| 1 | /* |
| 2 | * TAK decoder |
| 3 | * Copyright (c) 2012 Paul B Mahol |
| 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 | * TAK (Tom's lossless Audio Kompressor) decoder |
| 25 | * @author Paul B Mahol |
| 26 | */ |
| 27 | |
| 28 | #include "libavutil/internal.h" |
| 29 | #include "libavutil/samplefmt.h" |
| 30 | #include "tak.h" |
| 31 | #include "audiodsp.h" |
| 32 | #include "thread.h" |
| 33 | #include "avcodec.h" |
| 34 | #include "internal.h" |
| 35 | #include "unary.h" |
| 36 | |
| 37 | #define MAX_SUBFRAMES 8 ///< max number of subframes per channel |
| 38 | #define MAX_PREDICTORS 256 |
| 39 | |
| 40 | typedef struct MCDParam { |
| 41 | int8_t present; ///< decorrelation parameter availability for this channel |
| 42 | int8_t index; ///< index into array of decorrelation types |
| 43 | int8_t chan1; |
| 44 | int8_t chan2; |
| 45 | } MCDParam; |
| 46 | |
| 47 | typedef struct TAKDecContext { |
| 48 | AVCodecContext *avctx; ///< parent AVCodecContext |
| 49 | AudioDSPContext adsp; |
| 50 | TAKStreamInfo ti; |
| 51 | GetBitContext gb; ///< bitstream reader initialized to start at the current frame |
| 52 | |
| 53 | int uval; |
| 54 | int nb_samples; ///< number of samples in the current frame |
| 55 | uint8_t *decode_buffer; |
| 56 | unsigned int decode_buffer_size; |
| 57 | int32_t *decoded[TAK_MAX_CHANNELS]; ///< decoded samples for each channel |
| 58 | |
| 59 | int8_t lpc_mode[TAK_MAX_CHANNELS]; |
| 60 | int8_t sample_shift[TAK_MAX_CHANNELS]; ///< shift applied to every sample in the channel |
| 61 | int16_t predictors[MAX_PREDICTORS]; |
| 62 | int nb_subframes; ///< number of subframes in the current frame |
| 63 | int16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples |
| 64 | int subframe_scale; |
| 65 | |
| 66 | int8_t dmode; ///< channel decorrelation type in the current frame |
| 67 | |
| 68 | MCDParam mcdparams[TAK_MAX_CHANNELS]; ///< multichannel decorrelation parameters |
| 69 | |
| 70 | int8_t coding_mode[128]; |
| 71 | DECLARE_ALIGNED(16, int16_t, filter)[MAX_PREDICTORS]; |
| 72 | DECLARE_ALIGNED(16, int16_t, residues)[544]; |
| 73 | } TAKDecContext; |
| 74 | |
| 75 | static const int8_t mc_dmodes[] = { 1, 3, 4, 6, }; |
| 76 | |
| 77 | static const uint16_t predictor_sizes[] = { |
| 78 | 4, 8, 12, 16, 24, 32, 48, 64, 80, 96, 128, 160, 192, 224, 256, 0, |
| 79 | }; |
| 80 | |
| 81 | static const struct CParam { |
| 82 | int init; |
| 83 | int escape; |
| 84 | int scale; |
| 85 | int aescape; |
| 86 | int bias; |
| 87 | } xcodes[50] = { |
| 88 | { 0x01, 0x0000001, 0x0000001, 0x0000003, 0x0000008 }, |
| 89 | { 0x02, 0x0000003, 0x0000001, 0x0000007, 0x0000006 }, |
| 90 | { 0x03, 0x0000005, 0x0000002, 0x000000E, 0x000000D }, |
| 91 | { 0x03, 0x0000003, 0x0000003, 0x000000D, 0x0000018 }, |
| 92 | { 0x04, 0x000000B, 0x0000004, 0x000001C, 0x0000019 }, |
| 93 | { 0x04, 0x0000006, 0x0000006, 0x000001A, 0x0000030 }, |
| 94 | { 0x05, 0x0000016, 0x0000008, 0x0000038, 0x0000032 }, |
| 95 | { 0x05, 0x000000C, 0x000000C, 0x0000034, 0x0000060 }, |
| 96 | { 0x06, 0x000002C, 0x0000010, 0x0000070, 0x0000064 }, |
| 97 | { 0x06, 0x0000018, 0x0000018, 0x0000068, 0x00000C0 }, |
| 98 | { 0x07, 0x0000058, 0x0000020, 0x00000E0, 0x00000C8 }, |
| 99 | { 0x07, 0x0000030, 0x0000030, 0x00000D0, 0x0000180 }, |
| 100 | { 0x08, 0x00000B0, 0x0000040, 0x00001C0, 0x0000190 }, |
| 101 | { 0x08, 0x0000060, 0x0000060, 0x00001A0, 0x0000300 }, |
| 102 | { 0x09, 0x0000160, 0x0000080, 0x0000380, 0x0000320 }, |
| 103 | { 0x09, 0x00000C0, 0x00000C0, 0x0000340, 0x0000600 }, |
| 104 | { 0x0A, 0x00002C0, 0x0000100, 0x0000700, 0x0000640 }, |
| 105 | { 0x0A, 0x0000180, 0x0000180, 0x0000680, 0x0000C00 }, |
| 106 | { 0x0B, 0x0000580, 0x0000200, 0x0000E00, 0x0000C80 }, |
| 107 | { 0x0B, 0x0000300, 0x0000300, 0x0000D00, 0x0001800 }, |
| 108 | { 0x0C, 0x0000B00, 0x0000400, 0x0001C00, 0x0001900 }, |
| 109 | { 0x0C, 0x0000600, 0x0000600, 0x0001A00, 0x0003000 }, |
| 110 | { 0x0D, 0x0001600, 0x0000800, 0x0003800, 0x0003200 }, |
| 111 | { 0x0D, 0x0000C00, 0x0000C00, 0x0003400, 0x0006000 }, |
| 112 | { 0x0E, 0x0002C00, 0x0001000, 0x0007000, 0x0006400 }, |
| 113 | { 0x0E, 0x0001800, 0x0001800, 0x0006800, 0x000C000 }, |
| 114 | { 0x0F, 0x0005800, 0x0002000, 0x000E000, 0x000C800 }, |
| 115 | { 0x0F, 0x0003000, 0x0003000, 0x000D000, 0x0018000 }, |
| 116 | { 0x10, 0x000B000, 0x0004000, 0x001C000, 0x0019000 }, |
| 117 | { 0x10, 0x0006000, 0x0006000, 0x001A000, 0x0030000 }, |
| 118 | { 0x11, 0x0016000, 0x0008000, 0x0038000, 0x0032000 }, |
| 119 | { 0x11, 0x000C000, 0x000C000, 0x0034000, 0x0060000 }, |
| 120 | { 0x12, 0x002C000, 0x0010000, 0x0070000, 0x0064000 }, |
| 121 | { 0x12, 0x0018000, 0x0018000, 0x0068000, 0x00C0000 }, |
| 122 | { 0x13, 0x0058000, 0x0020000, 0x00E0000, 0x00C8000 }, |
| 123 | { 0x13, 0x0030000, 0x0030000, 0x00D0000, 0x0180000 }, |
| 124 | { 0x14, 0x00B0000, 0x0040000, 0x01C0000, 0x0190000 }, |
| 125 | { 0x14, 0x0060000, 0x0060000, 0x01A0000, 0x0300000 }, |
| 126 | { 0x15, 0x0160000, 0x0080000, 0x0380000, 0x0320000 }, |
| 127 | { 0x15, 0x00C0000, 0x00C0000, 0x0340000, 0x0600000 }, |
| 128 | { 0x16, 0x02C0000, 0x0100000, 0x0700000, 0x0640000 }, |
| 129 | { 0x16, 0x0180000, 0x0180000, 0x0680000, 0x0C00000 }, |
| 130 | { 0x17, 0x0580000, 0x0200000, 0x0E00000, 0x0C80000 }, |
| 131 | { 0x17, 0x0300000, 0x0300000, 0x0D00000, 0x1800000 }, |
| 132 | { 0x18, 0x0B00000, 0x0400000, 0x1C00000, 0x1900000 }, |
| 133 | { 0x18, 0x0600000, 0x0600000, 0x1A00000, 0x3000000 }, |
| 134 | { 0x19, 0x1600000, 0x0800000, 0x3800000, 0x3200000 }, |
| 135 | { 0x19, 0x0C00000, 0x0C00000, 0x3400000, 0x6000000 }, |
| 136 | { 0x1A, 0x2C00000, 0x1000000, 0x7000000, 0x6400000 }, |
| 137 | { 0x1A, 0x1800000, 0x1800000, 0x6800000, 0xC000000 }, |
| 138 | }; |
| 139 | |
| 140 | static int set_bps_params(AVCodecContext *avctx) |
| 141 | { |
| 142 | switch (avctx->bits_per_raw_sample) { |
| 143 | case 8: |
| 144 | avctx->sample_fmt = AV_SAMPLE_FMT_U8P; |
| 145 | break; |
| 146 | case 16: |
| 147 | avctx->sample_fmt = AV_SAMPLE_FMT_S16P; |
| 148 | break; |
| 149 | case 24: |
| 150 | avctx->sample_fmt = AV_SAMPLE_FMT_S32P; |
| 151 | break; |
| 152 | default: |
| 153 | av_log(avctx, AV_LOG_ERROR, "invalid/unsupported bits per sample: %d\n", |
| 154 | avctx->bits_per_raw_sample); |
| 155 | return AVERROR_INVALIDDATA; |
| 156 | } |
| 157 | |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | static void set_sample_rate_params(AVCodecContext *avctx) |
| 162 | { |
| 163 | TAKDecContext *s = avctx->priv_data; |
| 164 | int shift = 3 - (avctx->sample_rate / 11025); |
| 165 | shift = FFMAX(0, shift); |
| 166 | s->uval = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << shift; |
| 167 | s->subframe_scale = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << 1; |
| 168 | } |
| 169 | |
| 170 | static av_cold int tak_decode_init(AVCodecContext *avctx) |
| 171 | { |
| 172 | TAKDecContext *s = avctx->priv_data; |
| 173 | |
| 174 | ff_audiodsp_init(&s->adsp); |
| 175 | |
| 176 | s->avctx = avctx; |
| 177 | avctx->bits_per_raw_sample = avctx->bits_per_coded_sample; |
| 178 | |
| 179 | set_sample_rate_params(avctx); |
| 180 | |
| 181 | return set_bps_params(avctx); |
| 182 | } |
| 183 | |
| 184 | static void decode_lpc(int32_t *coeffs, int mode, int length) |
| 185 | { |
| 186 | int i; |
| 187 | |
| 188 | if (length < 2) |
| 189 | return; |
| 190 | |
| 191 | if (mode == 1) { |
| 192 | int a1 = *coeffs++; |
| 193 | for (i = 0; i < length - 1 >> 1; i++) { |
| 194 | *coeffs += a1; |
| 195 | coeffs[1] += *coeffs; |
| 196 | a1 = coeffs[1]; |
| 197 | coeffs += 2; |
| 198 | } |
| 199 | if (length - 1 & 1) |
| 200 | *coeffs += a1; |
| 201 | } else if (mode == 2) { |
| 202 | int a1 = coeffs[1]; |
| 203 | int a2 = a1 + *coeffs; |
| 204 | coeffs[1] = a2; |
| 205 | if (length > 2) { |
| 206 | coeffs += 2; |
| 207 | for (i = 0; i < length - 2 >> 1; i++) { |
| 208 | int a3 = *coeffs + a1; |
| 209 | int a4 = a3 + a2; |
| 210 | *coeffs = a4; |
| 211 | a1 = coeffs[1] + a3; |
| 212 | a2 = a1 + a4; |
| 213 | coeffs[1] = a2; |
| 214 | coeffs += 2; |
| 215 | } |
| 216 | if (length & 1) |
| 217 | *coeffs += a1 + a2; |
| 218 | } |
| 219 | } else if (mode == 3) { |
| 220 | int a1 = coeffs[1]; |
| 221 | int a2 = a1 + *coeffs; |
| 222 | coeffs[1] = a2; |
| 223 | if (length > 2) { |
| 224 | int a3 = coeffs[2]; |
| 225 | int a4 = a3 + a1; |
| 226 | int a5 = a4 + a2; |
| 227 | coeffs += 3; |
| 228 | for (i = 0; i < length - 3; i++) { |
| 229 | a3 += *coeffs; |
| 230 | a4 += a3; |
| 231 | a5 += a4; |
| 232 | *coeffs = a5; |
| 233 | coeffs++; |
| 234 | } |
| 235 | } |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | static int decode_segment(TAKDecContext *s, int8_t mode, int32_t *decoded, int len) |
| 240 | { |
| 241 | struct CParam code; |
| 242 | GetBitContext *gb = &s->gb; |
| 243 | int i; |
| 244 | |
| 245 | if (!mode) { |
| 246 | memset(decoded, 0, len * sizeof(*decoded)); |
| 247 | return 0; |
| 248 | } |
| 249 | |
| 250 | if (mode > FF_ARRAY_ELEMS(xcodes)) |
| 251 | return AVERROR_INVALIDDATA; |
| 252 | code = xcodes[mode - 1]; |
| 253 | |
| 254 | for (i = 0; i < len; i++) { |
| 255 | int x = get_bits_long(gb, code.init); |
| 256 | if (x >= code.escape && get_bits1(gb)) { |
| 257 | x |= 1 << code.init; |
| 258 | if (x >= code.aescape) { |
| 259 | int scale = get_unary(gb, 1, 9); |
| 260 | if (scale == 9) { |
| 261 | int scale_bits = get_bits(gb, 3); |
| 262 | if (scale_bits > 0) { |
| 263 | if (scale_bits == 7) { |
| 264 | scale_bits += get_bits(gb, 5); |
| 265 | if (scale_bits > 29) |
| 266 | return AVERROR_INVALIDDATA; |
| 267 | } |
| 268 | scale = get_bits_long(gb, scale_bits) + 1; |
| 269 | x += code.scale * scale; |
| 270 | } |
| 271 | x += code.bias; |
| 272 | } else |
| 273 | x += code.scale * scale - code.escape; |
| 274 | } else |
| 275 | x -= code.escape; |
| 276 | } |
| 277 | decoded[i] = (x >> 1) ^ -(x & 1); |
| 278 | } |
| 279 | |
| 280 | return 0; |
| 281 | } |
| 282 | |
| 283 | static int decode_residues(TAKDecContext *s, int32_t *decoded, int length) |
| 284 | { |
| 285 | GetBitContext *gb = &s->gb; |
| 286 | int i, mode, ret; |
| 287 | |
| 288 | if (length > s->nb_samples) |
| 289 | return AVERROR_INVALIDDATA; |
| 290 | |
| 291 | if (get_bits1(gb)) { |
| 292 | int wlength, rval; |
| 293 | |
| 294 | wlength = length / s->uval; |
| 295 | |
| 296 | rval = length - (wlength * s->uval); |
| 297 | |
| 298 | if (rval < s->uval / 2) |
| 299 | rval += s->uval; |
| 300 | else |
| 301 | wlength++; |
| 302 | |
| 303 | if (wlength <= 1 || wlength > 128) |
| 304 | return AVERROR_INVALIDDATA; |
| 305 | |
| 306 | s->coding_mode[0] = mode = get_bits(gb, 6); |
| 307 | |
| 308 | for (i = 1; i < wlength; i++) { |
| 309 | int c = get_unary(gb, 1, 6); |
| 310 | |
| 311 | switch (c) { |
| 312 | case 6: |
| 313 | mode = get_bits(gb, 6); |
| 314 | break; |
| 315 | case 5: |
| 316 | case 4: |
| 317 | case 3: { |
| 318 | /* mode += sign ? (1 - c) : (c - 1) */ |
| 319 | int sign = get_bits1(gb); |
| 320 | mode += (-sign ^ (c - 1)) + sign; |
| 321 | break; |
| 322 | } |
| 323 | case 2: |
| 324 | mode++; |
| 325 | break; |
| 326 | case 1: |
| 327 | mode--; |
| 328 | break; |
| 329 | } |
| 330 | s->coding_mode[i] = mode; |
| 331 | } |
| 332 | |
| 333 | i = 0; |
| 334 | while (i < wlength) { |
| 335 | int len = 0; |
| 336 | |
| 337 | mode = s->coding_mode[i]; |
| 338 | do { |
| 339 | if (i >= wlength - 1) |
| 340 | len += rval; |
| 341 | else |
| 342 | len += s->uval; |
| 343 | i++; |
| 344 | |
| 345 | if (i == wlength) |
| 346 | break; |
| 347 | } while (s->coding_mode[i] == mode); |
| 348 | |
| 349 | if ((ret = decode_segment(s, mode, decoded, len)) < 0) |
| 350 | return ret; |
| 351 | decoded += len; |
| 352 | } |
| 353 | } else { |
| 354 | mode = get_bits(gb, 6); |
| 355 | if ((ret = decode_segment(s, mode, decoded, length)) < 0) |
| 356 | return ret; |
| 357 | } |
| 358 | |
| 359 | return 0; |
| 360 | } |
| 361 | |
| 362 | static int get_bits_esc4(GetBitContext *gb) |
| 363 | { |
| 364 | if (get_bits1(gb)) |
| 365 | return get_bits(gb, 4) + 1; |
| 366 | else |
| 367 | return 0; |
| 368 | } |
| 369 | |
| 370 | static int decode_subframe(TAKDecContext *s, int32_t *decoded, |
| 371 | int subframe_size, int prev_subframe_size) |
| 372 | { |
| 373 | GetBitContext *gb = &s->gb; |
| 374 | int x, y, i, j, ret = 0; |
| 375 | int dshift, size, filter_quant, filter_order; |
| 376 | int tfilter[MAX_PREDICTORS]; |
| 377 | |
| 378 | if (!get_bits1(gb)) |
| 379 | return decode_residues(s, decoded, subframe_size); |
| 380 | |
| 381 | filter_order = predictor_sizes[get_bits(gb, 4)]; |
| 382 | |
| 383 | if (prev_subframe_size > 0 && get_bits1(gb)) { |
| 384 | if (filter_order > prev_subframe_size) |
| 385 | return AVERROR_INVALIDDATA; |
| 386 | |
| 387 | decoded -= filter_order; |
| 388 | subframe_size += filter_order; |
| 389 | |
| 390 | if (filter_order > subframe_size) |
| 391 | return AVERROR_INVALIDDATA; |
| 392 | } else { |
| 393 | int lpc_mode; |
| 394 | |
| 395 | if (filter_order > subframe_size) |
| 396 | return AVERROR_INVALIDDATA; |
| 397 | |
| 398 | lpc_mode = get_bits(gb, 2); |
| 399 | if (lpc_mode > 2) |
| 400 | return AVERROR_INVALIDDATA; |
| 401 | |
| 402 | if ((ret = decode_residues(s, decoded, filter_order)) < 0) |
| 403 | return ret; |
| 404 | |
| 405 | if (lpc_mode) |
| 406 | decode_lpc(decoded, lpc_mode, filter_order); |
| 407 | } |
| 408 | |
| 409 | dshift = get_bits_esc4(gb); |
| 410 | size = get_bits1(gb) + 6; |
| 411 | |
| 412 | filter_quant = 10; |
| 413 | if (get_bits1(gb)) { |
| 414 | filter_quant -= get_bits(gb, 3) + 1; |
| 415 | if (filter_quant < 3) |
| 416 | return AVERROR_INVALIDDATA; |
| 417 | } |
| 418 | |
| 419 | s->predictors[0] = get_sbits(gb, 10); |
| 420 | s->predictors[1] = get_sbits(gb, 10); |
| 421 | s->predictors[2] = get_sbits(gb, size) << (10 - size); |
| 422 | s->predictors[3] = get_sbits(gb, size) << (10 - size); |
| 423 | if (filter_order > 4) { |
| 424 | int tmp = size - get_bits1(gb); |
| 425 | |
| 426 | for (i = 4; i < filter_order; i++) { |
| 427 | if (!(i & 3)) |
| 428 | x = tmp - get_bits(gb, 2); |
| 429 | s->predictors[i] = get_sbits(gb, x) << (10 - size); |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | tfilter[0] = s->predictors[0] << 6; |
| 434 | for (i = 1; i < filter_order; i++) { |
| 435 | int32_t *p1 = &tfilter[0]; |
| 436 | int32_t *p2 = &tfilter[i - 1]; |
| 437 | |
| 438 | for (j = 0; j < (i + 1) / 2; j++) { |
| 439 | x = *p1 + (s->predictors[i] * *p2 + 256 >> 9); |
| 440 | *p2 += s->predictors[i] * *p1 + 256 >> 9; |
| 441 | *p1++ = x; |
| 442 | p2--; |
| 443 | } |
| 444 | |
| 445 | tfilter[i] = s->predictors[i] << 6; |
| 446 | } |
| 447 | |
| 448 | x = 1 << (32 - (15 - filter_quant)); |
| 449 | y = 1 << ((15 - filter_quant) - 1); |
| 450 | for (i = 0, j = filter_order - 1; i < filter_order / 2; i++, j--) { |
| 451 | s->filter[j] = x - ((tfilter[i] + y) >> (15 - filter_quant)); |
| 452 | s->filter[i] = x - ((tfilter[j] + y) >> (15 - filter_quant)); |
| 453 | } |
| 454 | |
| 455 | if ((ret = decode_residues(s, &decoded[filter_order], |
| 456 | subframe_size - filter_order)) < 0) |
| 457 | return ret; |
| 458 | |
| 459 | for (i = 0; i < filter_order; i++) |
| 460 | s->residues[i] = *decoded++ >> dshift; |
| 461 | |
| 462 | y = FF_ARRAY_ELEMS(s->residues) - filter_order; |
| 463 | x = subframe_size - filter_order; |
| 464 | while (x > 0) { |
| 465 | int tmp = FFMIN(y, x); |
| 466 | |
| 467 | for (i = 0; i < tmp; i++) { |
| 468 | int v = 1 << (filter_quant - 1); |
| 469 | |
| 470 | if (filter_order & -16) |
| 471 | v += s->adsp.scalarproduct_int16(&s->residues[i], s->filter, |
| 472 | filter_order & -16); |
| 473 | for (j = filter_order & -16; j < filter_order; j += 4) { |
| 474 | v += s->residues[i + j + 3] * s->filter[j + 3] + |
| 475 | s->residues[i + j + 2] * s->filter[j + 2] + |
| 476 | s->residues[i + j + 1] * s->filter[j + 1] + |
| 477 | s->residues[i + j ] * s->filter[j ]; |
| 478 | } |
| 479 | v = (av_clip(v >> filter_quant, -8192, 8191) << dshift) - *decoded; |
| 480 | *decoded++ = v; |
| 481 | s->residues[filter_order + i] = v >> dshift; |
| 482 | } |
| 483 | |
| 484 | x -= tmp; |
| 485 | if (x > 0) |
| 486 | memcpy(s->residues, &s->residues[y], 2 * filter_order); |
| 487 | } |
| 488 | |
| 489 | emms_c(); |
| 490 | |
| 491 | return 0; |
| 492 | } |
| 493 | |
| 494 | static int decode_channel(TAKDecContext *s, int chan) |
| 495 | { |
| 496 | AVCodecContext *avctx = s->avctx; |
| 497 | GetBitContext *gb = &s->gb; |
| 498 | int32_t *decoded = s->decoded[chan]; |
| 499 | int left = s->nb_samples - 1; |
| 500 | int i = 0, ret, prev = 0; |
| 501 | |
| 502 | s->sample_shift[chan] = get_bits_esc4(gb); |
| 503 | if (s->sample_shift[chan] >= avctx->bits_per_raw_sample) |
| 504 | return AVERROR_INVALIDDATA; |
| 505 | |
| 506 | *decoded++ = get_sbits(gb, avctx->bits_per_raw_sample - s->sample_shift[chan]); |
| 507 | s->lpc_mode[chan] = get_bits(gb, 2); |
| 508 | s->nb_subframes = get_bits(gb, 3) + 1; |
| 509 | |
| 510 | if (s->nb_subframes > 1) { |
| 511 | if (get_bits_left(gb) < (s->nb_subframes - 1) * 6) |
| 512 | return AVERROR_INVALIDDATA; |
| 513 | |
| 514 | for (; i < s->nb_subframes - 1; i++) { |
| 515 | int v = get_bits(gb, 6); |
| 516 | |
| 517 | s->subframe_len[i] = (v - prev) * s->subframe_scale; |
| 518 | if (s->subframe_len[i] <= 0) |
| 519 | return AVERROR_INVALIDDATA; |
| 520 | |
| 521 | left -= s->subframe_len[i]; |
| 522 | prev = v; |
| 523 | } |
| 524 | |
| 525 | if (left <= 0) |
| 526 | return AVERROR_INVALIDDATA; |
| 527 | } |
| 528 | s->subframe_len[i] = left; |
| 529 | |
| 530 | prev = 0; |
| 531 | for (i = 0; i < s->nb_subframes; i++) { |
| 532 | if ((ret = decode_subframe(s, decoded, s->subframe_len[i], prev)) < 0) |
| 533 | return ret; |
| 534 | decoded += s->subframe_len[i]; |
| 535 | prev = s->subframe_len[i]; |
| 536 | } |
| 537 | |
| 538 | return 0; |
| 539 | } |
| 540 | |
| 541 | static int decorrelate(TAKDecContext *s, int c1, int c2, int length) |
| 542 | { |
| 543 | GetBitContext *gb = &s->gb; |
| 544 | int32_t *p1 = s->decoded[c1] + 1; |
| 545 | int32_t *p2 = s->decoded[c2] + 1; |
| 546 | int i; |
| 547 | int dshift, dfactor; |
| 548 | |
| 549 | switch (s->dmode) { |
| 550 | case 1: /* left/side */ |
| 551 | for (i = 0; i < length; i++) { |
| 552 | int32_t a = p1[i]; |
| 553 | int32_t b = p2[i]; |
| 554 | p2[i] = a + b; |
| 555 | } |
| 556 | break; |
| 557 | case 2: /* side/right */ |
| 558 | for (i = 0; i < length; i++) { |
| 559 | int32_t a = p1[i]; |
| 560 | int32_t b = p2[i]; |
| 561 | p1[i] = b - a; |
| 562 | } |
| 563 | break; |
| 564 | case 3: /* side/mid */ |
| 565 | for (i = 0; i < length; i++) { |
| 566 | int32_t a = p1[i]; |
| 567 | int32_t b = p2[i]; |
| 568 | a -= b >> 1; |
| 569 | p1[i] = a; |
| 570 | p2[i] = a + b; |
| 571 | } |
| 572 | break; |
| 573 | case 4: /* side/left with scale factor */ |
| 574 | FFSWAP(int32_t*, p1, p2); |
| 575 | case 5: /* side/right with scale factor */ |
| 576 | dshift = get_bits_esc4(gb); |
| 577 | dfactor = get_sbits(gb, 10); |
| 578 | for (i = 0; i < length; i++) { |
| 579 | int32_t a = p1[i]; |
| 580 | int32_t b = p2[i]; |
| 581 | b = dfactor * (b >> dshift) + 128 >> 8 << dshift; |
| 582 | p1[i] = b - a; |
| 583 | } |
| 584 | break; |
| 585 | case 6: |
| 586 | FFSWAP(int32_t*, p1, p2); |
| 587 | case 7: { |
| 588 | int length2, order_half, filter_order, dval1, dval2; |
| 589 | int tmp, x, code_size; |
| 590 | |
| 591 | if (length < 256) |
| 592 | return AVERROR_INVALIDDATA; |
| 593 | |
| 594 | dshift = get_bits_esc4(gb); |
| 595 | filter_order = 8 << get_bits1(gb); |
| 596 | dval1 = get_bits1(gb); |
| 597 | dval2 = get_bits1(gb); |
| 598 | |
| 599 | for (i = 0; i < filter_order; i++) { |
| 600 | if (!(i & 3)) |
| 601 | code_size = 14 - get_bits(gb, 3); |
| 602 | s->filter[i] = get_sbits(gb, code_size); |
| 603 | } |
| 604 | |
| 605 | order_half = filter_order / 2; |
| 606 | length2 = length - (filter_order - 1); |
| 607 | |
| 608 | /* decorrelate beginning samples */ |
| 609 | if (dval1) { |
| 610 | for (i = 0; i < order_half; i++) { |
| 611 | int32_t a = p1[i]; |
| 612 | int32_t b = p2[i]; |
| 613 | p1[i] = a + b; |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | /* decorrelate ending samples */ |
| 618 | if (dval2) { |
| 619 | for (i = length2 + order_half; i < length; i++) { |
| 620 | int32_t a = p1[i]; |
| 621 | int32_t b = p2[i]; |
| 622 | p1[i] = a + b; |
| 623 | } |
| 624 | } |
| 625 | |
| 626 | |
| 627 | for (i = 0; i < filter_order; i++) |
| 628 | s->residues[i] = *p2++ >> dshift; |
| 629 | |
| 630 | p1 += order_half; |
| 631 | x = FF_ARRAY_ELEMS(s->residues) - filter_order; |
| 632 | for (; length2 > 0; length2 -= tmp) { |
| 633 | tmp = FFMIN(length2, x); |
| 634 | |
| 635 | for (i = 0; i < tmp; i++) |
| 636 | s->residues[filter_order + i] = *p2++ >> dshift; |
| 637 | |
| 638 | for (i = 0; i < tmp; i++) { |
| 639 | int v = 1 << 9; |
| 640 | |
| 641 | if (filter_order == 16) { |
| 642 | v += s->adsp.scalarproduct_int16(&s->residues[i], s->filter, |
| 643 | filter_order); |
| 644 | } else { |
| 645 | v += s->residues[i + 7] * s->filter[7] + |
| 646 | s->residues[i + 6] * s->filter[6] + |
| 647 | s->residues[i + 5] * s->filter[5] + |
| 648 | s->residues[i + 4] * s->filter[4] + |
| 649 | s->residues[i + 3] * s->filter[3] + |
| 650 | s->residues[i + 2] * s->filter[2] + |
| 651 | s->residues[i + 1] * s->filter[1] + |
| 652 | s->residues[i ] * s->filter[0]; |
| 653 | } |
| 654 | |
| 655 | v = (av_clip(v >> 10, -8192, 8191) << dshift) - *p1; |
| 656 | *p1++ = v; |
| 657 | } |
| 658 | |
| 659 | memcpy(s->residues, &s->residues[tmp], 2 * filter_order); |
| 660 | } |
| 661 | |
| 662 | emms_c(); |
| 663 | break; |
| 664 | } |
| 665 | } |
| 666 | |
| 667 | return 0; |
| 668 | } |
| 669 | |
| 670 | static int tak_decode_frame(AVCodecContext *avctx, void *data, |
| 671 | int *got_frame_ptr, AVPacket *pkt) |
| 672 | { |
| 673 | TAKDecContext *s = avctx->priv_data; |
| 674 | AVFrame *frame = data; |
| 675 | ThreadFrame tframe = { .f = data }; |
| 676 | GetBitContext *gb = &s->gb; |
| 677 | int chan, i, ret, hsize; |
| 678 | |
| 679 | if (pkt->size < TAK_MIN_FRAME_HEADER_BYTES) |
| 680 | return AVERROR_INVALIDDATA; |
| 681 | |
| 682 | if ((ret = init_get_bits8(gb, pkt->data, pkt->size)) < 0) |
| 683 | return ret; |
| 684 | |
| 685 | if ((ret = ff_tak_decode_frame_header(avctx, gb, &s->ti, 0)) < 0) |
| 686 | return ret; |
| 687 | |
| 688 | hsize = get_bits_count(gb) / 8; |
| 689 | if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_COMPLIANT)) { |
| 690 | if (ff_tak_check_crc(pkt->data, hsize)) { |
| 691 | av_log(avctx, AV_LOG_ERROR, "CRC error\n"); |
| 692 | if (avctx->err_recognition & AV_EF_EXPLODE) |
| 693 | return AVERROR_INVALIDDATA; |
| 694 | } |
| 695 | } |
| 696 | |
| 697 | if (s->ti.codec != TAK_CODEC_MONO_STEREO && |
| 698 | s->ti.codec != TAK_CODEC_MULTICHANNEL) { |
| 699 | av_log(avctx, AV_LOG_ERROR, "unsupported codec: %d\n", s->ti.codec); |
| 700 | return AVERROR_PATCHWELCOME; |
| 701 | } |
| 702 | if (s->ti.data_type) { |
| 703 | av_log(avctx, AV_LOG_ERROR, |
| 704 | "unsupported data type: %d\n", s->ti.data_type); |
| 705 | return AVERROR_INVALIDDATA; |
| 706 | } |
| 707 | if (s->ti.codec == TAK_CODEC_MONO_STEREO && s->ti.channels > 2) { |
| 708 | av_log(avctx, AV_LOG_ERROR, |
| 709 | "invalid number of channels: %d\n", s->ti.channels); |
| 710 | return AVERROR_INVALIDDATA; |
| 711 | } |
| 712 | if (s->ti.channels > 6) { |
| 713 | av_log(avctx, AV_LOG_ERROR, |
| 714 | "unsupported number of channels: %d\n", s->ti.channels); |
| 715 | return AVERROR_INVALIDDATA; |
| 716 | } |
| 717 | |
| 718 | if (s->ti.frame_samples <= 0) { |
| 719 | av_log(avctx, AV_LOG_ERROR, "unsupported/invalid number of samples\n"); |
| 720 | return AVERROR_INVALIDDATA; |
| 721 | } |
| 722 | |
| 723 | avctx->bits_per_raw_sample = s->ti.bps; |
| 724 | if ((ret = set_bps_params(avctx)) < 0) |
| 725 | return ret; |
| 726 | if (s->ti.sample_rate != avctx->sample_rate) { |
| 727 | avctx->sample_rate = s->ti.sample_rate; |
| 728 | set_sample_rate_params(avctx); |
| 729 | } |
| 730 | if (s->ti.ch_layout) |
| 731 | avctx->channel_layout = s->ti.ch_layout; |
| 732 | avctx->channels = s->ti.channels; |
| 733 | |
| 734 | s->nb_samples = s->ti.last_frame_samples ? s->ti.last_frame_samples |
| 735 | : s->ti.frame_samples; |
| 736 | |
| 737 | frame->nb_samples = s->nb_samples; |
| 738 | if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0) |
| 739 | return ret; |
| 740 | ff_thread_finish_setup(avctx); |
| 741 | |
| 742 | if (avctx->bits_per_raw_sample <= 16) { |
| 743 | int buf_size = av_samples_get_buffer_size(NULL, avctx->channels, |
| 744 | s->nb_samples, |
| 745 | AV_SAMPLE_FMT_S32P, 0); |
| 746 | av_fast_malloc(&s->decode_buffer, &s->decode_buffer_size, buf_size); |
| 747 | if (!s->decode_buffer) |
| 748 | return AVERROR(ENOMEM); |
| 749 | ret = av_samples_fill_arrays((uint8_t **)s->decoded, NULL, |
| 750 | s->decode_buffer, avctx->channels, |
| 751 | s->nb_samples, AV_SAMPLE_FMT_S32P, 0); |
| 752 | if (ret < 0) |
| 753 | return ret; |
| 754 | } else { |
| 755 | for (chan = 0; chan < avctx->channels; chan++) |
| 756 | s->decoded[chan] = (int32_t *)frame->extended_data[chan]; |
| 757 | } |
| 758 | |
| 759 | if (s->nb_samples < 16) { |
| 760 | for (chan = 0; chan < avctx->channels; chan++) { |
| 761 | int32_t *decoded = s->decoded[chan]; |
| 762 | for (i = 0; i < s->nb_samples; i++) |
| 763 | decoded[i] = get_sbits(gb, avctx->bits_per_raw_sample); |
| 764 | } |
| 765 | } else { |
| 766 | if (s->ti.codec == TAK_CODEC_MONO_STEREO) { |
| 767 | for (chan = 0; chan < avctx->channels; chan++) |
| 768 | if (ret = decode_channel(s, chan)) |
| 769 | return ret; |
| 770 | |
| 771 | if (avctx->channels == 2) { |
| 772 | s->nb_subframes = get_bits(gb, 1) + 1; |
| 773 | if (s->nb_subframes > 1) { |
| 774 | s->subframe_len[1] = get_bits(gb, 6); |
| 775 | } |
| 776 | |
| 777 | s->dmode = get_bits(gb, 3); |
| 778 | if (ret = decorrelate(s, 0, 1, s->nb_samples - 1)) |
| 779 | return ret; |
| 780 | } |
| 781 | } else if (s->ti.codec == TAK_CODEC_MULTICHANNEL) { |
| 782 | if (get_bits1(gb)) { |
| 783 | int ch_mask = 0; |
| 784 | |
| 785 | chan = get_bits(gb, 4) + 1; |
| 786 | if (chan > avctx->channels) |
| 787 | return AVERROR_INVALIDDATA; |
| 788 | |
| 789 | for (i = 0; i < chan; i++) { |
| 790 | int nbit = get_bits(gb, 4); |
| 791 | |
| 792 | if (nbit >= avctx->channels) |
| 793 | return AVERROR_INVALIDDATA; |
| 794 | |
| 795 | if (ch_mask & 1 << nbit) |
| 796 | return AVERROR_INVALIDDATA; |
| 797 | |
| 798 | s->mcdparams[i].present = get_bits1(gb); |
| 799 | if (s->mcdparams[i].present) { |
| 800 | s->mcdparams[i].index = get_bits(gb, 2); |
| 801 | s->mcdparams[i].chan2 = get_bits(gb, 4); |
| 802 | if (s->mcdparams[i].index == 1) { |
| 803 | if ((nbit == s->mcdparams[i].chan2) || |
| 804 | (ch_mask & 1 << s->mcdparams[i].chan2)) |
| 805 | return AVERROR_INVALIDDATA; |
| 806 | |
| 807 | ch_mask |= 1 << s->mcdparams[i].chan2; |
| 808 | } else if (!(ch_mask & 1 << s->mcdparams[i].chan2)) { |
| 809 | return AVERROR_INVALIDDATA; |
| 810 | } |
| 811 | } |
| 812 | s->mcdparams[i].chan1 = nbit; |
| 813 | |
| 814 | ch_mask |= 1 << nbit; |
| 815 | } |
| 816 | } else { |
| 817 | chan = avctx->channels; |
| 818 | for (i = 0; i < chan; i++) { |
| 819 | s->mcdparams[i].present = 0; |
| 820 | s->mcdparams[i].chan1 = i; |
| 821 | } |
| 822 | } |
| 823 | |
| 824 | for (i = 0; i < chan; i++) { |
| 825 | if (s->mcdparams[i].present && s->mcdparams[i].index == 1) |
| 826 | if (ret = decode_channel(s, s->mcdparams[i].chan2)) |
| 827 | return ret; |
| 828 | |
| 829 | if (ret = decode_channel(s, s->mcdparams[i].chan1)) |
| 830 | return ret; |
| 831 | |
| 832 | if (s->mcdparams[i].present) { |
| 833 | s->dmode = mc_dmodes[s->mcdparams[i].index]; |
| 834 | if (ret = decorrelate(s, |
| 835 | s->mcdparams[i].chan2, |
| 836 | s->mcdparams[i].chan1, |
| 837 | s->nb_samples - 1)) |
| 838 | return ret; |
| 839 | } |
| 840 | } |
| 841 | } |
| 842 | |
| 843 | for (chan = 0; chan < avctx->channels; chan++) { |
| 844 | int32_t *decoded = s->decoded[chan]; |
| 845 | |
| 846 | if (s->lpc_mode[chan]) |
| 847 | decode_lpc(decoded, s->lpc_mode[chan], s->nb_samples); |
| 848 | |
| 849 | if (s->sample_shift[chan] > 0) |
| 850 | for (i = 0; i < s->nb_samples; i++) |
| 851 | decoded[i] <<= s->sample_shift[chan]; |
| 852 | } |
| 853 | } |
| 854 | |
| 855 | align_get_bits(gb); |
| 856 | skip_bits(gb, 24); |
| 857 | if (get_bits_left(gb) < 0) |
| 858 | av_log(avctx, AV_LOG_DEBUG, "overread\n"); |
| 859 | else if (get_bits_left(gb) > 0) |
| 860 | av_log(avctx, AV_LOG_DEBUG, "underread\n"); |
| 861 | |
| 862 | if (avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_COMPLIANT)) { |
| 863 | if (ff_tak_check_crc(pkt->data + hsize, |
| 864 | get_bits_count(gb) / 8 - hsize)) { |
| 865 | av_log(avctx, AV_LOG_ERROR, "CRC error\n"); |
| 866 | if (avctx->err_recognition & AV_EF_EXPLODE) |
| 867 | return AVERROR_INVALIDDATA; |
| 868 | } |
| 869 | } |
| 870 | |
| 871 | /* convert to output buffer */ |
| 872 | switch (avctx->sample_fmt) { |
| 873 | case AV_SAMPLE_FMT_U8P: |
| 874 | for (chan = 0; chan < avctx->channels; chan++) { |
| 875 | uint8_t *samples = (uint8_t *)frame->extended_data[chan]; |
| 876 | int32_t *decoded = s->decoded[chan]; |
| 877 | for (i = 0; i < s->nb_samples; i++) |
| 878 | samples[i] = decoded[i] + 0x80; |
| 879 | } |
| 880 | break; |
| 881 | case AV_SAMPLE_FMT_S16P: |
| 882 | for (chan = 0; chan < avctx->channels; chan++) { |
| 883 | int16_t *samples = (int16_t *)frame->extended_data[chan]; |
| 884 | int32_t *decoded = s->decoded[chan]; |
| 885 | for (i = 0; i < s->nb_samples; i++) |
| 886 | samples[i] = decoded[i]; |
| 887 | } |
| 888 | break; |
| 889 | case AV_SAMPLE_FMT_S32P: |
| 890 | for (chan = 0; chan < avctx->channels; chan++) { |
| 891 | int32_t *samples = (int32_t *)frame->extended_data[chan]; |
| 892 | for (i = 0; i < s->nb_samples; i++) |
| 893 | samples[i] <<= 8; |
| 894 | } |
| 895 | break; |
| 896 | } |
| 897 | |
| 898 | *got_frame_ptr = 1; |
| 899 | |
| 900 | return pkt->size; |
| 901 | } |
| 902 | |
| 903 | static int init_thread_copy(AVCodecContext *avctx) |
| 904 | { |
| 905 | TAKDecContext *s = avctx->priv_data; |
| 906 | s->avctx = avctx; |
| 907 | return 0; |
| 908 | } |
| 909 | |
| 910 | static int update_thread_context(AVCodecContext *dst, |
| 911 | const AVCodecContext *src) |
| 912 | { |
| 913 | TAKDecContext *tsrc = src->priv_data; |
| 914 | TAKDecContext *tdst = dst->priv_data; |
| 915 | |
| 916 | if (dst == src) |
| 917 | return 0; |
| 918 | memcpy(&tdst->ti, &tsrc->ti, sizeof(TAKStreamInfo)); |
| 919 | return 0; |
| 920 | } |
| 921 | |
| 922 | static av_cold int tak_decode_close(AVCodecContext *avctx) |
| 923 | { |
| 924 | TAKDecContext *s = avctx->priv_data; |
| 925 | |
| 926 | av_freep(&s->decode_buffer); |
| 927 | |
| 928 | return 0; |
| 929 | } |
| 930 | |
| 931 | AVCodec ff_tak_decoder = { |
| 932 | .name = "tak", |
| 933 | .long_name = NULL_IF_CONFIG_SMALL("TAK (Tom's lossless Audio Kompressor)"), |
| 934 | .type = AVMEDIA_TYPE_AUDIO, |
| 935 | .id = AV_CODEC_ID_TAK, |
| 936 | .priv_data_size = sizeof(TAKDecContext), |
| 937 | .init = tak_decode_init, |
| 938 | .close = tak_decode_close, |
| 939 | .decode = tak_decode_frame, |
| 940 | .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy), |
| 941 | .update_thread_context = ONLY_IF_THREADS_ENABLED(update_thread_context), |
| 942 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS, |
| 943 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P, |
| 944 | AV_SAMPLE_FMT_S16P, |
| 945 | AV_SAMPLE_FMT_S32P, |
| 946 | AV_SAMPLE_FMT_NONE }, |
| 947 | }; |