ffmpeg / libavcodec / imc.c @ 9106a698
History | View | Annotate | Download (24.7 KB)
| 1 |
/*
|
|---|---|
| 2 |
* IMC compatible decoder
|
| 3 |
* Copyright (c) 2002-2004 Maxim Poliakovski
|
| 4 |
* Copyright (c) 2006 Benjamin Larsson
|
| 5 |
* Copyright (c) 2006 Konstantin Shishkov
|
| 6 |
*
|
| 7 |
* This file is part of FFmpeg.
|
| 8 |
*
|
| 9 |
* FFmpeg is free software; you can redistribute it and/or
|
| 10 |
* modify it under the terms of the GNU Lesser General Public
|
| 11 |
* License as published by the Free Software Foundation; either
|
| 12 |
* version 2.1 of the License, or (at your option) any later version.
|
| 13 |
*
|
| 14 |
* FFmpeg is distributed in the hope that it will be useful,
|
| 15 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 16 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 17 |
* Lesser General Public License for more details.
|
| 18 |
*
|
| 19 |
* You should have received a copy of the GNU Lesser General Public
|
| 20 |
* License along with FFmpeg; if not, write to the Free Software
|
| 21 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
| 22 |
*/
|
| 23 |
|
| 24 |
/**
|
| 25 |
* @file libavcodec/imc.c IMC - Intel Music Coder
|
| 26 |
* A mdct based codec using a 256 points large transform
|
| 27 |
* divied into 32 bands with some mix of scale factors.
|
| 28 |
* Only mono is supported.
|
| 29 |
*
|
| 30 |
*/
|
| 31 |
|
| 32 |
|
| 33 |
#include <math.h> |
| 34 |
#include <stddef.h> |
| 35 |
#include <stdio.h> |
| 36 |
|
| 37 |
#define ALT_BITSTREAM_READER
|
| 38 |
#include "avcodec.h" |
| 39 |
#include "get_bits.h" |
| 40 |
#include "dsputil.h" |
| 41 |
|
| 42 |
#include "imcdata.h" |
| 43 |
|
| 44 |
#define IMC_BLOCK_SIZE 64 |
| 45 |
#define IMC_FRAME_ID 0x21 |
| 46 |
#define BANDS 32 |
| 47 |
#define COEFFS 256 |
| 48 |
|
| 49 |
typedef struct { |
| 50 |
float old_floor[BANDS];
|
| 51 |
float flcoeffs1[BANDS];
|
| 52 |
float flcoeffs2[BANDS];
|
| 53 |
float flcoeffs3[BANDS];
|
| 54 |
float flcoeffs4[BANDS];
|
| 55 |
float flcoeffs5[BANDS];
|
| 56 |
float flcoeffs6[BANDS];
|
| 57 |
float CWdecoded[COEFFS];
|
| 58 |
|
| 59 |
/** MDCT tables */
|
| 60 |
//@{
|
| 61 |
float mdct_sine_window[COEFFS];
|
| 62 |
float post_cos[COEFFS];
|
| 63 |
float post_sin[COEFFS];
|
| 64 |
float pre_coef1[COEFFS];
|
| 65 |
float pre_coef2[COEFFS];
|
| 66 |
float last_fft_im[COEFFS];
|
| 67 |
//@}
|
| 68 |
|
| 69 |
int bandWidthT[BANDS]; ///< codewords per band |
| 70 |
int bitsBandT[BANDS]; ///< how many bits per codeword in band |
| 71 |
int CWlengthT[COEFFS]; ///< how many bits in each codeword |
| 72 |
int levlCoeffBuf[BANDS];
|
| 73 |
int bandFlagsBuf[BANDS]; ///< flags for each band |
| 74 |
int sumLenArr[BANDS]; ///< bits for all coeffs in band |
| 75 |
int skipFlagRaw[BANDS]; ///< skip flags are stored in raw form or not |
| 76 |
int skipFlagBits[BANDS]; ///< bits used to code skip flags |
| 77 |
int skipFlagCount[BANDS]; ///< skipped coeffients per band |
| 78 |
int skipFlags[COEFFS]; ///< skip coefficient decoding or not |
| 79 |
int codewords[COEFFS]; ///< raw codewords read from bitstream |
| 80 |
float sqrt_tab[30]; |
| 81 |
GetBitContext gb; |
| 82 |
int decoder_reset;
|
| 83 |
float one_div_log2;
|
| 84 |
|
| 85 |
DSPContext dsp; |
| 86 |
FFTContext fft; |
| 87 |
DECLARE_ALIGNED_16(FFTComplex, samples[COEFFS/2]);
|
| 88 |
DECLARE_ALIGNED_16(float, out_samples[COEFFS]);
|
| 89 |
} IMCContext; |
| 90 |
|
| 91 |
static VLC huffman_vlc[4][4]; |
| 92 |
|
| 93 |
#define VLC_TABLES_SIZE 9512 |
| 94 |
|
| 95 |
static const int vlc_offsets[17] = { |
| 96 |
0, 640, 1156, 1732, 2308, 2852, 3396, 3924, |
| 97 |
4452, 5220, 5860, 6628, 7268, 7908, 8424, 8936, VLC_TABLES_SIZE}; |
| 98 |
|
| 99 |
static VLC_TYPE vlc_tables[VLC_TABLES_SIZE][2]; |
| 100 |
|
| 101 |
static av_cold int imc_decode_init(AVCodecContext * avctx) |
| 102 |
{
|
| 103 |
int i, j;
|
| 104 |
IMCContext *q = avctx->priv_data; |
| 105 |
double r1, r2;
|
| 106 |
|
| 107 |
q->decoder_reset = 1;
|
| 108 |
|
| 109 |
for(i = 0; i < BANDS; i++) |
| 110 |
q->old_floor[i] = 1.0; |
| 111 |
|
| 112 |
/* Build mdct window, a simple sine window normalized with sqrt(2) */
|
| 113 |
ff_sine_window_init(q->mdct_sine_window, COEFFS); |
| 114 |
for(i = 0; i < COEFFS; i++) |
| 115 |
q->mdct_sine_window[i] *= sqrt(2.0); |
| 116 |
for(i = 0; i < COEFFS/2; i++){ |
| 117 |
q->post_cos[i] = cos(i / 256.0 * M_PI); |
| 118 |
q->post_sin[i] = sin(i / 256.0 * M_PI); |
| 119 |
|
| 120 |
r1 = sin((i * 4.0 + 1.0) / 1024.0 * M_PI); |
| 121 |
r2 = cos((i * 4.0 + 1.0) / 1024.0 * M_PI); |
| 122 |
|
| 123 |
if (i & 0x1) |
| 124 |
{
|
| 125 |
q->pre_coef1[i] = (r1 + r2) * sqrt(2.0); |
| 126 |
q->pre_coef2[i] = -(r1 - r2) * sqrt(2.0); |
| 127 |
} |
| 128 |
else
|
| 129 |
{
|
| 130 |
q->pre_coef1[i] = -(r1 + r2) * sqrt(2.0); |
| 131 |
q->pre_coef2[i] = (r1 - r2) * sqrt(2.0); |
| 132 |
} |
| 133 |
|
| 134 |
q->last_fft_im[i] = 0;
|
| 135 |
} |
| 136 |
|
| 137 |
/* Generate a square root table */
|
| 138 |
|
| 139 |
for(i = 0; i < 30; i++) { |
| 140 |
q->sqrt_tab[i] = sqrt(i); |
| 141 |
} |
| 142 |
|
| 143 |
/* initialize the VLC tables */
|
| 144 |
for(i = 0; i < 4 ; i++) { |
| 145 |
for(j = 0; j < 4; j++) { |
| 146 |
huffman_vlc[i][j].table = &vlc_tables[vlc_offsets[i * 4 + j]];
|
| 147 |
huffman_vlc[i][j].table_allocated = vlc_offsets[i * 4 + j + 1] - vlc_offsets[i * 4 + j]; |
| 148 |
init_vlc(&huffman_vlc[i][j], 9, imc_huffman_sizes[i],
|
| 149 |
imc_huffman_lens[i][j], 1, 1, |
| 150 |
imc_huffman_bits[i][j], 2, 2, INIT_VLC_USE_NEW_STATIC); |
| 151 |
} |
| 152 |
} |
| 153 |
q->one_div_log2 = 1/log(2); |
| 154 |
|
| 155 |
ff_fft_init(&q->fft, 7, 1); |
| 156 |
dsputil_init(&q->dsp, avctx); |
| 157 |
avctx->sample_fmt = SAMPLE_FMT_S16; |
| 158 |
avctx->channel_layout = (avctx->channels==2) ? CH_LAYOUT_STEREO : CH_LAYOUT_MONO;
|
| 159 |
return 0; |
| 160 |
} |
| 161 |
|
| 162 |
static void imc_calculate_coeffs(IMCContext* q, float* flcoeffs1, float* flcoeffs2, int* bandWidthT, |
| 163 |
float* flcoeffs3, float* flcoeffs5) |
| 164 |
{
|
| 165 |
float workT1[BANDS];
|
| 166 |
float workT2[BANDS];
|
| 167 |
float workT3[BANDS];
|
| 168 |
float snr_limit = 1.e-30; |
| 169 |
float accum = 0.0; |
| 170 |
int i, cnt2;
|
| 171 |
|
| 172 |
for(i = 0; i < BANDS; i++) { |
| 173 |
flcoeffs5[i] = workT2[i] = 0.0; |
| 174 |
if (bandWidthT[i]){
|
| 175 |
workT1[i] = flcoeffs1[i] * flcoeffs1[i]; |
| 176 |
flcoeffs3[i] = 2.0 * flcoeffs2[i]; |
| 177 |
} else {
|
| 178 |
workT1[i] = 0.0; |
| 179 |
flcoeffs3[i] = -30000.0; |
| 180 |
} |
| 181 |
workT3[i] = bandWidthT[i] * workT1[i] * 0.01; |
| 182 |
if (workT3[i] <= snr_limit)
|
| 183 |
workT3[i] = 0.0; |
| 184 |
} |
| 185 |
|
| 186 |
for(i = 0; i < BANDS; i++) { |
| 187 |
for(cnt2 = i; cnt2 < cyclTab[i]; cnt2++)
|
| 188 |
flcoeffs5[cnt2] = flcoeffs5[cnt2] + workT3[i]; |
| 189 |
workT2[cnt2-1] = workT2[cnt2-1] + workT3[i]; |
| 190 |
} |
| 191 |
|
| 192 |
for(i = 1; i < BANDS; i++) { |
| 193 |
accum = (workT2[i-1] + accum) * imc_weights1[i-1]; |
| 194 |
flcoeffs5[i] += accum; |
| 195 |
} |
| 196 |
|
| 197 |
for(i = 0; i < BANDS; i++) |
| 198 |
workT2[i] = 0.0; |
| 199 |
|
| 200 |
for(i = 0; i < BANDS; i++) { |
| 201 |
for(cnt2 = i-1; cnt2 > cyclTab2[i]; cnt2--) |
| 202 |
flcoeffs5[cnt2] += workT3[i]; |
| 203 |
workT2[cnt2+1] += workT3[i];
|
| 204 |
} |
| 205 |
|
| 206 |
accum = 0.0; |
| 207 |
|
| 208 |
for(i = BANDS-2; i >= 0; i--) { |
| 209 |
accum = (workT2[i+1] + accum) * imc_weights2[i];
|
| 210 |
flcoeffs5[i] += accum; |
| 211 |
//there is missing code here, but it seems to never be triggered
|
| 212 |
} |
| 213 |
} |
| 214 |
|
| 215 |
|
| 216 |
static void imc_read_level_coeffs(IMCContext* q, int stream_format_code, int* levlCoeffs) |
| 217 |
{
|
| 218 |
int i;
|
| 219 |
VLC *hufftab[4];
|
| 220 |
int start = 0; |
| 221 |
const uint8_t *cb_sel;
|
| 222 |
int s;
|
| 223 |
|
| 224 |
s = stream_format_code >> 1;
|
| 225 |
hufftab[0] = &huffman_vlc[s][0]; |
| 226 |
hufftab[1] = &huffman_vlc[s][1]; |
| 227 |
hufftab[2] = &huffman_vlc[s][2]; |
| 228 |
hufftab[3] = &huffman_vlc[s][3]; |
| 229 |
cb_sel = imc_cb_select[s]; |
| 230 |
|
| 231 |
if(stream_format_code & 4) |
| 232 |
start = 1;
|
| 233 |
if(start)
|
| 234 |
levlCoeffs[0] = get_bits(&q->gb, 7); |
| 235 |
for(i = start; i < BANDS; i++){
|
| 236 |
levlCoeffs[i] = get_vlc2(&q->gb, hufftab[cb_sel[i]]->table, hufftab[cb_sel[i]]->bits, 2);
|
| 237 |
if(levlCoeffs[i] == 17) |
| 238 |
levlCoeffs[i] += get_bits(&q->gb, 4);
|
| 239 |
} |
| 240 |
} |
| 241 |
|
| 242 |
static void imc_decode_level_coefficients(IMCContext* q, int* levlCoeffBuf, float* flcoeffs1, |
| 243 |
float* flcoeffs2)
|
| 244 |
{
|
| 245 |
int i, level;
|
| 246 |
float tmp, tmp2;
|
| 247 |
//maybe some frequency division thingy
|
| 248 |
|
| 249 |
flcoeffs1[0] = 20000.0 / pow (2, levlCoeffBuf[0] * 0.18945); // 0.18945 = log2(10) * 0.05703125 |
| 250 |
flcoeffs2[0] = log(flcoeffs1[0])/log(2); |
| 251 |
tmp = flcoeffs1[0];
|
| 252 |
tmp2 = flcoeffs2[0];
|
| 253 |
|
| 254 |
for(i = 1; i < BANDS; i++) { |
| 255 |
level = levlCoeffBuf[i]; |
| 256 |
if (level == 16) { |
| 257 |
flcoeffs1[i] = 1.0; |
| 258 |
flcoeffs2[i] = 0.0; |
| 259 |
} else {
|
| 260 |
if (level < 17) |
| 261 |
level -=7;
|
| 262 |
else if (level <= 24) |
| 263 |
level -=32;
|
| 264 |
else
|
| 265 |
level -=16;
|
| 266 |
|
| 267 |
tmp *= imc_exp_tab[15 + level];
|
| 268 |
tmp2 += 0.83048 * level; // 0.83048 = log2(10) * 0.25 |
| 269 |
flcoeffs1[i] = tmp; |
| 270 |
flcoeffs2[i] = tmp2; |
| 271 |
} |
| 272 |
} |
| 273 |
} |
| 274 |
|
| 275 |
|
| 276 |
static void imc_decode_level_coefficients2(IMCContext* q, int* levlCoeffBuf, float* old_floor, float* flcoeffs1, |
| 277 |
float* flcoeffs2) {
|
| 278 |
int i;
|
| 279 |
//FIXME maybe flag_buf = noise coding and flcoeffs1 = new scale factors
|
| 280 |
// and flcoeffs2 old scale factors
|
| 281 |
// might be incomplete due to a missing table that is in the binary code
|
| 282 |
for(i = 0; i < BANDS; i++) { |
| 283 |
flcoeffs1[i] = 0;
|
| 284 |
if(levlCoeffBuf[i] < 16) { |
| 285 |
flcoeffs1[i] = imc_exp_tab2[levlCoeffBuf[i]] * old_floor[i]; |
| 286 |
flcoeffs2[i] = (levlCoeffBuf[i]-7) * 0.83048 + flcoeffs2[i]; // 0.83048 = log2(10) * 0.25 |
| 287 |
} else {
|
| 288 |
flcoeffs1[i] = old_floor[i]; |
| 289 |
} |
| 290 |
} |
| 291 |
} |
| 292 |
|
| 293 |
/**
|
| 294 |
* Perform bit allocation depending on bits available
|
| 295 |
*/
|
| 296 |
static int bit_allocation (IMCContext* q, int stream_format_code, int freebits, int flag) { |
| 297 |
int i, j;
|
| 298 |
const float limit = -1.e20; |
| 299 |
float highest = 0.0; |
| 300 |
int indx;
|
| 301 |
int t1 = 0; |
| 302 |
int t2 = 1; |
| 303 |
float summa = 0.0; |
| 304 |
int iacc = 0; |
| 305 |
int summer = 0; |
| 306 |
int rres, cwlen;
|
| 307 |
float lowest = 1.e10; |
| 308 |
int low_indx = 0; |
| 309 |
float workT[32]; |
| 310 |
int flg;
|
| 311 |
int found_indx = 0; |
| 312 |
|
| 313 |
for(i = 0; i < BANDS; i++) |
| 314 |
highest = FFMAX(highest, q->flcoeffs1[i]); |
| 315 |
|
| 316 |
for(i = 0; i < BANDS-1; i++) { |
| 317 |
q->flcoeffs4[i] = q->flcoeffs3[i] - log(q->flcoeffs5[i])/log(2);
|
| 318 |
} |
| 319 |
q->flcoeffs4[BANDS - 1] = limit;
|
| 320 |
|
| 321 |
highest = highest * 0.25; |
| 322 |
|
| 323 |
for(i = 0; i < BANDS; i++) { |
| 324 |
indx = -1;
|
| 325 |
if ((band_tab[i+1] - band_tab[i]) == q->bandWidthT[i]) |
| 326 |
indx = 0;
|
| 327 |
|
| 328 |
if ((band_tab[i+1] - band_tab[i]) > q->bandWidthT[i]) |
| 329 |
indx = 1;
|
| 330 |
|
| 331 |
if (((band_tab[i+1] - band_tab[i])/2) >= q->bandWidthT[i]) |
| 332 |
indx = 2;
|
| 333 |
|
| 334 |
if (indx == -1) |
| 335 |
return -1; |
| 336 |
|
| 337 |
q->flcoeffs4[i] = q->flcoeffs4[i] + xTab[(indx*2 + (q->flcoeffs1[i] < highest)) * 2 + flag]; |
| 338 |
} |
| 339 |
|
| 340 |
if (stream_format_code & 0x2) { |
| 341 |
q->flcoeffs4[0] = limit;
|
| 342 |
q->flcoeffs4[1] = limit;
|
| 343 |
q->flcoeffs4[2] = limit;
|
| 344 |
q->flcoeffs4[3] = limit;
|
| 345 |
} |
| 346 |
|
| 347 |
for(i = (stream_format_code & 0x2)?4:0; i < BANDS-1; i++) { |
| 348 |
iacc += q->bandWidthT[i]; |
| 349 |
summa += q->bandWidthT[i] * q->flcoeffs4[i]; |
| 350 |
} |
| 351 |
q->bandWidthT[BANDS-1] = 0; |
| 352 |
summa = (summa * 0.5 - freebits) / iacc; |
| 353 |
|
| 354 |
|
| 355 |
for(i = 0; i < BANDS/2; i++) { |
| 356 |
rres = summer - freebits; |
| 357 |
if((rres >= -8) && (rres <= 8)) break; |
| 358 |
|
| 359 |
summer = 0;
|
| 360 |
iacc = 0;
|
| 361 |
|
| 362 |
for(j = (stream_format_code & 0x2)?4:0; j < BANDS; j++) { |
| 363 |
cwlen = av_clip((int)((q->flcoeffs4[j] * 0.5) - summa + 0.5), 0, 6); |
| 364 |
|
| 365 |
q->bitsBandT[j] = cwlen; |
| 366 |
summer += q->bandWidthT[j] * cwlen; |
| 367 |
|
| 368 |
if (cwlen > 0) |
| 369 |
iacc += q->bandWidthT[j]; |
| 370 |
} |
| 371 |
|
| 372 |
flg = t2; |
| 373 |
t2 = 1;
|
| 374 |
if (freebits < summer)
|
| 375 |
t2 = -1;
|
| 376 |
if (i == 0) |
| 377 |
flg = t2; |
| 378 |
if(flg != t2)
|
| 379 |
t1++; |
| 380 |
|
| 381 |
summa = (float)(summer - freebits) / ((t1 + 1) * iacc) + summa; |
| 382 |
} |
| 383 |
|
| 384 |
for(i = (stream_format_code & 0x2)?4:0; i < BANDS; i++) { |
| 385 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) |
| 386 |
q->CWlengthT[j] = q->bitsBandT[i]; |
| 387 |
} |
| 388 |
|
| 389 |
if (freebits > summer) {
|
| 390 |
for(i = 0; i < BANDS; i++) { |
| 391 |
workT[i] = (q->bitsBandT[i] == 6) ? -1.e20 : (q->bitsBandT[i] * -2 + q->flcoeffs4[i] - 0.415); |
| 392 |
} |
| 393 |
|
| 394 |
highest = 0.0; |
| 395 |
|
| 396 |
do{
|
| 397 |
if (highest <= -1.e20) |
| 398 |
break;
|
| 399 |
|
| 400 |
found_indx = 0;
|
| 401 |
highest = -1.e20;
|
| 402 |
|
| 403 |
for(i = 0; i < BANDS; i++) { |
| 404 |
if (workT[i] > highest) {
|
| 405 |
highest = workT[i]; |
| 406 |
found_indx = i; |
| 407 |
} |
| 408 |
} |
| 409 |
|
| 410 |
if (highest > -1.e20) { |
| 411 |
workT[found_indx] -= 2.0; |
| 412 |
if (++(q->bitsBandT[found_indx]) == 6) |
| 413 |
workT[found_indx] = -1.e20;
|
| 414 |
|
| 415 |
for(j = band_tab[found_indx]; j < band_tab[found_indx+1] && (freebits > summer); j++){ |
| 416 |
q->CWlengthT[j]++; |
| 417 |
summer++; |
| 418 |
} |
| 419 |
} |
| 420 |
}while (freebits > summer);
|
| 421 |
} |
| 422 |
if (freebits < summer) {
|
| 423 |
for(i = 0; i < BANDS; i++) { |
| 424 |
workT[i] = q->bitsBandT[i] ? (q->bitsBandT[i] * -2 + q->flcoeffs4[i] + 1.585) : 1.e20; |
| 425 |
} |
| 426 |
if (stream_format_code & 0x2) { |
| 427 |
workT[0] = 1.e20; |
| 428 |
workT[1] = 1.e20; |
| 429 |
workT[2] = 1.e20; |
| 430 |
workT[3] = 1.e20; |
| 431 |
} |
| 432 |
while (freebits < summer){
|
| 433 |
lowest = 1.e10;
|
| 434 |
low_indx = 0;
|
| 435 |
for(i = 0; i < BANDS; i++) { |
| 436 |
if (workT[i] < lowest) {
|
| 437 |
lowest = workT[i]; |
| 438 |
low_indx = i; |
| 439 |
} |
| 440 |
} |
| 441 |
//if(lowest >= 1.e10) break;
|
| 442 |
workT[low_indx] = lowest + 2.0; |
| 443 |
|
| 444 |
if (!(--q->bitsBandT[low_indx]))
|
| 445 |
workT[low_indx] = 1.e20;
|
| 446 |
|
| 447 |
for(j = band_tab[low_indx]; j < band_tab[low_indx+1] && (freebits < summer); j++){ |
| 448 |
if(q->CWlengthT[j] > 0){ |
| 449 |
q->CWlengthT[j]--; |
| 450 |
summer--; |
| 451 |
} |
| 452 |
} |
| 453 |
} |
| 454 |
} |
| 455 |
return 0; |
| 456 |
} |
| 457 |
|
| 458 |
static void imc_get_skip_coeff(IMCContext* q) { |
| 459 |
int i, j;
|
| 460 |
|
| 461 |
memset(q->skipFlagBits, 0, sizeof(q->skipFlagBits)); |
| 462 |
memset(q->skipFlagCount, 0, sizeof(q->skipFlagCount)); |
| 463 |
for(i = 0; i < BANDS; i++) { |
| 464 |
if (!q->bandFlagsBuf[i] || !q->bandWidthT[i])
|
| 465 |
continue;
|
| 466 |
|
| 467 |
if (!q->skipFlagRaw[i]) {
|
| 468 |
q->skipFlagBits[i] = band_tab[i+1] - band_tab[i];
|
| 469 |
|
| 470 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) { |
| 471 |
if ((q->skipFlags[j] = get_bits1(&q->gb)))
|
| 472 |
q->skipFlagCount[i]++; |
| 473 |
} |
| 474 |
} else {
|
| 475 |
for(j = band_tab[i]; j < (band_tab[i+1]-1); j += 2) { |
| 476 |
if(!get_bits1(&q->gb)){//0 |
| 477 |
q->skipFlagBits[i]++; |
| 478 |
q->skipFlags[j]=1;
|
| 479 |
q->skipFlags[j+1]=1; |
| 480 |
q->skipFlagCount[i] += 2;
|
| 481 |
}else{
|
| 482 |
if(get_bits1(&q->gb)){//11 |
| 483 |
q->skipFlagBits[i] +=2;
|
| 484 |
q->skipFlags[j]=0;
|
| 485 |
q->skipFlags[j+1]=1; |
| 486 |
q->skipFlagCount[i]++; |
| 487 |
}else{
|
| 488 |
q->skipFlagBits[i] +=3;
|
| 489 |
q->skipFlags[j+1]=0; |
| 490 |
if(!get_bits1(&q->gb)){//100 |
| 491 |
q->skipFlags[j]=1;
|
| 492 |
q->skipFlagCount[i]++; |
| 493 |
}else{//101 |
| 494 |
q->skipFlags[j]=0;
|
| 495 |
} |
| 496 |
} |
| 497 |
} |
| 498 |
} |
| 499 |
|
| 500 |
if (j < band_tab[i+1]) { |
| 501 |
q->skipFlagBits[i]++; |
| 502 |
if ((q->skipFlags[j] = get_bits1(&q->gb)))
|
| 503 |
q->skipFlagCount[i]++; |
| 504 |
} |
| 505 |
} |
| 506 |
} |
| 507 |
} |
| 508 |
|
| 509 |
/**
|
| 510 |
* Increase highest' band coefficient sizes as some bits won't be used
|
| 511 |
*/
|
| 512 |
static void imc_adjust_bit_allocation (IMCContext* q, int summer) { |
| 513 |
float workT[32]; |
| 514 |
int corrected = 0; |
| 515 |
int i, j;
|
| 516 |
float highest = 0; |
| 517 |
int found_indx=0; |
| 518 |
|
| 519 |
for(i = 0; i < BANDS; i++) { |
| 520 |
workT[i] = (q->bitsBandT[i] == 6) ? -1.e20 : (q->bitsBandT[i] * -2 + q->flcoeffs4[i] - 0.415); |
| 521 |
} |
| 522 |
|
| 523 |
while (corrected < summer) {
|
| 524 |
if(highest <= -1.e20) |
| 525 |
break;
|
| 526 |
|
| 527 |
highest = -1.e20;
|
| 528 |
|
| 529 |
for(i = 0; i < BANDS; i++) { |
| 530 |
if (workT[i] > highest) {
|
| 531 |
highest = workT[i]; |
| 532 |
found_indx = i; |
| 533 |
} |
| 534 |
} |
| 535 |
|
| 536 |
if (highest > -1.e20) { |
| 537 |
workT[found_indx] -= 2.0; |
| 538 |
if (++(q->bitsBandT[found_indx]) == 6) |
| 539 |
workT[found_indx] = -1.e20;
|
| 540 |
|
| 541 |
for(j = band_tab[found_indx]; j < band_tab[found_indx+1] && (corrected < summer); j++) { |
| 542 |
if (!q->skipFlags[j] && (q->CWlengthT[j] < 6)) { |
| 543 |
q->CWlengthT[j]++; |
| 544 |
corrected++; |
| 545 |
} |
| 546 |
} |
| 547 |
} |
| 548 |
} |
| 549 |
} |
| 550 |
|
| 551 |
static void imc_imdct256(IMCContext *q) { |
| 552 |
int i;
|
| 553 |
float re, im;
|
| 554 |
|
| 555 |
/* prerotation */
|
| 556 |
for(i=0; i < COEFFS/2; i++){ |
| 557 |
q->samples[i].re = -(q->pre_coef1[i] * q->CWdecoded[COEFFS-1-i*2]) - |
| 558 |
(q->pre_coef2[i] * q->CWdecoded[i*2]);
|
| 559 |
q->samples[i].im = (q->pre_coef2[i] * q->CWdecoded[COEFFS-1-i*2]) - |
| 560 |
(q->pre_coef1[i] * q->CWdecoded[i*2]);
|
| 561 |
} |
| 562 |
|
| 563 |
/* FFT */
|
| 564 |
ff_fft_permute(&q->fft, q->samples); |
| 565 |
ff_fft_calc (&q->fft, q->samples); |
| 566 |
|
| 567 |
/* postrotation, window and reorder */
|
| 568 |
for(i = 0; i < COEFFS/2; i++){ |
| 569 |
re = (q->samples[i].re * q->post_cos[i]) + (-q->samples[i].im * q->post_sin[i]); |
| 570 |
im = (-q->samples[i].im * q->post_cos[i]) - (q->samples[i].re * q->post_sin[i]); |
| 571 |
q->out_samples[i*2] = (q->mdct_sine_window[COEFFS-1-i*2] * q->last_fft_im[i]) + (q->mdct_sine_window[i*2] * re); |
| 572 |
q->out_samples[COEFFS-1-i*2] = (q->mdct_sine_window[i*2] * q->last_fft_im[i]) - (q->mdct_sine_window[COEFFS-1-i*2] * re); |
| 573 |
q->last_fft_im[i] = im; |
| 574 |
} |
| 575 |
} |
| 576 |
|
| 577 |
static int inverse_quant_coeff (IMCContext* q, int stream_format_code) { |
| 578 |
int i, j;
|
| 579 |
int middle_value, cw_len, max_size;
|
| 580 |
const float* quantizer; |
| 581 |
|
| 582 |
for(i = 0; i < BANDS; i++) { |
| 583 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) { |
| 584 |
q->CWdecoded[j] = 0;
|
| 585 |
cw_len = q->CWlengthT[j]; |
| 586 |
|
| 587 |
if (cw_len <= 0 || q->skipFlags[j]) |
| 588 |
continue;
|
| 589 |
|
| 590 |
max_size = 1 << cw_len;
|
| 591 |
middle_value = max_size >> 1;
|
| 592 |
|
| 593 |
if (q->codewords[j] >= max_size || q->codewords[j] < 0) |
| 594 |
return -1; |
| 595 |
|
| 596 |
if (cw_len >= 4){ |
| 597 |
quantizer = imc_quantizer2[(stream_format_code & 2) >> 1]; |
| 598 |
if (q->codewords[j] >= middle_value)
|
| 599 |
q->CWdecoded[j] = quantizer[q->codewords[j] - 8] * q->flcoeffs6[i];
|
| 600 |
else
|
| 601 |
q->CWdecoded[j] = -quantizer[max_size - q->codewords[j] - 8 - 1] * q->flcoeffs6[i]; |
| 602 |
}else{
|
| 603 |
quantizer = imc_quantizer1[((stream_format_code & 2) >> 1) | (q->bandFlagsBuf[i] << 1)]; |
| 604 |
if (q->codewords[j] >= middle_value)
|
| 605 |
q->CWdecoded[j] = quantizer[q->codewords[j] - 1] * q->flcoeffs6[i];
|
| 606 |
else
|
| 607 |
q->CWdecoded[j] = -quantizer[max_size - 2 - q->codewords[j]] * q->flcoeffs6[i];
|
| 608 |
} |
| 609 |
} |
| 610 |
} |
| 611 |
return 0; |
| 612 |
} |
| 613 |
|
| 614 |
|
| 615 |
static int imc_get_coeffs (IMCContext* q) { |
| 616 |
int i, j, cw_len, cw;
|
| 617 |
|
| 618 |
for(i = 0; i < BANDS; i++) { |
| 619 |
if(!q->sumLenArr[i]) continue; |
| 620 |
if (q->bandFlagsBuf[i] || q->bandWidthT[i]) {
|
| 621 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) { |
| 622 |
cw_len = q->CWlengthT[j]; |
| 623 |
cw = 0;
|
| 624 |
|
| 625 |
if (get_bits_count(&q->gb) + cw_len > 512){ |
| 626 |
//av_log(NULL,0,"Band %i coeff %i cw_len %i\n",i,j,cw_len);
|
| 627 |
return -1; |
| 628 |
} |
| 629 |
|
| 630 |
if(cw_len && (!q->bandFlagsBuf[i] || !q->skipFlags[j]))
|
| 631 |
cw = get_bits(&q->gb, cw_len); |
| 632 |
|
| 633 |
q->codewords[j] = cw; |
| 634 |
} |
| 635 |
} |
| 636 |
} |
| 637 |
return 0; |
| 638 |
} |
| 639 |
|
| 640 |
static int imc_decode_frame(AVCodecContext * avctx, |
| 641 |
void *data, int *data_size, |
| 642 |
AVPacket *avpkt) |
| 643 |
{
|
| 644 |
const uint8_t *buf = avpkt->data;
|
| 645 |
int buf_size = avpkt->size;
|
| 646 |
|
| 647 |
IMCContext *q = avctx->priv_data; |
| 648 |
|
| 649 |
int stream_format_code;
|
| 650 |
int imc_hdr, i, j;
|
| 651 |
int flag;
|
| 652 |
int bits, summer;
|
| 653 |
int counter, bitscount;
|
| 654 |
uint16_t buf16[IMC_BLOCK_SIZE / 2];
|
| 655 |
|
| 656 |
if (buf_size < IMC_BLOCK_SIZE) {
|
| 657 |
av_log(avctx, AV_LOG_ERROR, "imc frame too small!\n");
|
| 658 |
return -1; |
| 659 |
} |
| 660 |
for(i = 0; i < IMC_BLOCK_SIZE / 2; i++) |
| 661 |
buf16[i] = bswap_16(((const uint16_t*)buf)[i]);
|
| 662 |
|
| 663 |
init_get_bits(&q->gb, (const uint8_t*)buf16, IMC_BLOCK_SIZE * 8); |
| 664 |
|
| 665 |
/* Check the frame header */
|
| 666 |
imc_hdr = get_bits(&q->gb, 9);
|
| 667 |
if (imc_hdr != IMC_FRAME_ID) {
|
| 668 |
av_log(avctx, AV_LOG_ERROR, "imc frame header check failed!\n");
|
| 669 |
av_log(avctx, AV_LOG_ERROR, "got %x instead of 0x21.\n", imc_hdr);
|
| 670 |
return -1; |
| 671 |
} |
| 672 |
stream_format_code = get_bits(&q->gb, 3);
|
| 673 |
|
| 674 |
if(stream_format_code & 1){ |
| 675 |
av_log(avctx, AV_LOG_ERROR, "Stream code format %X is not supported\n", stream_format_code);
|
| 676 |
return -1; |
| 677 |
} |
| 678 |
|
| 679 |
// av_log(avctx, AV_LOG_DEBUG, "stream_format_code = %d\n", stream_format_code);
|
| 680 |
|
| 681 |
if (stream_format_code & 0x04) |
| 682 |
q->decoder_reset = 1;
|
| 683 |
|
| 684 |
if(q->decoder_reset) {
|
| 685 |
memset(q->out_samples, 0, sizeof(q->out_samples)); |
| 686 |
for(i = 0; i < BANDS; i++)q->old_floor[i] = 1.0; |
| 687 |
for(i = 0; i < COEFFS; i++)q->CWdecoded[i] = 0; |
| 688 |
q->decoder_reset = 0;
|
| 689 |
} |
| 690 |
|
| 691 |
flag = get_bits1(&q->gb); |
| 692 |
imc_read_level_coeffs(q, stream_format_code, q->levlCoeffBuf); |
| 693 |
|
| 694 |
if (stream_format_code & 0x4) |
| 695 |
imc_decode_level_coefficients(q, q->levlCoeffBuf, q->flcoeffs1, q->flcoeffs2); |
| 696 |
else
|
| 697 |
imc_decode_level_coefficients2(q, q->levlCoeffBuf, q->old_floor, q->flcoeffs1, q->flcoeffs2); |
| 698 |
|
| 699 |
memcpy(q->old_floor, q->flcoeffs1, 32 * sizeof(float)); |
| 700 |
|
| 701 |
counter = 0;
|
| 702 |
for (i=0 ; i<BANDS ; i++) { |
| 703 |
if (q->levlCoeffBuf[i] == 16) { |
| 704 |
q->bandWidthT[i] = 0;
|
| 705 |
counter++; |
| 706 |
} else
|
| 707 |
q->bandWidthT[i] = band_tab[i+1] - band_tab[i];
|
| 708 |
} |
| 709 |
memset(q->bandFlagsBuf, 0, BANDS * sizeof(int)); |
| 710 |
for(i = 0; i < BANDS-1; i++) { |
| 711 |
if (q->bandWidthT[i])
|
| 712 |
q->bandFlagsBuf[i] = get_bits1(&q->gb); |
| 713 |
} |
| 714 |
|
| 715 |
imc_calculate_coeffs(q, q->flcoeffs1, q->flcoeffs2, q->bandWidthT, q->flcoeffs3, q->flcoeffs5); |
| 716 |
|
| 717 |
bitscount = 0;
|
| 718 |
/* first 4 bands will be assigned 5 bits per coefficient */
|
| 719 |
if (stream_format_code & 0x2) { |
| 720 |
bitscount += 15;
|
| 721 |
|
| 722 |
q->bitsBandT[0] = 5; |
| 723 |
q->CWlengthT[0] = 5; |
| 724 |
q->CWlengthT[1] = 5; |
| 725 |
q->CWlengthT[2] = 5; |
| 726 |
for(i = 1; i < 4; i++){ |
| 727 |
bits = (q->levlCoeffBuf[i] == 16) ? 0 : 5; |
| 728 |
q->bitsBandT[i] = bits; |
| 729 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) { |
| 730 |
q->CWlengthT[j] = bits; |
| 731 |
bitscount += bits; |
| 732 |
} |
| 733 |
} |
| 734 |
} |
| 735 |
|
| 736 |
if(bit_allocation (q, stream_format_code, 512 - bitscount - get_bits_count(&q->gb), flag) < 0) { |
| 737 |
av_log(avctx, AV_LOG_ERROR, "Bit allocations failed\n");
|
| 738 |
q->decoder_reset = 1;
|
| 739 |
return -1; |
| 740 |
} |
| 741 |
|
| 742 |
for(i = 0; i < BANDS; i++) { |
| 743 |
q->sumLenArr[i] = 0;
|
| 744 |
q->skipFlagRaw[i] = 0;
|
| 745 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) |
| 746 |
q->sumLenArr[i] += q->CWlengthT[j]; |
| 747 |
if (q->bandFlagsBuf[i])
|
| 748 |
if( (((band_tab[i+1] - band_tab[i]) * 1.5) > q->sumLenArr[i]) && (q->sumLenArr[i] > 0)) |
| 749 |
q->skipFlagRaw[i] = 1;
|
| 750 |
} |
| 751 |
|
| 752 |
imc_get_skip_coeff(q); |
| 753 |
|
| 754 |
for(i = 0; i < BANDS; i++) { |
| 755 |
q->flcoeffs6[i] = q->flcoeffs1[i]; |
| 756 |
/* band has flag set and at least one coded coefficient */
|
| 757 |
if (q->bandFlagsBuf[i] && (band_tab[i+1] - band_tab[i]) != q->skipFlagCount[i]){ |
| 758 |
q->flcoeffs6[i] *= q->sqrt_tab[band_tab[i+1] - band_tab[i]] /
|
| 759 |
q->sqrt_tab[(band_tab[i+1] - band_tab[i] - q->skipFlagCount[i])];
|
| 760 |
} |
| 761 |
} |
| 762 |
|
| 763 |
/* calculate bits left, bits needed and adjust bit allocation */
|
| 764 |
bits = summer = 0;
|
| 765 |
|
| 766 |
for(i = 0; i < BANDS; i++) { |
| 767 |
if (q->bandFlagsBuf[i]) {
|
| 768 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) { |
| 769 |
if(q->skipFlags[j]) {
|
| 770 |
summer += q->CWlengthT[j]; |
| 771 |
q->CWlengthT[j] = 0;
|
| 772 |
} |
| 773 |
} |
| 774 |
bits += q->skipFlagBits[i]; |
| 775 |
summer -= q->skipFlagBits[i]; |
| 776 |
} |
| 777 |
} |
| 778 |
imc_adjust_bit_allocation(q, summer); |
| 779 |
|
| 780 |
for(i = 0; i < BANDS; i++) { |
| 781 |
q->sumLenArr[i] = 0;
|
| 782 |
|
| 783 |
for(j = band_tab[i]; j < band_tab[i+1]; j++) |
| 784 |
if (!q->skipFlags[j])
|
| 785 |
q->sumLenArr[i] += q->CWlengthT[j]; |
| 786 |
} |
| 787 |
|
| 788 |
memset(q->codewords, 0, sizeof(q->codewords)); |
| 789 |
|
| 790 |
if(imc_get_coeffs(q) < 0) { |
| 791 |
av_log(avctx, AV_LOG_ERROR, "Read coefficients failed\n");
|
| 792 |
q->decoder_reset = 1;
|
| 793 |
return 0; |
| 794 |
} |
| 795 |
|
| 796 |
if(inverse_quant_coeff(q, stream_format_code) < 0) { |
| 797 |
av_log(avctx, AV_LOG_ERROR, "Inverse quantization of coefficients failed\n");
|
| 798 |
q->decoder_reset = 1;
|
| 799 |
return 0; |
| 800 |
} |
| 801 |
|
| 802 |
memset(q->skipFlags, 0, sizeof(q->skipFlags)); |
| 803 |
|
| 804 |
imc_imdct256(q); |
| 805 |
|
| 806 |
q->dsp.float_to_int16(data, q->out_samples, COEFFS); |
| 807 |
|
| 808 |
*data_size = COEFFS * sizeof(int16_t);
|
| 809 |
|
| 810 |
return IMC_BLOCK_SIZE;
|
| 811 |
} |
| 812 |
|
| 813 |
|
| 814 |
static av_cold int imc_decode_close(AVCodecContext * avctx) |
| 815 |
{
|
| 816 |
IMCContext *q = avctx->priv_data; |
| 817 |
|
| 818 |
ff_fft_end(&q->fft); |
| 819 |
return 0; |
| 820 |
} |
| 821 |
|
| 822 |
|
| 823 |
AVCodec imc_decoder = {
|
| 824 |
.name = "imc",
|
| 825 |
.type = CODEC_TYPE_AUDIO, |
| 826 |
.id = CODEC_ID_IMC, |
| 827 |
.priv_data_size = sizeof(IMCContext),
|
| 828 |
.init = imc_decode_init, |
| 829 |
.close = imc_decode_close, |
| 830 |
.decode = imc_decode_frame, |
| 831 |
.long_name = NULL_IF_CONFIG_SMALL("IMC (Intel Music Coder)"),
|
| 832 |
}; |