2 * The simplest mpeg audio layer 2 encoder
3 * Copyright (c) 2000, 2001 Fabrice Bellard
5 * This file is part of Libav.
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.
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.
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
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];
67 float scale_factor_inv_table[64];
68 unsigned short total_quant_bits[17]; /* total number of bits per allocation group */
71 static av_cold int MPA_encode_init(AVCodecContext *avctx)
73 MpegAudioContext *s = avctx->priv_data;
74 int freq = avctx->sample_rate;
75 int bitrate = avctx->bit_rate;
76 int channels = avctx->channels;
80 if (channels <= 0 || channels > 2){
81 av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels);
82 return AVERROR(EINVAL);
84 bitrate = bitrate / 1000;
85 s->nb_channels = channels;
86 avctx->frame_size = MPA_FRAME_SIZE;
87 avctx->initial_padding = 512 - 32 + 1;
92 if (avpriv_mpa_freq_tab[i] == freq)
94 if ((avpriv_mpa_freq_tab[i] / 2) == freq) {
100 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq);
101 return AVERROR(EINVAL);
105 /* encoding bitrate & frequency */
107 if (avpriv_mpa_bitrate_tab[s->lsf][1][i] == bitrate)
111 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate);
112 return AVERROR(EINVAL);
114 s->bitrate_index = i;
116 /* compute total header size & pad bit */
118 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
119 s->frame_size = ((int)a) * 8;
121 /* frame fractional size to compute padding */
123 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
125 /* select the right allocation table */
126 table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf);
128 /* number of used subbands */
129 s->sblimit = ff_mpa_sblimit_table[table];
130 s->alloc_table = ff_mpa_alloc_tables[table];
132 ff_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
133 bitrate, freq, s->frame_size, table, s->frame_frac_incr);
135 for(i=0;i<s->nb_channels;i++)
136 s->samples_offset[i] = 0;
140 v = ff_mpa_enwindow[i];
142 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
144 s->filter_bank[i] = v;
148 s->filter_bank[512 - i] = v;
152 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20));
155 s->scale_factor_table[i] = v;
156 s->scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20);
170 s->scale_diff_table[i] = v;
174 v = ff_mpa_quant_bits[i];
179 s->total_quant_bits[i] = 12 * v;
185 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
186 static void idct32(int *out, int *tab)
190 const int *xp = costab32;
192 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
231 x3 = MUL(t[16], FIX(SQRT2*0.5));
235 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
236 x1 = MUL((t[8] - x2), xp[0]);
237 x2 = MUL((t[8] + x2), xp[1]);
250 xr = MUL(t[28],xp[0]);
254 xr = MUL(t[4],xp[1]);
255 t[ 4] = (t[24] - xr);
256 t[24] = (t[24] + xr);
258 xr = MUL(t[20],xp[2]);
262 xr = MUL(t[12],xp[3]);
263 t[12] = (t[16] - xr);
264 t[16] = (t[16] + xr);
269 for (i = 0; i < 4; i++) {
270 xr = MUL(tab[30-i*4],xp[0]);
271 tab[30-i*4] = (tab[i*4] - xr);
272 tab[ i*4] = (tab[i*4] + xr);
274 xr = MUL(tab[ 2+i*4],xp[1]);
275 tab[ 2+i*4] = (tab[28-i*4] - xr);
276 tab[28-i*4] = (tab[28-i*4] + xr);
278 xr = MUL(tab[31-i*4],xp[0]);
279 tab[31-i*4] = (tab[1+i*4] - xr);
280 tab[ 1+i*4] = (tab[1+i*4] + xr);
282 xr = MUL(tab[ 3+i*4],xp[1]);
283 tab[ 3+i*4] = (tab[29-i*4] - xr);
284 tab[29-i*4] = (tab[29-i*4] + xr);
292 xr = MUL(t1[0], *xp);
301 out[i] = tab[bitinv32[i]];
305 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS)
307 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr)
310 int sum, offset, i, j;
315 offset = s->samples_offset[ch];
316 out = &s->sb_samples[ch][0][0][0];
318 /* 32 samples at once */
320 s->samples_buf[ch][offset + (31 - i)] = samples[0];
325 p = s->samples_buf[ch] + offset;
329 sum = p[0*64] * q[0*64];
330 sum += p[1*64] * q[1*64];
331 sum += p[2*64] * q[2*64];
332 sum += p[3*64] * q[3*64];
333 sum += p[4*64] * q[4*64];
334 sum += p[5*64] * q[5*64];
335 sum += p[6*64] * q[6*64];
336 sum += p[7*64] * q[7*64];
341 tmp1[0] = tmp[16] >> WSHIFT;
342 for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT;
343 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT;
347 /* advance of 32 samples */
350 /* handle the wrap around */
352 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
353 s->samples_buf[ch], (512 - 32) * 2);
354 offset = SAMPLES_BUF_SIZE - 512;
357 s->samples_offset[ch] = offset;
360 static void compute_scale_factors(MpegAudioContext *s,
361 unsigned char scale_code[SBLIMIT],
362 unsigned char scale_factors[SBLIMIT][3],
363 int sb_samples[3][12][SBLIMIT],
366 int *p, vmax, v, n, i, j, k, code;
368 unsigned char *sf = &scale_factors[0][0];
370 for(j=0;j<sblimit;j++) {
372 /* find the max absolute value */
373 p = &sb_samples[i][0][j];
381 /* compute the scale factor index using log 2 computations */
384 /* n is the position of the MSB of vmax. now
385 use at most 2 compares to find the index */
386 index = (21 - n) * 3 - 3;
388 while (vmax <= s->scale_factor_table[index+1])
391 index = 0; /* very unlikely case of overflow */
394 index = 62; /* value 63 is not allowed */
397 ff_dlog(NULL, "%2d:%d in=%x %x %d\n",
398 j, i, vmax, s->scale_factor_table[index], index);
399 /* store the scale factor */
400 assert(index >=0 && index <= 63);
404 /* compute the transmission factor : look if the scale factors
405 are close enough to each other */
406 d1 = s->scale_diff_table[sf[0] - sf[1] + 64];
407 d2 = s->scale_diff_table[sf[1] - sf[2] + 64];
409 /* handle the 25 cases */
410 switch(d1 * 5 + d2) {
442 sf[1] = sf[2] = sf[0];
447 sf[0] = sf[1] = sf[2];
453 sf[0] = sf[2] = sf[1];
459 sf[1] = sf[2] = sf[0];
462 assert(0); //cannot happen
463 code = 0; /* kill warning */
466 ff_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j,
467 sf[0], sf[1], sf[2], d1, d2, code);
468 scale_code[j] = code;
473 /* The most important function : psycho acoustic module. In this
474 encoder there is basically none, so this is the worst you can do,
475 but also this is the simpler. */
476 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
480 for(i=0;i<s->sblimit;i++) {
481 smr[i] = (int)(fixed_smr[i] * 10);
486 #define SB_NOTALLOCATED 0
487 #define SB_ALLOCATED 1
490 /* Try to maximize the smr while using a number of bits inferior to
491 the frame size. I tried to make the code simpler, faster and
492 smaller than other encoders :-) */
493 static void compute_bit_allocation(MpegAudioContext *s,
494 short smr1[MPA_MAX_CHANNELS][SBLIMIT],
495 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
498 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
500 short smr[MPA_MAX_CHANNELS][SBLIMIT];
501 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
502 const unsigned char *alloc;
504 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
505 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
506 memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
508 /* compute frame size and padding */
509 max_frame_size = s->frame_size;
510 s->frame_frac += s->frame_frac_incr;
511 if (s->frame_frac >= 65536) {
512 s->frame_frac -= 65536;
519 /* compute the header + bit alloc size */
520 current_frame_size = 32;
521 alloc = s->alloc_table;
522 for(i=0;i<s->sblimit;i++) {
524 current_frame_size += incr * s->nb_channels;
528 /* look for the subband with the largest signal to mask ratio */
532 for(ch=0;ch<s->nb_channels;ch++) {
533 for(i=0;i<s->sblimit;i++) {
534 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
535 max_smr = smr[ch][i];
543 ff_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n",
544 current_frame_size, max_frame_size, max_sb, max_ch,
545 bit_alloc[max_ch][max_sb]);
547 /* find alloc table entry (XXX: not optimal, should use
549 alloc = s->alloc_table;
550 for(i=0;i<max_sb;i++) {
551 alloc += 1 << alloc[0];
554 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
555 /* nothing was coded for this band: add the necessary bits */
556 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
557 incr += s->total_quant_bits[alloc[1]];
559 /* increments bit allocation */
560 b = bit_alloc[max_ch][max_sb];
561 incr = s->total_quant_bits[alloc[b + 1]] -
562 s->total_quant_bits[alloc[b]];
565 if (current_frame_size + incr <= max_frame_size) {
566 /* can increase size */
567 b = ++bit_alloc[max_ch][max_sb];
568 current_frame_size += incr;
569 /* decrease smr by the resolution we added */
570 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
571 /* max allocation size reached ? */
572 if (b == ((1 << alloc[0]) - 1))
573 subband_status[max_ch][max_sb] = SB_NOMORE;
575 subband_status[max_ch][max_sb] = SB_ALLOCATED;
577 /* cannot increase the size of this subband */
578 subband_status[max_ch][max_sb] = SB_NOMORE;
581 *padding = max_frame_size - current_frame_size;
582 assert(*padding >= 0);
586 * Output the mpeg audio layer 2 frame. Note how the code is small
587 * compared to other encoders :-)
589 static void encode_frame(MpegAudioContext *s,
590 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
593 int i, j, k, l, bit_alloc_bits, b, ch;
596 PutBitContext *p = &s->pb;
600 put_bits(p, 12, 0xfff);
601 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
602 put_bits(p, 2, 4-2); /* layer 2 */
603 put_bits(p, 1, 1); /* no error protection */
604 put_bits(p, 4, s->bitrate_index);
605 put_bits(p, 2, s->freq_index);
606 put_bits(p, 1, s->do_padding); /* use padding */
607 put_bits(p, 1, 0); /* private_bit */
608 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
609 put_bits(p, 2, 0); /* mode_ext */
610 put_bits(p, 1, 0); /* no copyright */
611 put_bits(p, 1, 1); /* original */
612 put_bits(p, 2, 0); /* no emphasis */
616 for(i=0;i<s->sblimit;i++) {
617 bit_alloc_bits = s->alloc_table[j];
618 for(ch=0;ch<s->nb_channels;ch++) {
619 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
621 j += 1 << bit_alloc_bits;
625 for(i=0;i<s->sblimit;i++) {
626 for(ch=0;ch<s->nb_channels;ch++) {
627 if (bit_alloc[ch][i])
628 put_bits(p, 2, s->scale_code[ch][i]);
633 for(i=0;i<s->sblimit;i++) {
634 for(ch=0;ch<s->nb_channels;ch++) {
635 if (bit_alloc[ch][i]) {
636 sf = &s->scale_factors[ch][i][0];
637 switch(s->scale_code[ch][i]) {
639 put_bits(p, 6, sf[0]);
640 put_bits(p, 6, sf[1]);
641 put_bits(p, 6, sf[2]);
645 put_bits(p, 6, sf[0]);
646 put_bits(p, 6, sf[2]);
649 put_bits(p, 6, sf[0]);
656 /* quantization & write sub band samples */
661 for(i=0;i<s->sblimit;i++) {
662 bit_alloc_bits = s->alloc_table[j];
663 for(ch=0;ch<s->nb_channels;ch++) {
664 b = bit_alloc[ch][i];
666 int qindex, steps, m, sample, bits;
667 /* we encode 3 sub band samples of the same sub band at a time */
668 qindex = s->alloc_table[j+b];
669 steps = ff_mpa_quant_steps[qindex];
672 sample = s->sb_samples[ch][k][l + m][i];
673 /* divide by scale factor */
674 a = (float)sample * s->scale_factor_inv_table[s->scale_factors[ch][i][k]];
675 q[m] = (int)((a + 1.0) * steps * 0.5);
678 assert(q[m] >= 0 && q[m] < steps);
680 bits = ff_mpa_quant_bits[qindex];
682 /* group the 3 values to save bits */
684 q[0] + steps * (q[1] + steps * q[2]));
686 put_bits(p, bits, q[0]);
687 put_bits(p, bits, q[1]);
688 put_bits(p, bits, q[2]);
692 /* next subband in alloc table */
693 j += 1 << bit_alloc_bits;
699 for(i=0;i<padding;i++)
706 static int MPA_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
707 const AVFrame *frame, int *got_packet_ptr)
709 MpegAudioContext *s = avctx->priv_data;
710 const int16_t *samples = (const int16_t *)frame->data[0];
711 short smr[MPA_MAX_CHANNELS][SBLIMIT];
712 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
715 for(i=0;i<s->nb_channels;i++) {
716 filter(s, i, samples + i, s->nb_channels);
719 for(i=0;i<s->nb_channels;i++) {
720 compute_scale_factors(s, s->scale_code[i], s->scale_factors[i],
721 s->sb_samples[i], s->sblimit);
723 for(i=0;i<s->nb_channels;i++) {
724 psycho_acoustic_model(s, smr[i]);
726 compute_bit_allocation(s, smr, bit_alloc, &padding);
728 if ((ret = ff_alloc_packet(avpkt, MPA_MAX_CODED_FRAME_SIZE))) {
729 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
733 init_put_bits(&s->pb, avpkt->data, avpkt->size);
735 encode_frame(s, bit_alloc, padding);
737 if (frame->pts != AV_NOPTS_VALUE)
738 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding);
740 avpkt->size = put_bits_count(&s->pb) / 8;
745 static const AVCodecDefault mp2_defaults[] = {
750 AVCodec ff_mp2_encoder = {
752 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
753 .type = AVMEDIA_TYPE_AUDIO,
754 .id = AV_CODEC_ID_MP2,
755 .priv_data_size = sizeof(MpegAudioContext),
756 .init = MPA_encode_init,
757 .encode2 = MPA_encode_frame,
758 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
759 AV_SAMPLE_FMT_NONE },
760 .supported_samplerates = (const int[]){
761 44100, 48000, 32000, 22050, 24000, 16000, 0
763 .channel_layouts = (const uint64_t[]){ AV_CH_LAYOUT_MONO,
766 .defaults = mp2_defaults,