libavcodec/lagarith.c

   1 /*
   2  * Lagarith lossless decoder
   3  * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
   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 libavcodec/lagarith.c
  24  * Lagarith lossless decoder
  25  * @author Nathan Caldwell
  26  *
  27  */
  28
  29 #include "avcodec.h"
  30 #include "get_bits.h"
  31 #include "mathops.h"
  32 #include "dsputil.h"
  33 #include "lagarithrac.h"
  34
  35 enum LagarithFrameType {
  36     FRAME_RAW           = 1,    /*!< Uncompressed */
  37     FRAME_U_RGB24       = 2,    /*!< Unaligned RGB24 */
  38     FRAME_ARITH_YUY2    = 3,    /*!< Arith coded YUY2 */
  39     FRAME_ARITH_RGB24   = 4,    /*!< Arith coded RGB24 */
  40     FRAME_SOLID_GRAY    = 5,    /*!< Solid grayscale color frame */
  41     FRAME_SOLID_COLOR   = 6,    /*!< Solid non-grayscale color frame */
  42     FRAME_OLD_ARITH_RGB = 7,    /*!< Obsolete arithmetic coded RGB (Maintained for backwards compatibility) */
  43     FRAME_ARITH_RGBA    = 8,    /*!< Arithmetic coded RGBA */
  44     FRAME_SOLID_RGBA    = 9,    /*!< Solid RGBA color frame */
  45     FRAME_ARITH_YV12    = 10,   /*!< Arithmetic coded YV12 */
  46     FRAME_REDUCED_RES   = 11,   /*!< Reduced resolution frame */
  47 };
  48
  49 typedef struct LagarithContext {
  50     AVCodecContext *avctx;
  51     AVFrame picture;
  52     DSPContext dsp;
  53     int zeros;                  /*!< number of consecutave zero bytes encountered */
  54     int zeros_rem;              /*!< number of zero bytes remaining to output */
  55 } LagarithContext;
  56
  57 /**
  58  * Compute the 52bit mantissa of 1/(double)denom.
  59  * This is used because it is more portable across architectures and
  60  * toolchains than using floats.
  61  * @param denom denominator
  62  * @return 52bit mantissa
  63  * @see softfloat_mul
  64  */
  65 static uint64_t softfloat_reciprocal(uint32_t denom)
  66 {
  67     int shift = av_log2(denom - 1) + 1;
  68     uint64_t ret = (1ULL << 52) / denom;
  69     uint64_t err = (1ULL << 52) - ret * denom;
  70     ret <<= shift;
  71     err <<= shift;
  72     err +=  denom / 2;
  73     return ret + err / denom;
  74 }
  75
  76 /**
  77  * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
  78  * Used in combination with softfloat_reciprocal computes x/(double)denom.
  79  * @param x 32bit integer factor
  80  * @param mantissa mantissa of f with exponent 0
  81  * @return 32bit integer value (x*f)
  82  * @see softfloat_reciprocal
  83  */
  84 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
  85 {
  86     uint64_t l = x * (mantissa & 0xffffffff);
  87     uint64_t h = x * (mantissa >> 32);
  88     h += l >> 32;
  89     l &= 0xffffffff;
  90     l += 1 << av_log2(h >> 21);
  91     h += l >> 32;
  92     return h >> 20;
  93 }
  94
  95 static void lag_memset(uint8_t *s, uint8_t c, size_t n, int step)
  96 {
  97     int i;
  98     if (step == 1) {
  99         memset(s, c, n);
 100         return;
 101     }
 102
 103     for (i = 0; i < n * step; i += step)
 104         s[i] = c;
 105 }
 106
 107 static uint8_t *lag_memcpy(uint8_t *dest, const uint8_t *src, size_t n,
 108                            int step)
 109 {
 110     int i, j;
 111     if (step == 1)
 112         return memcpy(dest, src, n);
 113
 114     for (i = j = 0; i < n; i++) {
 115         dest[j] = src[i];
 116         j += step;
 117     }
 118     return dest;
 119 }
 120
 121 static uint8_t lag_calc_zero_run(int8_t x)
 122 {
 123     return (x << 1) ^ (x >> 7);
 124 }
 125
 126 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
 127 {
 128     static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
 129     int i;
 130     int bit     = 0;
 131     int bits    = 0;
 132     int prevbit = 0;
 133     unsigned val;
 134
 135     for (i = 0; i < 7; i++) {
 136         if (prevbit && bit)
 137             break;
 138         prevbit = bit;
 139         bit = get_bits1(gb);
 140         if (bit && !prevbit)
 141             bits += series[i];
 142     }
 143     bits--;
 144     if (bits < 0 || bits > 31) {
 145         *value = 0;
 146         return -1;
 147     } else if (bits == 0) {
 148         *value = 0;
 149         return 0;
 150     }
 151
 152     val  = get_bits_long(gb, bits);
 153     val |= 1 << bits;
 154
 155     *value = val - 1;
 156
 157     return 0;
 158 }
 159
 160 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
 161 {
 162     int i, j, scale_factor;
 163     unsigned prob, cumulative_target;
 164     unsigned cumul_prob = 0;
 165     unsigned scaled_cumul_prob = 0;
 166
 167     rac->prob[0] = 0;
 168     rac->prob[257] = UINT_MAX;
 169     /* Read probabilities from bitstream */
 170     for (i = 1; i < 257; i++) {
 171         if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
 172             av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
 173             return -1;
 174         }
 175         cumul_prob += rac->prob[i];
 176         if (!rac->prob[i]) {
 177             if (lag_decode_prob(gb, &prob)) {
 178                 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
 179                 return -1;
 180             }
 181             if (prob > 257 - i)
 182                 prob = 257 - i;
 183             for (j = 0; j < prob; j++)
 184                 rac->prob[++i] = 0;
 185         }
 186     }
 187
 188     if (!cumul_prob) {
 189         av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
 190         return -1;
 191     }
 192
 193     /* Scale probabilities so cumulative probability is an even power of 2. */
 194     scale_factor = av_log2(cumul_prob);
 195
 196     if (cumul_prob & (cumul_prob - 1)) {
 197         uint64_t mul = softfloat_reciprocal(cumul_prob);
 198         for (i = 1; i < 257; i++) {
 199             rac->prob[i] = softfloat_mul(rac->prob[i], mul);
 200             scaled_cumul_prob += rac->prob[i];
 201         }
 202
 203         scale_factor++;
 204         cumulative_target = 1 << scale_factor;
 205
 206         if (scaled_cumul_prob > cumulative_target) {
 207             av_log(rac->avctx, AV_LOG_ERROR,
 208                    "Scaled probabilities are larger than target!\n");
 209             return -1;
 210         }
 211
 212         scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
 213
 214         for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
 215             if (rac->prob[i]) {
 216                 rac->prob[i]++;
 217                 scaled_cumul_prob--;
 218             }
 219             /* Comment from reference source:
 220              * if (b & 0x80 == 0) {     // order of operations is 'wrong'; it has been left this way
 221              *                          // since the compression change is negligable and fixing it
 222              *                          // breaks backwards compatibilty
 223              *      b =- (signed int)b;
 224              *      b &= 0xFF;
 225              * } else {
 226              *      b++;
 227              *      b &= 0x7f;
 228              * }
 229              */
 230         }
 231     }
 232
 233     rac->scale = scale_factor;
 234
 235     /* Fill probability array with cumulative probability for each symbol. */
 236     for (i = 1; i < 257; i++)
 237         rac->prob[i] += rac->prob[i - 1];
 238
 239     return 0;
 240 }
 241
 242 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
 243                                       uint8_t *diff, int w, int *left,
 244                                       int *left_top)
 245 {
 246     /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
 247      * However the &0xFF on the gradient predictor yealds incorrect output
 248      * for lagarith.
 249      */
 250     int i;
 251     uint8_t l, lt;
 252
 253     l  = *left;
 254     lt = *left_top;
 255
 256     for (i = 0; i < w; i++) {
 257         l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
 258         lt = src1[i];
 259         dst[i] = l;
 260     }
 261
 262     *left     = l;
 263     *left_top = lt;
 264 }
 265
 266 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
 267                           int width, int stride, int step, int line)
 268 {
 269     int i = 0;
 270     int L, TL;
 271     int width_scaled = width * step;
 272
 273     if (!line) {
 274         /* Left prediction only for first line */
 275         L = l->dsp.add_hfyu_left_prediction(buf + step, buf + step,
 276                                             width - step, buf[0]);
 277         return;
 278     } else if (line == 1) {
 279         /* Second line, left predict first pixel, the rest of the line is median predicted */
 280         /* FIXME: In the case of RGB this pixel is top predicted */
 281         TL = buf[-stride];
 282         L = l->dsp.add_hfyu_left_prediction(buf, buf, 1,
 283                                             buf[width_scaled - stride - step]);
 284         i += step;
 285     } else {
 286         /* Left pixel is actually prev_row[width] */
 287         L = buf[width_scaled - stride - step];
 288         /* Top left is 2 rows back, last pixel */
 289         TL = buf[width_scaled - (2 * stride) - step];
 290     }
 291
 292     if (i < width_scaled)
 293         add_lag_median_prediction(buf + i, buf - stride + i, buf + i,
 294                                   width_scaled - i, &L, &TL);
 295 }
 296
 297 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
 298                            uint8_t *dst, int width, int stride,
 299                            int step, int esc_count)
 300 {
 301     int i = 0;
 302     int ret = 0;
 303
 304     if (!esc_count)
 305         esc_count = -1;
 306
 307     /* Output any zeros remaining from the previous run */
 308 handle_zeros:
 309     if (l->zeros_rem) {
 310         int count = FFMIN(l->zeros_rem, width - i);
 311         lag_memset(dst + i * step, 0, count, step);
 312         i += count;
 313         l->zeros_rem -= count;
 314     }
 315
 316     while (i < width) {
 317
 318         dst[i * step] = lag_get_rac(rac);
 319         ret++;
 320
 321         if (dst[i * step])
 322             l->zeros = 0;
 323         else
 324             l->zeros++;
 325
 326         i++;
 327         if (l->zeros == esc_count) {
 328             int index = lag_get_rac(rac);
 329             ret++;
 330
 331             l->zeros = 0;
 332
 333             l->zeros_rem = lag_calc_zero_run(index);
 334             goto handle_zeros;
 335         }
 336     }
 337     return ret;
 338 }
 339
 340 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
 341                                     const uint8_t *src, int width,
 342                                     int step, int esc_count)
 343 {
 344     int i = 0;
 345     int count;
 346     uint8_t zero_run = 0;
 347     const uint8_t *start = src;
 348     uint8_t mask1 = -(esc_count < 2);
 349     uint8_t mask2 = -(esc_count < 3);
 350     uint8_t *end = dst + (width - 2) * step;
 351
 352 output_zeros:
 353     if (l->zeros_rem) {
 354         count = FFMIN(l->zeros_rem, width - i);
 355         lag_memset(dst, 0, count, step);
 356         l->zeros_rem -= count;
 357         dst += count;
 358     }
 359
 360     while (dst < end) {
 361         i = 0;
 362         while (!zero_run && dst + i * step < end) {
 363             i++;
 364             zero_run =
 365                 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
 366         }
 367         if (zero_run) {
 368             zero_run = 0;
 369             i += esc_count;
 370             lag_memcpy(dst, src, i, step);
 371             dst += i;
 372             l->zeros_rem = lag_calc_zero_run(src[i]);
 373
 374             src += i + 1;
 375             goto output_zeros;
 376         } else {
 377             lag_memcpy(dst, src, i, step);
 378             src += i;
 379         }
 380     }
 381     return start - src;
 382 }
 383
 384
 385
 386 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
 387                                   int width, int height, int stride,
 388                                   int step, const uint8_t *src,
 389                                   int src_size)
 390 {
 391     int i = 0;
 392     int read = 0;
 393     uint32_t length;
 394     uint32_t offset = 1;
 395     int esc_count = src[0];
 396     GetBitContext gb;
 397     lag_rac rac;
 398
 399     rac.avctx = l->avctx;
 400     l->zeros = 0;
 401
 402     if (esc_count < 4) {
 403         length = width * height;
 404         if (esc_count && AV_RL32(src + 1) < length) {
 405             length = AV_RL32(src + 1);
 406             offset += 4;
 407         }
 408
 409         init_get_bits(&gb, src + offset, src_size * 8);
 410
 411         if (lag_read_prob_header(&rac, &gb) < 0)
 412             return -1;
 413
 414         lag_rac_init(&rac, &gb, length - stride);
 415
 416         for (i = 0; i < height; i++)
 417             read +=
 418                 lag_decode_line(l, &rac, dst + (i * stride), width,
 419                                 stride, step, esc_count);
 420
 421         if (read > length)
 422             av_log(l->avctx, AV_LOG_WARNING,
 423                    "Output more bytes than length (%d of %d)\n", read,
 424                    length);
 425     } else if (esc_count < 8) {
 426         esc_count -= 4;
 427         if (esc_count > 0) {
 428             /* Zero run coding only, no range coding. */
 429             for (i = 0; i < height; i++)
 430                 src +=
 431                     lag_decode_zero_run_line(l, dst + (i * stride),
 432                                              src, width, step, esc_count);
 433         } else {
 434             /* Plane is stored uncompressed */
 435             for (i = 0; i < height; i++) {
 436                 lag_memcpy(dst + (i * stride), src, width, step);
 437                 src += width;
 438             }
 439         }
 440     } else if (esc_count == 0xff) {
 441         /* Plane is a solid run of 0 bytes */
 442         for (i = 0; i < height; i++)
 443             lag_memset(dst + i * stride, 0, width, step);
 444     } else {
 445         av_log(l->avctx, AV_LOG_ERROR,
 446                "Invalid zero run escape code! (%#x)\n", esc_count);
 447         return -1;
 448     }
 449
 450     for (i = 0; i < height; i++) {
 451         lag_pred_line(l, dst, width, stride, step, i);
 452         dst += stride;
 453     }
 454
 455     return 0;
 456 }
 457
 458 /**
 459  * Decode a frame.
 460  * @param avctx codec context
 461  * @param data output AVFrame
 462  * @param data_size size of output data or 0 if no picture is returned
 463  * @param avpkt input packet
 464  * @return number of consumed bytes on success or negative if decode fails
 465  */
 466 static int lag_decode_frame(AVCodecContext *avctx,
 467                             void *data, int *data_size, AVPacket *avpkt)
 468 {
 469     const uint8_t *buf = avpkt->data;
 470     int buf_size = avpkt->size;
 471     LagarithContext *l = avctx->priv_data;
 472     AVFrame *const p = &l->picture;
 473     uint8_t frametype = 0;
 474     uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
 475
 476     AVFrame *picture = data;
 477
 478     if (!l->avctx)
 479         l->avctx = avctx;
 480
 481     if (p->data[0])
 482         avctx->release_buffer(avctx, p);
 483
 484     p->reference = 0;
 485     p->key_frame = 1;
 486
 487     frametype = buf[0];
 488
 489     offset_gu = AV_RL32(buf + 1);
 490     offset_bv = AV_RL32(buf + 5);
 491
 492     switch (frametype) {
 493     case FRAME_ARITH_YV12:
 494         avctx->pix_fmt = PIX_FMT_YUV420P;
 495
 496         if (avctx->get_buffer(avctx, p) < 0) {
 497             av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 498             return -1;
 499         }
 500
 501         lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
 502                                p->linesize[0], 1, buf + offset_ry,
 503                                buf_size);
 504         lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
 505                                avctx->height / 2, p->linesize[2], 1,
 506                                buf + offset_gu, buf_size);
 507         lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
 508                                avctx->height / 2, p->linesize[1], 1,
 509                                buf + offset_bv, buf_size);
 510         break;
 511     default:
 512         av_log(avctx, AV_LOG_ERROR,
 513                "Unsupported Lagarith frame type: %#x\n", frametype);
 514         return -1;
 515     }
 516
 517     *picture = *p;
 518     *data_size = sizeof(AVFrame);
 519
 520     return buf_size;
 521 }
 522
 523 static av_cold int lag_decode_init(AVCodecContext *avctx)
 524 {
 525     LagarithContext *l = avctx->priv_data;
 526
 527     dsputil_init(&l->dsp, avctx);
 528
 529     return 0;
 530 }
 531
 532 AVCodec lagarith_decoder = {
 533     "lagarith",
 534     CODEC_TYPE_VIDEO,
 535     CODEC_ID_LAGARITH,
 536     sizeof(LagarithContext),
 537     lag_decode_init,
 538     NULL,
 539     NULL,
 540     lag_decode_frame,
 541     CODEC_CAP_DR1,
 542     .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
 543 };