    "4:2:2 video requires even width.\n");
  92             return AVERROR_INVALIDDATA;
  93         }
  94         c->planes        = 3;
  95         avctx->codec_tag = MKTAG('U', 'L', 'Y', '2');
  96         original_format  = UTVIDEO_422;
  97         break;
  98     default:
  99         av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
 100                avctx->pix_fmt);
 101         return AVERROR_INVALIDDATA;
 102     }
 103
 104     ff_dsputil_init(&c->dsp, avctx);
 105
 106     /* Check the prediction method, and error out if unsupported */
 107     if (avctx->prediction_method < 0 || avctx->prediction_method > 4) {
 108         av_log(avctx, AV_LOG_WARNING,
 109                "Prediction method %d is not supported in Ut Video.\n",
 110                avctx->prediction_method);
 111         return AVERROR_OPTION_NOT_FOUND;
 112     }
 113
 114     if (avctx->prediction_method == FF_PRED_PLANE) {
 115         av_log(avctx, AV_LOG_ERROR,
 116                "Plane prediction is not supported in Ut Video.\n");
 117         return AVERROR_OPTION_NOT_FOUND;
 118     }
 119
 120     /* Convert from libavcodec prediction type to Ut Video's */
 121     c->frame_pred = ff_ut_pred_order[avctx->prediction_method];
 122
 123     if (c->frame_pred == PRED_GRADIENT) {
 124         av_log(avctx, AV_LOG_ERROR, "Gradient prediction is not supported.\n");
 125         return AVERROR_OPTION_NOT_FOUND;
 126     }
 127
 128     avctx->coded_frame = avcodec_alloc_frame();
 129
 130     if (!avctx->coded_frame) {
 131         av_log(avctx, AV_LOG_ERROR, "Could not allocate frame.\n");
 132         utvideo_encode_close(avctx);
 133         return AVERROR(ENOMEM);
 134     }
 135
 136     /* extradata size is 4 * 32bit */
 137     avctx->extradata_size = 16;
 138
 139     avctx->extradata = av_mallocz(avctx->extradata_size +
 140                                   FF_INPUT_BUFFER_PADDING_SIZE);
 141
 142     if (!avctx->extradata) {
 143         av_log(avctx, AV_LOG_ERROR, "Could not allocate extradata.\n");
 144         utvideo_encode_close(avctx);
 145         return AVERROR(ENOMEM);
 146     }
 147
 148     for (i = 0; i < c->planes; i++) {
 149         c->slice_buffer[i] = av_malloc(c->slice_stride * (avctx->height + 2) +
 150                                        FF_INPUT_BUFFER_PADDING_SIZE);
 151         if (!c->slice_buffer[i]) {
 152             av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 1.\n");
 153             utvideo_encode_close(avctx);
 154             return AVERROR(ENOMEM);
 155         }
 156     }
 157
 158     /*
 159      * Set the version of the encoder.
 160      * Last byte is "implementation ID", which is
 161      * obtained from the creator of the format.
 162      * Libavcodec has been assigned with the ID 0xF0.
 163      */
 164     AV_WB32(avctx->extradata, MKTAG(1, 0, 0, 0xF0));
 165
 166     /*
 167      * Set the "original format"
 168      * Not used for anything during decoding.
 169      */
 170     AV_WL32(avctx->extradata + 4, original_format);
 171
 172     /* Write 4 as the 'frame info size' */
 173     AV_WL32(avctx->extradata + 8, c->frame_info_size);
 174
 175     /*
 176      * Set how many slices are going to be used.
 177      * Set one slice for now.
 178      */
 179     c->slices = 1;
 180
 181     /* Set compression mode */
 182     c->compression = COMP_HUFF;
 183
 184     /*
 185      * Set the encoding flags:
 186      * - Slice count minus 1
 187      * - Interlaced encoding mode flag, set to zero for now.
 188      * - Compression mode (none/huff)
 189      * And write the flags.
 190      */
 191     c->flags  = (c->slices - 1) << 24;
 192     c->flags |= 0 << 11; // bit field to signal interlaced encoding mode
 193     c->flags |= c->compression;
 194
 195     AV_WL32(avctx->extradata + 12, c->flags);
 196
 197     return 0;
 198 }
 199
 200 static void mangle_rgb_planes(uint8_t *dst[4], int dst_stride, uint8_t *src,
 201                               int step, int stride, int width, int height)
 202 {
 203     int i, j;
 204     int k = 2 * dst_stride;
 205     unsigned int g;
 206
 207     for (j = 0; j < height; j++) {
 208         if (step == 3) {
 209             for (i = 0; i < width * step; i += step) {
 210                 g         = src[i + 1];
 211                 dst[0][k] = g;
 212                 g        += 0x80;
 213                 dst[1][k] = src[i + 2] - g;
 214                 dst[2][k] = src[i + 0] - g;
 215                 k++;
 216             }
 217         } else {
 218             for (i = 0; i < width * step; i += step) {
 219                 g         = src[i + 1];
 220                 dst[0][k] = g;
 221                 g        += 0x80;
 222                 dst[1][k] = src[i + 2] - g;
 223                 dst[2][k] = src[i + 0] - g;
 224                 dst[3][k] = src[i + 3];
 225                 k++;
 226             }
 227         }
 228         k += dst_stride - width;
 229         src += stride;
 230     }
 231 }
 232
 233 /* Write data to a plane, no prediction applied */
 234 static void write_plane(uint8_t *src, uint8_t *dst, int stride,
 235                         int width, int height)
 236 {
 237     int i, j;
 238
 239     for (j = 0; j < height; j++) {
 240         for (i = 0; i < width; i++)
 241             *dst++ = src[i];
 242
 243         src += stride;
 244     }
 245 }
 246
 247 /* Write data to a plane with left prediction */
 248 static void left_predict(uint8_t *src, uint8_t *dst, int stride,
 249                          int width, int height)
 250 {
 251     int i, j;
 252     uint8_t prev;
 253
 254     prev = 0x80; /* Set the initial value */
 255     for (j = 0; j < height; j++) {
 256         for (i = 0; i < width; i++) {
 257             *dst++ = src[i] - prev;
 258             prev   = src[i];
 259         }
 260         src += stride;
 261     }
 262 }
 263
 264 /* Write data to a plane with median prediction */
 265 static void median_predict(UtvideoContext *c, uint8_t *src, uint8_t *dst, int stride,
 266                            int width, int height)
 267 {
 268     int i, j;
 269     int A, B;
 270     uint8_t prev;
 271
 272     /* First line uses left neighbour prediction */
 273     prev = 0x80; /* Set the initial value */
 274     for (i = 0; i < width; i++) {
 275         *dst++ = src[i] - prev;
 276         prev   = src[i];
 277     }
 278
 279     if (height == 1)
 280         return;
 281
 282     src += stride;
 283
 284     /*
 285      * Second line uses top prediction for the first sample,
 286      * and median for the rest.
 287      */
 288     A = B = 0;
 289
 290     /* Rest of the coded part uses median prediction */
 291     for (j = 1; j < height; j++) {
 292         c->dsp.sub_hfyu_median_prediction(dst, src - stride, src, width, &A, &B);
 293         dst += width;
 294         src += stride;
 295     }
 296 }
 297
 298 /* Count the usage of values in a plane */
 299 static void count_usage(uint8_t *src, int width,
 300                         int height, uint64_t *counts)
 301 {
 302     int i, j;
 303
 304     for (j = 0; j < height; j++) {
 305         for (i = 0; i < width; i++) {
 306             counts[src[i]]++;
 307         }
 308         src += width;
 309     }
 310 }
 311
 312 /* Calculate the actual huffman codes from the code lengths */
 313 static void calculate_codes(HuffEntry *he)
 314 {
 315     int last, i;
 316     uint32_t code;
 317
 318     qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len);
 319
 320     last = 255;
 321     while (he[last].len == 255 && last)
 322         last--;
 323
 324     code = 1;
 325     for (i = last; i >= 0; i--) {
 326         he[i].code  = code >> (32 - he[i].len);
 327         code       += 0x80000000u >> (he[i].len - 1);
 328     }
 329
 330     qsort(he, 256, sizeof(*he), huff_cmp_sym);
 331 }
 332
 333 /* Write huffman bit codes to a memory block */
 334 static int write_huff_codes(uint8_t *src, uint8_t *dst, int dst_size,
 335                             int width, int height, HuffEntry *he)
 336 {
 337     PutBitContext pb;
 338     int i, j;
 339     int count;
 340
 341     init_put_bits(&pb, dst, dst_size);
 342
 343     /* Write the codes */
 344     for (j = 0; j < height; j++) {
 345         for (i = 0; i < width; i++)
 346             put_bits(&pb, he[src[i]].len, he[src[i]].code);
 347
 348         src += width;
 349     }
 350
 351     /* Pad output to a 32bit boundary */
 352     count = put_bits_count(&pb) & 0x1F;
 353
 354     if (count)
 355         put_bits(&pb, 32 - count, 0);
 356
 357     /* Get the amount of bits written */
 358     count = put_bits_count(&pb);
 359
 360     /* Flush the rest with zeroes */
 361     flush_put_bits(&pb);
 362
 363     return count;
 364 }
 365
 366 static int encode_plane(AVCodecContext *avctx, uint8_t *src,
 367                         uint8_t *dst, int stride,
 368                         int width, int height, PutByteContext *pb)
 369 {
 370     UtvideoContext *c        = avctx->priv_data;
 371     uint8_t  lengths[256];
 372     uint64_t counts[256]     = { 0 };
 373
 374     HuffEntry he[256];
 375
 376     uint32_t offset = 0, slice_len = 0;
 377     int      i, sstart, send = 0;
 378     int      symbol;
 379
 380     /* Do prediction / make planes */
 381     switch (c->frame_pred) {
 382     case PRED_NONE:
 383         for (i = 0; i < c->slices; i++) {
 384             sstart = send;
 385             send   = height * (i + 1) / c->slices;
 386             write_plane(src + sstart * stride, dst + sstart * width,
 387                         stride, width, send - sstart);
 388         }
 389         break;
 390     case PRED_LEFT:
 391         for (i = 0; i < c->slices; i++) {
 392             sstart = send;
 393             send   = height * (i + 1) / c->slices;
 394             left_predict(src + sstart * stride, dst + sstart * width,
 395                          stride, width, send - sstart);
 396         }
 397         break;
 398     case PRED_MEDIAN:
 399         for (i = 0; i < c->slices; i++) {
 400             sstart = send;
 401             send   = height * (i + 1) / c->slices;
 402             median_predict(c, src + sstart * stride, dst + sstart * width,
 403                            stride, width, send - sstart);
 404         }
 405         break;
 406     default:
 407         av_log(avctx, AV_LOG_ERROR, "Unknown prediction mode: %d\n",
 408                c->frame_pred);
 409         return AVERROR_OPTION_NOT_FOUND;
 410     }
 411
 412     /* Count the usage of values */
 413     count_usage(dst, width, height, counts);
 414
 415     /* Check for a special case where only one symbol was used */
 416     for (symbol = 0; symbol < 256; symbol++) {
 417         /* If non-zero count is found, see if it matches width * height */
 418         if (counts[symbol]) {
 419             /* Special case if only one symbol was used */
 420             if (counts[symbol] == width * height) {
 421                 /*
 422                  * Write a zero for the single symbol
 423                  * used in the plane, else 0xFF.
 424                  */
 425                 for (i = 0; i < 256; i++) {
 426                     if (i == symbol)
 427                         bytestream2_put_byte(pb, 0);
 428                     else
 429                         bytestream2_put_byte(pb, 0xFF);
 430                 }
 431
 432                 /* Write zeroes for lengths */
 433                 for (i = 0; i < c->slices; i++)
 434                     bytestream2_put_le32(pb, 0);
 435
 436                 /* And that's all for that plane folks */
 437                 return 0;
 438             }
 439             break;
 440         }
 441     }
 442
 443     /* Calculate huffman lengths */
 444     ff_huff_gen_len_table(lengths, counts);
 445
 446     /*
 447      * Write the plane's header into the output packet:
 448      * - huffman code lengths (256 bytes)
 449      * - slice end offsets (gotten from the slice lengths)
 450      */
 451     for (i = 0; i < 256; i++) {
 452         bytestream2_put_byte(pb, lengths[i]);
 453
 454         he[i].len = lengths[i];
 455         he[i].sym = i;
 456     }
 457
 458     /* Calculate the huffman codes themselves */
 459     calculate_codes(he);
 460
 461     send = 0;
 462     for (i = 0; i < c->slices; i++) {
 463         sstart  = send;
 464         send    = height * (i + 1) / c->slices;
 465
 466         /*
 467          * Write the huffman codes to a buffer,
 468          * get the offset in bits and convert to bytes.
 469          */
 470         offset += write_huff_codes(dst + sstart * width, c->slice_bits,
 471                                    width * (send - sstart), width,
 472                                    send - sstart, he) >> 3;
 473
 474         slice_len = offset - slice_len;
 475
 476         /* Byteswap the written huffman codes */
 477         c->dsp.bswap_buf((uint32_t *) c->slice_bits,
 478                          (uint32_t *) c->slice_bits,
 479                          slice_len >> 2);
 480
 481         /* Write the offset to the stream */
 482         bytestream2_put_le32(pb, offset);
 483
 484         /* Seek to the data part of the packet */
 485         bytestream2_seek_p(pb, 4 * (c->slices - i - 1) +
 486                            offset - slice_len, SEEK_CUR);
 487
 488         /* Write the slices' data into the output packet */
 489         bytestream2_put_buffer(pb, c->slice_bits, slice_len);
 490
 491         /* Seek back to the slice offsets */
 492         bytestream2_seek_p(pb, -4 * (c->slices - i - 1) - offset,
 493                            SEEK_CUR);
 494
 495         slice_len = offset;
 496     }
 497
 498     /* And at the end seek to the end of written slice(s) */
 499     bytestream2_seek_p(pb, offset, SEEK_CUR);
 500
 501     return 0;
 502 }
 503
 504 static int utvideo_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
 505                                 const AVFrame *pic, int *got_packet)
 506 {
 507     UtvideoContext *c = avctx->priv_data;
 508     PutByteContext pb;
 509
 510     uint32_t frame_info;
 511
 512     uint8_t *dst;
 513
 514     int width = avctx->width, height = avctx->height;
 515     int i, ret = 0;
 516
 517     /* Allocate a new packet if needed, and set it to the pointer dst */
 518     ret = ff_alloc_packet(pkt, (256 + 4 * c->slices + width * height) *
 519                           c->planes + 4);
 520
 521     if (ret < 0) {
 522         av_log(avctx, AV_LOG_ERROR,
 523                "Error allocating the output packet, or the provided packet "
 524                "was too small.\n");
 525         return ret;
 526     }
 527
 528     dst = pkt->data;
 529
 530     bytestream2_init_writer(&pb, dst, pkt->size);
 531
 532     av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
 533                    width * height + FF_INPUT_BUFFER_PADDING_SIZE);
 534
 535     if (!c->slice_bits) {
 536         av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer 2.\n");
 537         return AVERROR(ENOMEM);
 538     }
 539
 540     /* In case of RGB, mangle the planes to Ut Video's format */
 541     if (avctx->pix_fmt == AV_PIX_FMT_RGBA || avctx->pix_fmt == AV_PIX_FMT_RGB24)
 542         mangle_rgb_planes(c->slice_buffer, c->slice_stride, pic->data[0],
 543                           c->planes, pic->linesize[0], width, height);
 544
 545     /* Deal with the planes */
 546     switch (avctx->pix_fmt) {
 547     case AV_PIX_FMT_RGB24:
 548     case AV_PIX_FMT_RGBA:
 549         for (i = 0; i < c->planes; i++) {
 550             ret = encode_plane(avctx, c->slice_buffer[i] + 2 * c->slice_stride,
 551                                c->slice_buffer[i], c->slice_stride,
 552                                width, height, &pb);
 553
 554             if (ret) {
 555                 av_log(avctx, AV_LOG_ERROR, "Error encoding plane %d.\n", i);
 556                 return ret;
 557             }
 558         }
 559         break;
 560     case AV_PIX_FMT_YUV422P:
 561         for (i = 0; i < c->planes; i++) {
 562             ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
 563                                pic->linesize[i], width >> !!i, 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_YUV420P:
 572         for (i = 0; i < c->planes; i++) {
 573             ret = encode_plane(avctx, pic->data[i], c->slice_buffer[0],
 574                                pic->linesize[i], width >> !!i, height >> !!i,
 575                                &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     default:
 584         av_log(avctx, AV_LOG_ERROR, "Unknown pixel format: %d\n",
 585                avctx->pix_fmt);
 586         return AVERROR_INVALIDDATA;
 587     }
 588
 589     /*
 590      * Write frame information (LE 32bit unsigned)
 591      * into the output packet.
 592      * Contains the prediction method.
 593      */
 594     frame_info = c->frame_pred << 8;
 595     bytestream2_put_le32(&pb, frame_info);
 596
 597     /*
 598      * At least currently Ut Video is IDR only.
 599      * Set flags accordingly.
 600      */
 601     avctx->coded_frame->key_frame = 1;
 602     avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
 603
 604     pkt->size   = bytestream2_tell_p(&pb);
 605     pkt->flags |= AV_PKT_FLAG_KEY;
 606
 607     /* Packet should be done */
 608     *got_packet = 1;
 609
 610     return 0;
 611 }
 612
 613 AVCodec ff_utvideo_encoder = {
 614     .name           = "utvideo",
 615     .type           = AVMEDIA_TYPE_VIDEO,
 616     .id             = AV_CODEC_ID_UTVIDEO,
 617     .priv_data_size = sizeof(UtvideoContext),
 618     .init           = utvideo_encode_init,
 619     .encode2        = utvideo_encode_frame,
 620     .close          = utvideo_encode_close,
 621     .pix_fmts       = (const enum AVPixelFormat[]) {
 622                           AV_PIX_FMT_RGB24, AV_PIX_FMT_RGBA, AV_PIX_FMT_YUV422P,
 623                           AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE
 624                       },
 625     .long_name      = NULL_IF_CONFIG_SMALL("Ut Video"),
 626 };