Libav 0.7.1
|
00001 /* 00002 * The simplest mpeg audio layer 2 encoder 00003 * Copyright (c) 2000, 2001 Fabrice Bellard 00004 * 00005 * This file is part of Libav. 00006 * 00007 * Libav is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * Libav is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with Libav; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 00027 #include "avcodec.h" 00028 #include "put_bits.h" 00029 00030 #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */ 00031 #define WFRAC_BITS 14 /* fractional bits for window */ 00032 00033 #include "mpegaudio.h" 00034 00035 /* currently, cannot change these constants (need to modify 00036 quantization stage) */ 00037 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) 00038 00039 #define SAMPLES_BUF_SIZE 4096 00040 00041 typedef struct MpegAudioContext { 00042 PutBitContext pb; 00043 int nb_channels; 00044 int lsf; /* 1 if mpeg2 low bitrate selected */ 00045 int bitrate_index; /* bit rate */ 00046 int freq_index; 00047 int frame_size; /* frame size, in bits, without padding */ 00048 /* padding computation */ 00049 int frame_frac, frame_frac_incr, do_padding; 00050 short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ 00051 int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ 00052 int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; 00053 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ 00054 /* code to group 3 scale factors */ 00055 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; 00056 int sblimit; /* number of used subbands */ 00057 const unsigned char *alloc_table; 00058 } MpegAudioContext; 00059 00060 /* define it to use floats in quantization (I don't like floats !) */ 00061 #define USE_FLOATS 00062 00063 #include "mpegaudiodata.h" 00064 #include "mpegaudiotab.h" 00065 00066 static av_cold int MPA_encode_init(AVCodecContext *avctx) 00067 { 00068 MpegAudioContext *s = avctx->priv_data; 00069 int freq = avctx->sample_rate; 00070 int bitrate = avctx->bit_rate; 00071 int channels = avctx->channels; 00072 int i, v, table; 00073 float a; 00074 00075 if (channels <= 0 || channels > 2){ 00076 av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels); 00077 return -1; 00078 } 00079 bitrate = bitrate / 1000; 00080 s->nb_channels = channels; 00081 avctx->frame_size = MPA_FRAME_SIZE; 00082 00083 /* encoding freq */ 00084 s->lsf = 0; 00085 for(i=0;i<3;i++) { 00086 if (ff_mpa_freq_tab[i] == freq) 00087 break; 00088 if ((ff_mpa_freq_tab[i] / 2) == freq) { 00089 s->lsf = 1; 00090 break; 00091 } 00092 } 00093 if (i == 3){ 00094 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq); 00095 return -1; 00096 } 00097 s->freq_index = i; 00098 00099 /* encoding bitrate & frequency */ 00100 for(i=0;i<15;i++) { 00101 if (ff_mpa_bitrate_tab[s->lsf][1][i] == bitrate) 00102 break; 00103 } 00104 if (i == 15){ 00105 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate); 00106 return -1; 00107 } 00108 s->bitrate_index = i; 00109 00110 /* compute total header size & pad bit */ 00111 00112 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); 00113 s->frame_size = ((int)a) * 8; 00114 00115 /* frame fractional size to compute padding */ 00116 s->frame_frac = 0; 00117 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); 00118 00119 /* select the right allocation table */ 00120 table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf); 00121 00122 /* number of used subbands */ 00123 s->sblimit = ff_mpa_sblimit_table[table]; 00124 s->alloc_table = ff_mpa_alloc_tables[table]; 00125 00126 av_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", 00127 bitrate, freq, s->frame_size, table, s->frame_frac_incr); 00128 00129 for(i=0;i<s->nb_channels;i++) 00130 s->samples_offset[i] = 0; 00131 00132 for(i=0;i<257;i++) { 00133 int v; 00134 v = ff_mpa_enwindow[i]; 00135 #if WFRAC_BITS != 16 00136 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); 00137 #endif 00138 filter_bank[i] = v; 00139 if ((i & 63) != 0) 00140 v = -v; 00141 if (i != 0) 00142 filter_bank[512 - i] = v; 00143 } 00144 00145 for(i=0;i<64;i++) { 00146 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20)); 00147 if (v <= 0) 00148 v = 1; 00149 scale_factor_table[i] = v; 00150 #ifdef USE_FLOATS 00151 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20); 00152 #else 00153 #define P 15 00154 scale_factor_shift[i] = 21 - P - (i / 3); 00155 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0); 00156 #endif 00157 } 00158 for(i=0;i<128;i++) { 00159 v = i - 64; 00160 if (v <= -3) 00161 v = 0; 00162 else if (v < 0) 00163 v = 1; 00164 else if (v == 0) 00165 v = 2; 00166 else if (v < 3) 00167 v = 3; 00168 else 00169 v = 4; 00170 scale_diff_table[i] = v; 00171 } 00172 00173 for(i=0;i<17;i++) { 00174 v = ff_mpa_quant_bits[i]; 00175 if (v < 0) 00176 v = -v; 00177 else 00178 v = v * 3; 00179 total_quant_bits[i] = 12 * v; 00180 } 00181 00182 avctx->coded_frame= avcodec_alloc_frame(); 00183 avctx->coded_frame->key_frame= 1; 00184 00185 return 0; 00186 } 00187 00188 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */ 00189 static void idct32(int *out, int *tab) 00190 { 00191 int i, j; 00192 int *t, *t1, xr; 00193 const int *xp = costab32; 00194 00195 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; 00196 00197 t = tab + 30; 00198 t1 = tab + 2; 00199 do { 00200 t[0] += t[-4]; 00201 t[1] += t[1 - 4]; 00202 t -= 4; 00203 } while (t != t1); 00204 00205 t = tab + 28; 00206 t1 = tab + 4; 00207 do { 00208 t[0] += t[-8]; 00209 t[1] += t[1-8]; 00210 t[2] += t[2-8]; 00211 t[3] += t[3-8]; 00212 t -= 8; 00213 } while (t != t1); 00214 00215 t = tab; 00216 t1 = tab + 32; 00217 do { 00218 t[ 3] = -t[ 3]; 00219 t[ 6] = -t[ 6]; 00220 00221 t[11] = -t[11]; 00222 t[12] = -t[12]; 00223 t[13] = -t[13]; 00224 t[15] = -t[15]; 00225 t += 16; 00226 } while (t != t1); 00227 00228 00229 t = tab; 00230 t1 = tab + 8; 00231 do { 00232 int x1, x2, x3, x4; 00233 00234 x3 = MUL(t[16], FIX(SQRT2*0.5)); 00235 x4 = t[0] - x3; 00236 x3 = t[0] + x3; 00237 00238 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5)); 00239 x1 = MUL((t[8] - x2), xp[0]); 00240 x2 = MUL((t[8] + x2), xp[1]); 00241 00242 t[ 0] = x3 + x1; 00243 t[ 8] = x4 - x2; 00244 t[16] = x4 + x2; 00245 t[24] = x3 - x1; 00246 t++; 00247 } while (t != t1); 00248 00249 xp += 2; 00250 t = tab; 00251 t1 = tab + 4; 00252 do { 00253 xr = MUL(t[28],xp[0]); 00254 t[28] = (t[0] - xr); 00255 t[0] = (t[0] + xr); 00256 00257 xr = MUL(t[4],xp[1]); 00258 t[ 4] = (t[24] - xr); 00259 t[24] = (t[24] + xr); 00260 00261 xr = MUL(t[20],xp[2]); 00262 t[20] = (t[8] - xr); 00263 t[ 8] = (t[8] + xr); 00264 00265 xr = MUL(t[12],xp[3]); 00266 t[12] = (t[16] - xr); 00267 t[16] = (t[16] + xr); 00268 t++; 00269 } while (t != t1); 00270 xp += 4; 00271 00272 for (i = 0; i < 4; i++) { 00273 xr = MUL(tab[30-i*4],xp[0]); 00274 tab[30-i*4] = (tab[i*4] - xr); 00275 tab[ i*4] = (tab[i*4] + xr); 00276 00277 xr = MUL(tab[ 2+i*4],xp[1]); 00278 tab[ 2+i*4] = (tab[28-i*4] - xr); 00279 tab[28-i*4] = (tab[28-i*4] + xr); 00280 00281 xr = MUL(tab[31-i*4],xp[0]); 00282 tab[31-i*4] = (tab[1+i*4] - xr); 00283 tab[ 1+i*4] = (tab[1+i*4] + xr); 00284 00285 xr = MUL(tab[ 3+i*4],xp[1]); 00286 tab[ 3+i*4] = (tab[29-i*4] - xr); 00287 tab[29-i*4] = (tab[29-i*4] + xr); 00288 00289 xp += 2; 00290 } 00291 00292 t = tab + 30; 00293 t1 = tab + 1; 00294 do { 00295 xr = MUL(t1[0], *xp); 00296 t1[0] = (t[0] - xr); 00297 t[0] = (t[0] + xr); 00298 t -= 2; 00299 t1 += 2; 00300 xp++; 00301 } while (t >= tab); 00302 00303 for(i=0;i<32;i++) { 00304 out[i] = tab[bitinv32[i]]; 00305 } 00306 } 00307 00308 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) 00309 00310 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr) 00311 { 00312 short *p, *q; 00313 int sum, offset, i, j; 00314 int tmp[64]; 00315 int tmp1[32]; 00316 int *out; 00317 00318 // print_pow1(samples, 1152); 00319 00320 offset = s->samples_offset[ch]; 00321 out = &s->sb_samples[ch][0][0][0]; 00322 for(j=0;j<36;j++) { 00323 /* 32 samples at once */ 00324 for(i=0;i<32;i++) { 00325 s->samples_buf[ch][offset + (31 - i)] = samples[0]; 00326 samples += incr; 00327 } 00328 00329 /* filter */ 00330 p = s->samples_buf[ch] + offset; 00331 q = filter_bank; 00332 /* maxsum = 23169 */ 00333 for(i=0;i<64;i++) { 00334 sum = p[0*64] * q[0*64]; 00335 sum += p[1*64] * q[1*64]; 00336 sum += p[2*64] * q[2*64]; 00337 sum += p[3*64] * q[3*64]; 00338 sum += p[4*64] * q[4*64]; 00339 sum += p[5*64] * q[5*64]; 00340 sum += p[6*64] * q[6*64]; 00341 sum += p[7*64] * q[7*64]; 00342 tmp[i] = sum; 00343 p++; 00344 q++; 00345 } 00346 tmp1[0] = tmp[16] >> WSHIFT; 00347 for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; 00348 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; 00349 00350 idct32(out, tmp1); 00351 00352 /* advance of 32 samples */ 00353 offset -= 32; 00354 out += 32; 00355 /* handle the wrap around */ 00356 if (offset < 0) { 00357 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), 00358 s->samples_buf[ch], (512 - 32) * 2); 00359 offset = SAMPLES_BUF_SIZE - 512; 00360 } 00361 } 00362 s->samples_offset[ch] = offset; 00363 00364 // print_pow(s->sb_samples, 1152); 00365 } 00366 00367 static void compute_scale_factors(unsigned char scale_code[SBLIMIT], 00368 unsigned char scale_factors[SBLIMIT][3], 00369 int sb_samples[3][12][SBLIMIT], 00370 int sblimit) 00371 { 00372 int *p, vmax, v, n, i, j, k, code; 00373 int index, d1, d2; 00374 unsigned char *sf = &scale_factors[0][0]; 00375 00376 for(j=0;j<sblimit;j++) { 00377 for(i=0;i<3;i++) { 00378 /* find the max absolute value */ 00379 p = &sb_samples[i][0][j]; 00380 vmax = abs(*p); 00381 for(k=1;k<12;k++) { 00382 p += SBLIMIT; 00383 v = abs(*p); 00384 if (v > vmax) 00385 vmax = v; 00386 } 00387 /* compute the scale factor index using log 2 computations */ 00388 if (vmax > 1) { 00389 n = av_log2(vmax); 00390 /* n is the position of the MSB of vmax. now 00391 use at most 2 compares to find the index */ 00392 index = (21 - n) * 3 - 3; 00393 if (index >= 0) { 00394 while (vmax <= scale_factor_table[index+1]) 00395 index++; 00396 } else { 00397 index = 0; /* very unlikely case of overflow */ 00398 } 00399 } else { 00400 index = 62; /* value 63 is not allowed */ 00401 } 00402 00403 av_dlog(NULL, "%2d:%d in=%x %x %d\n", 00404 j, i, vmax, scale_factor_table[index], index); 00405 /* store the scale factor */ 00406 assert(index >=0 && index <= 63); 00407 sf[i] = index; 00408 } 00409 00410 /* compute the transmission factor : look if the scale factors 00411 are close enough to each other */ 00412 d1 = scale_diff_table[sf[0] - sf[1] + 64]; 00413 d2 = scale_diff_table[sf[1] - sf[2] + 64]; 00414 00415 /* handle the 25 cases */ 00416 switch(d1 * 5 + d2) { 00417 case 0*5+0: 00418 case 0*5+4: 00419 case 3*5+4: 00420 case 4*5+0: 00421 case 4*5+4: 00422 code = 0; 00423 break; 00424 case 0*5+1: 00425 case 0*5+2: 00426 case 4*5+1: 00427 case 4*5+2: 00428 code = 3; 00429 sf[2] = sf[1]; 00430 break; 00431 case 0*5+3: 00432 case 4*5+3: 00433 code = 3; 00434 sf[1] = sf[2]; 00435 break; 00436 case 1*5+0: 00437 case 1*5+4: 00438 case 2*5+4: 00439 code = 1; 00440 sf[1] = sf[0]; 00441 break; 00442 case 1*5+1: 00443 case 1*5+2: 00444 case 2*5+0: 00445 case 2*5+1: 00446 case 2*5+2: 00447 code = 2; 00448 sf[1] = sf[2] = sf[0]; 00449 break; 00450 case 2*5+3: 00451 case 3*5+3: 00452 code = 2; 00453 sf[0] = sf[1] = sf[2]; 00454 break; 00455 case 3*5+0: 00456 case 3*5+1: 00457 case 3*5+2: 00458 code = 2; 00459 sf[0] = sf[2] = sf[1]; 00460 break; 00461 case 1*5+3: 00462 code = 2; 00463 if (sf[0] > sf[2]) 00464 sf[0] = sf[2]; 00465 sf[1] = sf[2] = sf[0]; 00466 break; 00467 default: 00468 assert(0); //cannot happen 00469 code = 0; /* kill warning */ 00470 } 00471 00472 av_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j, 00473 sf[0], sf[1], sf[2], d1, d2, code); 00474 scale_code[j] = code; 00475 sf += 3; 00476 } 00477 } 00478 00479 /* The most important function : psycho acoustic module. In this 00480 encoder there is basically none, so this is the worst you can do, 00481 but also this is the simpler. */ 00482 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) 00483 { 00484 int i; 00485 00486 for(i=0;i<s->sblimit;i++) { 00487 smr[i] = (int)(fixed_smr[i] * 10); 00488 } 00489 } 00490 00491 00492 #define SB_NOTALLOCATED 0 00493 #define SB_ALLOCATED 1 00494 #define SB_NOMORE 2 00495 00496 /* Try to maximize the smr while using a number of bits inferior to 00497 the frame size. I tried to make the code simpler, faster and 00498 smaller than other encoders :-) */ 00499 static void compute_bit_allocation(MpegAudioContext *s, 00500 short smr1[MPA_MAX_CHANNELS][SBLIMIT], 00501 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], 00502 int *padding) 00503 { 00504 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; 00505 int incr; 00506 short smr[MPA_MAX_CHANNELS][SBLIMIT]; 00507 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; 00508 const unsigned char *alloc; 00509 00510 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); 00511 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); 00512 memset(bit_alloc, 0, s->nb_channels * SBLIMIT); 00513 00514 /* compute frame size and padding */ 00515 max_frame_size = s->frame_size; 00516 s->frame_frac += s->frame_frac_incr; 00517 if (s->frame_frac >= 65536) { 00518 s->frame_frac -= 65536; 00519 s->do_padding = 1; 00520 max_frame_size += 8; 00521 } else { 00522 s->do_padding = 0; 00523 } 00524 00525 /* compute the header + bit alloc size */ 00526 current_frame_size = 32; 00527 alloc = s->alloc_table; 00528 for(i=0;i<s->sblimit;i++) { 00529 incr = alloc[0]; 00530 current_frame_size += incr * s->nb_channels; 00531 alloc += 1 << incr; 00532 } 00533 for(;;) { 00534 /* look for the subband with the largest signal to mask ratio */ 00535 max_sb = -1; 00536 max_ch = -1; 00537 max_smr = INT_MIN; 00538 for(ch=0;ch<s->nb_channels;ch++) { 00539 for(i=0;i<s->sblimit;i++) { 00540 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { 00541 max_smr = smr[ch][i]; 00542 max_sb = i; 00543 max_ch = ch; 00544 } 00545 } 00546 } 00547 if (max_sb < 0) 00548 break; 00549 av_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n", 00550 current_frame_size, max_frame_size, max_sb, max_ch, 00551 bit_alloc[max_ch][max_sb]); 00552 00553 /* find alloc table entry (XXX: not optimal, should use 00554 pointer table) */ 00555 alloc = s->alloc_table; 00556 for(i=0;i<max_sb;i++) { 00557 alloc += 1 << alloc[0]; 00558 } 00559 00560 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { 00561 /* nothing was coded for this band: add the necessary bits */ 00562 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; 00563 incr += total_quant_bits[alloc[1]]; 00564 } else { 00565 /* increments bit allocation */ 00566 b = bit_alloc[max_ch][max_sb]; 00567 incr = total_quant_bits[alloc[b + 1]] - 00568 total_quant_bits[alloc[b]]; 00569 } 00570 00571 if (current_frame_size + incr <= max_frame_size) { 00572 /* can increase size */ 00573 b = ++bit_alloc[max_ch][max_sb]; 00574 current_frame_size += incr; 00575 /* decrease smr by the resolution we added */ 00576 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; 00577 /* max allocation size reached ? */ 00578 if (b == ((1 << alloc[0]) - 1)) 00579 subband_status[max_ch][max_sb] = SB_NOMORE; 00580 else 00581 subband_status[max_ch][max_sb] = SB_ALLOCATED; 00582 } else { 00583 /* cannot increase the size of this subband */ 00584 subband_status[max_ch][max_sb] = SB_NOMORE; 00585 } 00586 } 00587 *padding = max_frame_size - current_frame_size; 00588 assert(*padding >= 0); 00589 } 00590 00591 /* 00592 * Output the mpeg audio layer 2 frame. Note how the code is small 00593 * compared to other encoders :-) 00594 */ 00595 static void encode_frame(MpegAudioContext *s, 00596 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], 00597 int padding) 00598 { 00599 int i, j, k, l, bit_alloc_bits, b, ch; 00600 unsigned char *sf; 00601 int q[3]; 00602 PutBitContext *p = &s->pb; 00603 00604 /* header */ 00605 00606 put_bits(p, 12, 0xfff); 00607 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */ 00608 put_bits(p, 2, 4-2); /* layer 2 */ 00609 put_bits(p, 1, 1); /* no error protection */ 00610 put_bits(p, 4, s->bitrate_index); 00611 put_bits(p, 2, s->freq_index); 00612 put_bits(p, 1, s->do_padding); /* use padding */ 00613 put_bits(p, 1, 0); /* private_bit */ 00614 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); 00615 put_bits(p, 2, 0); /* mode_ext */ 00616 put_bits(p, 1, 0); /* no copyright */ 00617 put_bits(p, 1, 1); /* original */ 00618 put_bits(p, 2, 0); /* no emphasis */ 00619 00620 /* bit allocation */ 00621 j = 0; 00622 for(i=0;i<s->sblimit;i++) { 00623 bit_alloc_bits = s->alloc_table[j]; 00624 for(ch=0;ch<s->nb_channels;ch++) { 00625 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); 00626 } 00627 j += 1 << bit_alloc_bits; 00628 } 00629 00630 /* scale codes */ 00631 for(i=0;i<s->sblimit;i++) { 00632 for(ch=0;ch<s->nb_channels;ch++) { 00633 if (bit_alloc[ch][i]) 00634 put_bits(p, 2, s->scale_code[ch][i]); 00635 } 00636 } 00637 00638 /* scale factors */ 00639 for(i=0;i<s->sblimit;i++) { 00640 for(ch=0;ch<s->nb_channels;ch++) { 00641 if (bit_alloc[ch][i]) { 00642 sf = &s->scale_factors[ch][i][0]; 00643 switch(s->scale_code[ch][i]) { 00644 case 0: 00645 put_bits(p, 6, sf[0]); 00646 put_bits(p, 6, sf[1]); 00647 put_bits(p, 6, sf[2]); 00648 break; 00649 case 3: 00650 case 1: 00651 put_bits(p, 6, sf[0]); 00652 put_bits(p, 6, sf[2]); 00653 break; 00654 case 2: 00655 put_bits(p, 6, sf[0]); 00656 break; 00657 } 00658 } 00659 } 00660 } 00661 00662 /* quantization & write sub band samples */ 00663 00664 for(k=0;k<3;k++) { 00665 for(l=0;l<12;l+=3) { 00666 j = 0; 00667 for(i=0;i<s->sblimit;i++) { 00668 bit_alloc_bits = s->alloc_table[j]; 00669 for(ch=0;ch<s->nb_channels;ch++) { 00670 b = bit_alloc[ch][i]; 00671 if (b) { 00672 int qindex, steps, m, sample, bits; 00673 /* we encode 3 sub band samples of the same sub band at a time */ 00674 qindex = s->alloc_table[j+b]; 00675 steps = ff_mpa_quant_steps[qindex]; 00676 for(m=0;m<3;m++) { 00677 sample = s->sb_samples[ch][k][l + m][i]; 00678 /* divide by scale factor */ 00679 #ifdef USE_FLOATS 00680 { 00681 float a; 00682 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]]; 00683 q[m] = (int)((a + 1.0) * steps * 0.5); 00684 } 00685 #else 00686 { 00687 int q1, e, shift, mult; 00688 e = s->scale_factors[ch][i][k]; 00689 shift = scale_factor_shift[e]; 00690 mult = scale_factor_mult[e]; 00691 00692 /* normalize to P bits */ 00693 if (shift < 0) 00694 q1 = sample << (-shift); 00695 else 00696 q1 = sample >> shift; 00697 q1 = (q1 * mult) >> P; 00698 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1); 00699 } 00700 #endif 00701 if (q[m] >= steps) 00702 q[m] = steps - 1; 00703 assert(q[m] >= 0 && q[m] < steps); 00704 } 00705 bits = ff_mpa_quant_bits[qindex]; 00706 if (bits < 0) { 00707 /* group the 3 values to save bits */ 00708 put_bits(p, -bits, 00709 q[0] + steps * (q[1] + steps * q[2])); 00710 } else { 00711 put_bits(p, bits, q[0]); 00712 put_bits(p, bits, q[1]); 00713 put_bits(p, bits, q[2]); 00714 } 00715 } 00716 } 00717 /* next subband in alloc table */ 00718 j += 1 << bit_alloc_bits; 00719 } 00720 } 00721 } 00722 00723 /* padding */ 00724 for(i=0;i<padding;i++) 00725 put_bits(p, 1, 0); 00726 00727 /* flush */ 00728 flush_put_bits(p); 00729 } 00730 00731 static int MPA_encode_frame(AVCodecContext *avctx, 00732 unsigned char *frame, int buf_size, void *data) 00733 { 00734 MpegAudioContext *s = avctx->priv_data; 00735 const short *samples = data; 00736 short smr[MPA_MAX_CHANNELS][SBLIMIT]; 00737 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; 00738 int padding, i; 00739 00740 for(i=0;i<s->nb_channels;i++) { 00741 filter(s, i, samples + i, s->nb_channels); 00742 } 00743 00744 for(i=0;i<s->nb_channels;i++) { 00745 compute_scale_factors(s->scale_code[i], s->scale_factors[i], 00746 s->sb_samples[i], s->sblimit); 00747 } 00748 for(i=0;i<s->nb_channels;i++) { 00749 psycho_acoustic_model(s, smr[i]); 00750 } 00751 compute_bit_allocation(s, smr, bit_alloc, &padding); 00752 00753 init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE); 00754 00755 encode_frame(s, bit_alloc, padding); 00756 00757 return put_bits_ptr(&s->pb) - s->pb.buf; 00758 } 00759 00760 static av_cold int MPA_encode_close(AVCodecContext *avctx) 00761 { 00762 av_freep(&avctx->coded_frame); 00763 return 0; 00764 } 00765 00766 AVCodec ff_mp2_encoder = { 00767 "mp2", 00768 AVMEDIA_TYPE_AUDIO, 00769 CODEC_ID_MP2, 00770 sizeof(MpegAudioContext), 00771 MPA_encode_init, 00772 MPA_encode_frame, 00773 MPA_encode_close, 00774 NULL, 00775 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, 00776 .supported_samplerates= (const int[]){44100, 48000, 32000, 22050, 24000, 16000, 0}, 00777 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"), 00778 };