2 * The simplest mpeg audio layer 2 encoder
3 * Copyright (c) 2000, 2001 Fabrice Bellard
5 * This file is part of FFmpeg.
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.
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.
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
24 * The simplest mpeg audio layer 2 encoder.
27 #include "libavutil/channel_layout.h"
33 #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */
34 #define WFRAC_BITS 14 /* fractional bits for window */
36 #include "mpegaudio.h"
37 #include "mpegaudiodsp.h"
38 #include "mpegaudiodata.h"
39 #include "mpegaudiotab.h"
41 /* currently, cannot change these constants (need to modify
42 quantization stage) */
43 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
45 #define SAMPLES_BUF_SIZE 4096
47 typedef struct MpegAudioContext {
50 int lsf; /* 1 if mpeg2 low bitrate selected */
51 int bitrate_index; /* bit rate */
53 int frame_size; /* frame size, in bits, without padding */
54 /* padding computation */
55 int frame_frac, frame_frac_incr, do_padding;
56 short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */
57 int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */
58 int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT];
59 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */
60 /* code to group 3 scale factors */
61 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
62 int sblimit; /* number of used subbands */
63 const unsigned char *alloc_table;
64 int16_t filter_bank[512];
65 int scale_factor_table[64];
66 unsigned char scale_diff_table[128];
68 float scale_factor_inv_table[64];
70 int8_t scale_factor_shift[64];
71 unsigned short scale_factor_mult[64];
73 unsigned short total_quant_bits[17]; /* total number of bits per allocation group */
76 static av_cold int MPA_encode_init(AVCodecContext *avctx)
78 MpegAudioContext *s = avctx->priv_data;
79 int freq = avctx->sample_rate;
80 int bitrate = avctx->bit_rate;
81 int channels = avctx->channels;
85 if (channels <= 0 || channels > 2){
86 av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels);
87 return AVERROR(EINVAL);
89 bitrate = bitrate / 1000;
90 s->nb_channels = channels;
91 avctx->frame_size = MPA_FRAME_SIZE;
92 avctx->delay = 512 - 32 + 1;
97 if (avpriv_mpa_freq_tab[i] == freq)
99 if ((avpriv_mpa_freq_tab[i] / 2) == freq) {
105 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq);
106 return AVERROR(EINVAL);
110 /* encoding bitrate & frequency */
112 if (avpriv_mpa_bitrate_tab[s->lsf][1][i] == bitrate)
116 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate);
117 return AVERROR(EINVAL);
119 s->bitrate_index = i;
121 /* compute total header size & pad bit */
123 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
124 s->frame_size = ((int)a) * 8;
126 /* frame fractional size to compute padding */
128 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
130 /* select the right allocation table */
131 table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf);
133 /* number of used subbands */
134 s->sblimit = ff_mpa_sblimit_table[table];
135 s->alloc_table = ff_mpa_alloc_tables[table];
137 av_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
138 bitrate, freq, s->frame_size, table, s->frame_frac_incr);
140 for(i=0;i<s->nb_channels;i++)
141 s->samples_offset[i] = 0;
145 v = ff_mpa_enwindow[i];
147 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
149 s->filter_bank[i] = v;
153 s->filter_bank[512 - i] = v;
157 v = (int)(exp2((3 - i) / 3.0) * (1 << 20));
160 s->scale_factor_table[i] = v;
162 s->scale_factor_inv_table[i] = exp2(-(3 - i) / 3.0) / (float)(1 << 20);
165 s->scale_factor_shift[i] = 21 - P - (i / 3);
166 s->scale_factor_mult[i] = (1 << P) * exp2((i % 3) / 3.0);
181 s->scale_diff_table[i] = v;
185 v = ff_mpa_quant_bits[i];
190 s->total_quant_bits[i] = 12 * v;
196 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
197 static void idct32(int *out, int *tab)
201 const int *xp = costab32;
203 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
242 x3 = MUL(t[16], FIX(SQRT2*0.5));
246 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
247 x1 = MUL((t[8] - x2), xp[0]);
248 x2 = MUL((t[8] + x2), xp[1]);
261 xr = MUL(t[28],xp[0]);
265 xr = MUL(t[4],xp[1]);
266 t[ 4] = (t[24] - xr);
267 t[24] = (t[24] + xr);
269 xr = MUL(t[20],xp[2]);
273 xr = MUL(t[12],xp[3]);
274 t[12] = (t[16] - xr);
275 t[16] = (t[16] + xr);
280 for (i = 0; i < 4; i++) {
281 xr = MUL(tab[30-i*4],xp[0]);
282 tab[30-i*4] = (tab[i*4] - xr);
283 tab[ i*4] = (tab[i*4] + xr);
285 xr = MUL(tab[ 2+i*4],xp[1]);
286 tab[ 2+i*4] = (tab[28-i*4] - xr);
287 tab[28-i*4] = (tab[28-i*4] + xr);
289 xr = MUL(tab[31-i*4],xp[0]);
290 tab[31-i*4] = (tab[1+i*4] - xr);
291 tab[ 1+i*4] = (tab[1+i*4] + xr);
293 xr = MUL(tab[ 3+i*4],xp[1]);
294 tab[ 3+i*4] = (tab[29-i*4] - xr);
295 tab[29-i*4] = (tab[29-i*4] + xr);
303 xr = MUL(t1[0], *xp);
312 out[i] = tab[bitinv32[i]];
316 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS)
318 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr)
321 int sum, offset, i, j;
326 offset = s->samples_offset[ch];
327 out = &s->sb_samples[ch][0][0][0];
329 /* 32 samples at once */
331 s->samples_buf[ch][offset + (31 - i)] = samples[0];
336 p = s->samples_buf[ch] + offset;
340 sum = p[0*64] * q[0*64];
341 sum += p[1*64] * q[1*64];
342 sum += p[2*64] * q[2*64];
343 sum += p[3*64] * q[3*64];
344 sum += p[4*64] * q[4*64];
345 sum += p[5*64] * q[5*64];
346 sum += p[6*64] * q[6*64];
347 sum += p[7*64] * q[7*64];
352 tmp1[0] = tmp[16] >> WSHIFT;
353 for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT;
354 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT;
358 /* advance of 32 samples */
361 /* handle the wrap around */
363 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
364 s->samples_buf[ch], (512 - 32) * 2);
365 offset = SAMPLES_BUF_SIZE - 512;
368 s->samples_offset[ch] = offset;
371 static void compute_scale_factors(MpegAudioContext *s,
372 unsigned char scale_code[SBLIMIT],
373 unsigned char scale_factors[SBLIMIT][3],
374 int sb_samples[3][12][SBLIMIT],
377 int *p, vmax, v, n, i, j, k, code;
379 unsigned char *sf = &scale_factors[0][0];
381 for(j=0;j<sblimit;j++) {
383 /* find the max absolute value */
384 p = &sb_samples[i][0][j];
392 /* compute the scale factor index using log 2 computations */
395 /* n is the position of the MSB of vmax. now
396 use at most 2 compares to find the index */
397 index = (21 - n) * 3 - 3;
399 while (vmax <= s->scale_factor_table[index+1])
402 index = 0; /* very unlikely case of overflow */
405 index = 62; /* value 63 is not allowed */
408 av_dlog(NULL, "%2d:%d in=%x %x %d\n",
409 j, i, vmax, s->scale_factor_table[index], index);
410 /* store the scale factor */
411 av_assert2(index >=0 && index <= 63);
415 /* compute the transmission factor : look if the scale factors
416 are close enough to each other */
417 d1 = s->scale_diff_table[sf[0] - sf[1] + 64];
418 d2 = s->scale_diff_table[sf[1] - sf[2] + 64];
420 /* handle the 25 cases */
421 switch(d1 * 5 + d2) {
453 sf[1] = sf[2] = sf[0];
458 sf[0] = sf[1] = sf[2];
464 sf[0] = sf[2] = sf[1];
470 sf[1] = sf[2] = sf[0];
473 av_assert2(0); //cannot happen
474 code = 0; /* kill warning */
477 av_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j,
478 sf[0], sf[1], sf[2], d1, d2, code);
479 scale_code[j] = code;
484 /* The most important function : psycho acoustic module. In this
485 encoder there is basically none, so this is the worst you can do,
486 but also this is the simpler. */
487 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
491 for(i=0;i<s->sblimit;i++) {
492 smr[i] = (int)(fixed_smr[i] * 10);
497 #define SB_NOTALLOCATED 0
498 #define SB_ALLOCATED 1
501 /* Try to maximize the smr while using a number of bits inferior to
502 the frame size. I tried to make the code simpler, faster and
503 smaller than other encoders :-) */
504 static void compute_bit_allocation(MpegAudioContext *s,
505 short smr1[MPA_MAX_CHANNELS][SBLIMIT],
506 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
509 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
511 short smr[MPA_MAX_CHANNELS][SBLIMIT];
512 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
513 const unsigned char *alloc;
515 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
516 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
517 memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
519 /* compute frame size and padding */
520 max_frame_size = s->frame_size;
521 s->frame_frac += s->frame_frac_incr;
522 if (s->frame_frac >= 65536) {
523 s->frame_frac -= 65536;
530 /* compute the header + bit alloc size */
531 current_frame_size = 32;
532 alloc = s->alloc_table;
533 for(i=0;i<s->sblimit;i++) {
535 current_frame_size += incr * s->nb_channels;
539 /* look for the subband with the largest signal to mask ratio */
543 for(ch=0;ch<s->nb_channels;ch++) {
544 for(i=0;i<s->sblimit;i++) {
545 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
546 max_smr = smr[ch][i];
554 av_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n",
555 current_frame_size, max_frame_size, max_sb, max_ch,
556 bit_alloc[max_ch][max_sb]);
558 /* find alloc table entry (XXX: not optimal, should use
560 alloc = s->alloc_table;
561 for(i=0;i<max_sb;i++) {
562 alloc += 1 << alloc[0];
565 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
566 /* nothing was coded for this band: add the necessary bits */
567 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
568 incr += s->total_quant_bits[alloc[1]];
570 /* increments bit allocation */
571 b = bit_alloc[max_ch][max_sb];
572 incr = s->total_quant_bits[alloc[b + 1]] -
573 s->total_quant_bits[alloc[b]];
576 if (current_frame_size + incr <= max_frame_size) {
577 /* can increase size */
578 b = ++bit_alloc[max_ch][max_sb];
579 current_frame_size += incr;
580 /* decrease smr by the resolution we added */
581 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
582 /* max allocation size reached ? */
583 if (b == ((1 << alloc[0]) - 1))
584 subband_status[max_ch][max_sb] = SB_NOMORE;
586 subband_status[max_ch][max_sb] = SB_ALLOCATED;
588 /* cannot increase the size of this subband */
589 subband_status[max_ch][max_sb] = SB_NOMORE;
592 *padding = max_frame_size - current_frame_size;
593 av_assert0(*padding >= 0);
597 * Output the mpeg audio layer 2 frame. Note how the code is small
598 * compared to other encoders :-)
600 static void encode_frame(MpegAudioContext *s,
601 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
604 int i, j, k, l, bit_alloc_bits, b, ch;
607 PutBitContext *p = &s->pb;
611 put_bits(p, 12, 0xfff);
612 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
613 put_bits(p, 2, 4-2); /* layer 2 */
614 put_bits(p, 1, 1); /* no error protection */
615 put_bits(p, 4, s->bitrate_index);
616 put_bits(p, 2, s->freq_index);
617 put_bits(p, 1, s->do_padding); /* use padding */
618 put_bits(p, 1, 0); /* private_bit */
619 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
620 put_bits(p, 2, 0); /* mode_ext */
621 put_bits(p, 1, 0); /* no copyright */
622 put_bits(p, 1, 1); /* original */
623 put_bits(p, 2, 0); /* no emphasis */
627 for(i=0;i<s->sblimit;i++) {
628 bit_alloc_bits = s->alloc_table[j];
629 for(ch=0;ch<s->nb_channels;ch++) {
630 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
632 j += 1 << bit_alloc_bits;
636 for(i=0;i<s->sblimit;i++) {
637 for(ch=0;ch<s->nb_channels;ch++) {
638 if (bit_alloc[ch][i])
639 put_bits(p, 2, s->scale_code[ch][i]);
644 for(i=0;i<s->sblimit;i++) {
645 for(ch=0;ch<s->nb_channels;ch++) {
646 if (bit_alloc[ch][i]) {
647 sf = &s->scale_factors[ch][i][0];
648 switch(s->scale_code[ch][i]) {
650 put_bits(p, 6, sf[0]);
651 put_bits(p, 6, sf[1]);
652 put_bits(p, 6, sf[2]);
656 put_bits(p, 6, sf[0]);
657 put_bits(p, 6, sf[2]);
660 put_bits(p, 6, sf[0]);
667 /* quantization & write sub band samples */
672 for(i=0;i<s->sblimit;i++) {
673 bit_alloc_bits = s->alloc_table[j];
674 for(ch=0;ch<s->nb_channels;ch++) {
675 b = bit_alloc[ch][i];
677 int qindex, steps, m, sample, bits;
678 /* we encode 3 sub band samples of the same sub band at a time */
679 qindex = s->alloc_table[j+b];
680 steps = ff_mpa_quant_steps[qindex];
682 sample = s->sb_samples[ch][k][l + m][i];
683 /* divide by scale factor */
687 a = (float)sample * s->scale_factor_inv_table[s->scale_factors[ch][i][k]];
688 q[m] = (int)((a + 1.0) * steps * 0.5);
692 int q1, e, shift, mult;
693 e = s->scale_factors[ch][i][k];
694 shift = s->scale_factor_shift[e];
695 mult = s->scale_factor_mult[e];
697 /* normalize to P bits */
699 q1 = sample << (-shift);
701 q1 = sample >> shift;
702 q1 = (q1 * mult) >> P;
703 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1);
710 av_assert2(q[m] >= 0 && q[m] < steps);
712 bits = ff_mpa_quant_bits[qindex];
714 /* group the 3 values to save bits */
716 q[0] + steps * (q[1] + steps * q[2]));
718 put_bits(p, bits, q[0]);
719 put_bits(p, bits, q[1]);
720 put_bits(p, bits, q[2]);
724 /* next subband in alloc table */
725 j += 1 << bit_alloc_bits;
731 for(i=0;i<padding;i++)
738 static int MPA_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
739 const AVFrame *frame, int *got_packet_ptr)
741 MpegAudioContext *s = avctx->priv_data;
742 const int16_t *samples = (const int16_t *)frame->data[0];
743 short smr[MPA_MAX_CHANNELS][SBLIMIT];
744 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
747 for(i=0;i<s->nb_channels;i++) {
748 filter(s, i, samples + i, s->nb_channels);
751 for(i=0;i<s->nb_channels;i++) {
752 compute_scale_factors(s, s->scale_code[i], s->scale_factors[i],
753 s->sb_samples[i], s->sblimit);
755 for(i=0;i<s->nb_channels;i++) {
756 psycho_acoustic_model(s, smr[i]);
758 compute_bit_allocation(s, smr, bit_alloc, &padding);
760 if ((ret = ff_alloc_packet2(avctx, avpkt, MPA_MAX_CODED_FRAME_SIZE)) < 0)
763 init_put_bits(&s->pb, avpkt->data, avpkt->size);
765 encode_frame(s, bit_alloc, padding);
767 if (frame->pts != AV_NOPTS_VALUE)
768 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);
770 avpkt->size = put_bits_count(&s->pb) / 8;
775 static const AVCodecDefault mp2_defaults[] = {