1/* 2 * WMA compatible encoder 3 * Copyright (c) 2007 Michael Niedermayer 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#include "avcodec.h" 23#include "wma.h" 24 25#undef NDEBUG 26#include <assert.h> 27 28 29static int encode_init(AVCodecContext * avctx){ 30 WMACodecContext *s = avctx->priv_data; 31 int i, flags1, flags2; 32 uint8_t *extradata; 33 34 s->avctx = avctx; 35 36 if(avctx->channels > MAX_CHANNELS) { 37 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer", 38 avctx->channels, MAX_CHANNELS); 39 return AVERROR(EINVAL); 40 } 41 42 if (avctx->sample_rate > 48000) { 43 av_log(avctx, AV_LOG_ERROR, "sample rate is too high: %d > 48kHz", 44 avctx->sample_rate); 45 return AVERROR(EINVAL); 46 } 47 48 if(avctx->bit_rate < 24*1000) { 49 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n", 50 avctx->bit_rate); 51 return AVERROR(EINVAL); 52 } 53 54 /* extract flag infos */ 55 flags1 = 0; 56 flags2 = 1; 57 if (avctx->codec->id == CODEC_ID_WMAV1) { 58 extradata= av_malloc(4); 59 avctx->extradata_size= 4; 60 AV_WL16(extradata, flags1); 61 AV_WL16(extradata+2, flags2); 62 } else if (avctx->codec->id == CODEC_ID_WMAV2) { 63 extradata= av_mallocz(10); 64 avctx->extradata_size= 10; 65 AV_WL32(extradata, flags1); 66 AV_WL16(extradata+4, flags2); 67 }else 68 assert(0); 69 avctx->extradata= extradata; 70 s->use_exp_vlc = flags2 & 0x0001; 71 s->use_bit_reservoir = flags2 & 0x0002; 72 s->use_variable_block_len = flags2 & 0x0004; 73 if (avctx->channels == 2) 74 s->ms_stereo = 1; 75 76 ff_wma_init(avctx, flags2); 77 78 /* init MDCT */ 79 for(i = 0; i < s->nb_block_sizes; i++) 80 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0); 81 82 s->block_align = avctx->bit_rate * (int64_t)s->frame_len / 83 (avctx->sample_rate * 8); 84 s->block_align = FFMIN(s->block_align, MAX_CODED_SUPERFRAME_SIZE); 85 avctx->block_align = s->block_align; 86 avctx->bit_rate = avctx->block_align * 8LL * avctx->sample_rate / 87 s->frame_len; 88//av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate); 89 avctx->frame_size= s->frame_len; 90 91 return 0; 92} 93 94 95static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) { 96 WMACodecContext *s = avctx->priv_data; 97 int window_index= s->frame_len_bits - s->block_len_bits; 98 FFTContext *mdct = &s->mdct_ctx[window_index]; 99 int i, j, channel; 100 const float * win = s->windows[window_index]; 101 int window_len = 1 << s->block_len_bits; 102 float n = window_len/2; 103 104 for (channel = 0; channel < avctx->channels; channel++) { 105 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len); 106 j = channel; 107 for (i = 0; i < len; i++, j += avctx->channels){ 108 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1]; 109 s->frame_out[channel][i] = audio[j] / n * win[i]; 110 } 111 mdct->mdct_calc(mdct, s->coefs[channel], s->output); 112 } 113} 114 115//FIXME use for decoding too 116static void init_exp(WMACodecContext *s, int ch, const int *exp_param){ 117 int n; 118 const uint16_t *ptr; 119 float v, *q, max_scale, *q_end; 120 121 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; 122 q = s->exponents[ch]; 123 q_end = q + s->block_len; 124 max_scale = 0; 125 while (q < q_end) { 126 /* XXX: use a table */ 127 v = pow(10, *exp_param++ * (1.0 / 16.0)); 128 max_scale= FFMAX(max_scale, v); 129 n = *ptr++; 130 do { 131 *q++ = v; 132 } while (--n); 133 } 134 s->max_exponent[ch] = max_scale; 135} 136 137static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){ 138 int last_exp; 139 const uint16_t *ptr; 140 float *q, *q_end; 141 142 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; 143 q = s->exponents[ch]; 144 q_end = q + s->block_len; 145 if (s->version == 1) { 146 last_exp= *exp_param++; 147 assert(last_exp-10 >= 0 && last_exp-10 < 32); 148 put_bits(&s->pb, 5, last_exp - 10); 149 q+= *ptr++; 150 }else 151 last_exp = 36; 152 while (q < q_end) { 153 int exp = *exp_param++; 154 int code = exp - last_exp + 60; 155 assert(code >= 0 && code < 120); 156 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]); 157 /* XXX: use a table */ 158 q+= *ptr++; 159 last_exp= exp; 160 } 161} 162 163static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){ 164 int v, bsize, ch, coef_nb_bits, parse_exponents; 165 float mdct_norm; 166 int nb_coefs[MAX_CHANNELS]; 167 static const int fixed_exp[25]={20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20}; 168 169 //FIXME remove duplication relative to decoder 170 if (s->use_variable_block_len) { 171 assert(0); //FIXME not implemented 172 }else{ 173 /* fixed block len */ 174 s->next_block_len_bits = s->frame_len_bits; 175 s->prev_block_len_bits = s->frame_len_bits; 176 s->block_len_bits = s->frame_len_bits; 177 } 178 179 s->block_len = 1 << s->block_len_bits; 180// assert((s->block_pos + s->block_len) <= s->frame_len); 181 bsize = s->frame_len_bits - s->block_len_bits; 182 183 //FIXME factor 184 v = s->coefs_end[bsize] - s->coefs_start; 185 for(ch = 0; ch < s->nb_channels; ch++) 186 nb_coefs[ch] = v; 187 { 188 int n4 = s->block_len / 2; 189 mdct_norm = 1.0 / (float)n4; 190 if (s->version == 1) { 191 mdct_norm *= sqrt(n4); 192 } 193 } 194 195 if (s->nb_channels == 2) { 196 put_bits(&s->pb, 1, !!s->ms_stereo); 197 } 198 199 for(ch = 0; ch < s->nb_channels; ch++) { 200 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always 201 if (s->channel_coded[ch]) { 202 init_exp(s, ch, fixed_exp); 203 } 204 } 205 206 for(ch = 0; ch < s->nb_channels; ch++) { 207 if (s->channel_coded[ch]) { 208 WMACoef *coefs1; 209 float *coefs, *exponents, mult; 210 int i, n; 211 212 coefs1 = s->coefs1[ch]; 213 exponents = s->exponents[ch]; 214 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; 215 mult *= mdct_norm; 216 coefs = src_coefs[ch]; 217 if (s->use_noise_coding && 0) { 218 assert(0); //FIXME not implemented 219 } else { 220 coefs += s->coefs_start; 221 n = nb_coefs[ch]; 222 for(i = 0;i < n; i++){ 223 double t= *coefs++ / (exponents[i] * mult); 224 if(t<-32768 || t>32767) 225 return -1; 226 227 coefs1[i] = lrint(t); 228 } 229 } 230 } 231 } 232 233 v = 0; 234 for(ch = 0; ch < s->nb_channels; ch++) { 235 int a = s->channel_coded[ch]; 236 put_bits(&s->pb, 1, a); 237 v |= a; 238 } 239 240 if (!v) 241 return 1; 242 243 for(v= total_gain-1; v>=127; v-= 127) 244 put_bits(&s->pb, 7, 127); 245 put_bits(&s->pb, 7, v); 246 247 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain); 248 249 if (s->use_noise_coding) { 250 for(ch = 0; ch < s->nb_channels; ch++) { 251 if (s->channel_coded[ch]) { 252 int i, n; 253 n = s->exponent_high_sizes[bsize]; 254 for(i=0;i<n;i++) { 255 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0); 256 if (0) 257 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; 258 } 259 } 260 } 261 } 262 263 parse_exponents = 1; 264 if (s->block_len_bits != s->frame_len_bits) { 265 put_bits(&s->pb, 1, parse_exponents); 266 } 267 268 if (parse_exponents) { 269 for(ch = 0; ch < s->nb_channels; ch++) { 270 if (s->channel_coded[ch]) { 271 if (s->use_exp_vlc) { 272 encode_exp_vlc(s, ch, fixed_exp); 273 } else { 274 assert(0); //FIXME not implemented 275// encode_exp_lsp(s, ch); 276 } 277 } 278 } 279 } else { 280 assert(0); //FIXME not implemented 281 } 282 283 for(ch = 0; ch < s->nb_channels; ch++) { 284 if (s->channel_coded[ch]) { 285 int run, tindex; 286 WMACoef *ptr, *eptr; 287 tindex = (ch == 1 && s->ms_stereo); 288 ptr = &s->coefs1[ch][0]; 289 eptr = ptr + nb_coefs[ch]; 290 291 run=0; 292 for(;ptr < eptr; ptr++){ 293 if(*ptr){ 294 int level= *ptr; 295 int abs_level= FFABS(level); 296 int code= 0; 297 if(abs_level <= s->coef_vlcs[tindex]->max_level){ 298 if(run < s->coef_vlcs[tindex]->levels[abs_level-1]) 299 code= run + s->int_table[tindex][abs_level-1]; 300 } 301 302 assert(code < s->coef_vlcs[tindex]->n); 303 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]); 304 305 if(code == 0){ 306 if(1<<coef_nb_bits <= abs_level) 307 return -1; 308 309 310 //Workaround minor rounding differences for the regression tests, FIXME we should find and replace the problematic float by fixpoint for reg tests 311 if(abs_level == 0x71B && (s->avctx->flags & CODEC_FLAG_BITEXACT)) abs_level=0x71A; 312 313 put_bits(&s->pb, coef_nb_bits, abs_level); 314 put_bits(&s->pb, s->frame_len_bits, run); 315 } 316 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere 317 run=0; 318 }else{ 319 run++; 320 } 321 } 322 if(run) 323 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]); 324 } 325 if (s->version == 1 && s->nb_channels >= 2) { 326 avpriv_align_put_bits(&s->pb); 327 } 328 } 329 return 0; 330} 331 332static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){ 333 init_put_bits(&s->pb, buf, buf_size); 334 335 if (s->use_bit_reservoir) { 336 assert(0);//FIXME not implemented 337 }else{ 338 if(encode_block(s, src_coefs, total_gain) < 0) 339 return INT_MAX; 340 } 341 342 avpriv_align_put_bits(&s->pb); 343 344 return put_bits_count(&s->pb)/8 - s->block_align; 345} 346 347static int encode_superframe(AVCodecContext *avctx, 348 unsigned char *buf, int buf_size, void *data){ 349 WMACodecContext *s = avctx->priv_data; 350 const short *samples = data; 351 int i, total_gain; 352 353 s->block_len_bits= s->frame_len_bits; //required by non variable block len 354 s->block_len = 1 << s->block_len_bits; 355 356 apply_window_and_mdct(avctx, samples, avctx->frame_size); 357 358 if (s->ms_stereo) { 359 float a, b; 360 int i; 361 362 for(i = 0; i < s->block_len; i++) { 363 a = s->coefs[0][i]*0.5; 364 b = s->coefs[1][i]*0.5; 365 s->coefs[0][i] = a + b; 366 s->coefs[1][i] = a - b; 367 } 368 } 369 370 if (buf_size < 2 * MAX_CODED_SUPERFRAME_SIZE) { 371 av_log(avctx, AV_LOG_ERROR, "output buffer size is too small\n"); 372 return AVERROR(EINVAL); 373 } 374 375#if 1 376 total_gain= 128; 377 for(i=64; i; i>>=1){ 378 int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i); 379 if(error<0) 380 total_gain-= i; 381 } 382#else 383 total_gain= 90; 384 best= encode_frame(s, s->coefs, buf, buf_size, total_gain); 385 for(i=32; i; i>>=1){ 386 int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i); 387 int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i); 388 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain); 389 if(scoreL < FFMIN(best, scoreR)){ 390 best = scoreL; 391 total_gain -= i; 392 }else if(scoreR < best){ 393 best = scoreR; 394 total_gain += i; 395 } 396 } 397#endif 398 399 encode_frame(s, s->coefs, buf, buf_size, total_gain); 400 assert((put_bits_count(&s->pb) & 7) == 0); 401 i= s->block_align - (put_bits_count(&s->pb)+7)/8; 402 assert(i>=0); 403 while(i--) 404 put_bits(&s->pb, 8, 'N'); 405 406 flush_put_bits(&s->pb); 407 return put_bits_ptr(&s->pb) - s->pb.buf; 408} 409 410AVCodec ff_wmav1_encoder = { 411 .name = "wmav1", 412 .type = AVMEDIA_TYPE_AUDIO, 413 .id = CODEC_ID_WMAV1, 414 .priv_data_size = sizeof(WMACodecContext), 415 .init = encode_init, 416 .encode = encode_superframe, 417 .close = ff_wma_end, 418 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, 419 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"), 420}; 421 422AVCodec ff_wmav2_encoder = { 423 .name = "wmav2", 424 .type = AVMEDIA_TYPE_AUDIO, 425 .id = CODEC_ID_WMAV2, 426 .priv_data_size = sizeof(WMACodecContext), 427 .init = encode_init, 428 .encode = encode_superframe, 429 .close = ff_wma_end, 430 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, 431 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"), 432}; 433