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/audioconvert.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"
38 /* currently, cannot change these constants (need to modify
39 quantization stage) */
40 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
42 #define SAMPLES_BUF_SIZE 4096
44 typedef struct MpegAudioContext {
47 int lsf; /* 1 if mpeg2 low bitrate selected */
48 int bitrate_index; /* bit rate */
50 int frame_size; /* frame size, in bits, without padding */
51 /* padding computation */
52 int frame_frac, frame_frac_incr, do_padding;
53 short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */
54 int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */
55 int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT];
56 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */
57 /* code to group 3 scale factors */
58 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
59 int sblimit; /* number of used subbands */
60 const unsigned char *alloc_table;
63 /* define it to use floats in quantization (I don't like floats !) */
66 #include "mpegaudiodata.h"
67 #include "mpegaudiotab.h"
69 static av_cold int MPA_encode_init(AVCodecContext *avctx)
71 MpegAudioContext *s = avctx->priv_data;
72 int freq = avctx->sample_rate;
73 int bitrate = avctx->bit_rate;
74 int channels = avctx->channels;
78 if (channels <= 0 || channels > 2){
79 av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels);
80 return AVERROR(EINVAL);
82 bitrate = bitrate / 1000;
83 s->nb_channels = channels;
84 avctx->frame_size = MPA_FRAME_SIZE;
85 avctx->delay = 512 - 32 + 1;
90 if (avpriv_mpa_freq_tab[i] == freq)
92 if ((avpriv_mpa_freq_tab[i] / 2) == freq) {
98 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq);
99 return AVERROR(EINVAL);
103 /* encoding bitrate & frequency */
105 if (avpriv_mpa_bitrate_tab[s->lsf][1][i] == bitrate)
109 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate);
110 return AVERROR(EINVAL);
112 s->bitrate_index = i;
114 /* compute total header size & pad bit */
116 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
117 s->frame_size = ((int)a) * 8;
119 /* frame fractional size to compute padding */
121 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
123 /* select the right allocation table */
124 table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf);
126 /* number of used subbands */
127 s->sblimit = ff_mpa_sblimit_table[table];
128 s->alloc_table = ff_mpa_alloc_tables[table];
130 av_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
131 bitrate, freq, s->frame_size, table, s->frame_frac_incr);
133 for(i=0;i<s->nb_channels;i++)
134 s->samples_offset[i] = 0;
138 v = ff_mpa_enwindow[i];
140 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
146 filter_bank[512 - i] = v;
150 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20));
153 scale_factor_table[i] = v;
155 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20);
158 scale_factor_shift[i] = 21 - P - (i / 3);
159 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0);
174 scale_diff_table[i] = v;
178 v = ff_mpa_quant_bits[i];
183 total_quant_bits[i] = 12 * v;
186 #if FF_API_OLD_ENCODE_AUDIO
187 avctx->coded_frame= avcodec_alloc_frame();
188 if (!avctx->coded_frame)
189 return AVERROR(ENOMEM);
195 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
196 static void idct32(int *out, int *tab)
200 const int *xp = costab32;
202 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
241 x3 = MUL(t[16], FIX(SQRT2*0.5));
245 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
246 x1 = MUL((t[8] - x2), xp[0]);
247 x2 = MUL((t[8] + x2), xp[1]);
260 xr = MUL(t[28],xp[0]);
264 xr = MUL(t[4],xp[1]);
265 t[ 4] = (t[24] - xr);
266 t[24] = (t[24] + xr);
268 xr = MUL(t[20],xp[2]);
272 xr = MUL(t[12],xp[3]);
273 t[12] = (t[16] - xr);
274 t[16] = (t[16] + xr);
279 for (i = 0; i < 4; i++) {
280 xr = MUL(tab[30-i*4],xp[0]);
281 tab[30-i*4] = (tab[i*4] - xr);
282 tab[ i*4] = (tab[i*4] + xr);
284 xr = MUL(tab[ 2+i*4],xp[1]);
285 tab[ 2+i*4] = (tab[28-i*4] - xr);
286 tab[28-i*4] = (tab[28-i*4] + xr);
288 xr = MUL(tab[31-i*4],xp[0]);
289 tab[31-i*4] = (tab[1+i*4] - xr);
290 tab[ 1+i*4] = (tab[1+i*4] + xr);
292 xr = MUL(tab[ 3+i*4],xp[1]);
293 tab[ 3+i*4] = (tab[29-i*4] - xr);
294 tab[29-i*4] = (tab[29-i*4] + xr);
302 xr = MUL(t1[0], *xp);
311 out[i] = tab[bitinv32[i]];
315 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS)
317 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr)
320 int sum, offset, i, j;
325 offset = s->samples_offset[ch];
326 out = &s->sb_samples[ch][0][0][0];
328 /* 32 samples at once */
330 s->samples_buf[ch][offset + (31 - i)] = samples[0];
335 p = s->samples_buf[ch] + offset;
339 sum = p[0*64] * q[0*64];
340 sum += p[1*64] * q[1*64];
341 sum += p[2*64] * q[2*64];
342 sum += p[3*64] * q[3*64];
343 sum += p[4*64] * q[4*64];
344 sum += p[5*64] * q[5*64];
345 sum += p[6*64] * q[6*64];
346 sum += p[7*64] * q[7*64];
351 tmp1[0] = tmp[16] >> WSHIFT;
352 for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT;
353 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT;
357 /* advance of 32 samples */
360 /* handle the wrap around */
362 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
363 s->samples_buf[ch], (512 - 32) * 2);
364 offset = SAMPLES_BUF_SIZE - 512;
367 s->samples_offset[ch] = offset;
370 static void compute_scale_factors(unsigned char scale_code[SBLIMIT],
371 unsigned char scale_factors[SBLIMIT][3],
372 int sb_samples[3][12][SBLIMIT],
375 int *p, vmax, v, n, i, j, k, code;
377 unsigned char *sf = &scale_factors[0][0];
379 for(j=0;j<sblimit;j++) {
381 /* find the max absolute value */
382 p = &sb_samples[i][0][j];
390 /* compute the scale factor index using log 2 computations */
393 /* n is the position of the MSB of vmax. now
394 use at most 2 compares to find the index */
395 index = (21 - n) * 3 - 3;
397 while (vmax <= scale_factor_table[index+1])
400 index = 0; /* very unlikely case of overflow */
403 index = 62; /* value 63 is not allowed */
406 av_dlog(NULL, "%2d:%d in=%x %x %d\n",
407 j, i, vmax, scale_factor_table[index], index);
408 /* store the scale factor */
409 assert(index >=0 && index <= 63);
413 /* compute the transmission factor : look if the scale factors
414 are close enough to each other */
415 d1 = scale_diff_table[sf[0] - sf[1] + 64];
416 d2 = scale_diff_table[sf[1] - sf[2] + 64];
418 /* handle the 25 cases */
419 switch(d1 * 5 + d2) {
451 sf[1] = sf[2] = sf[0];
456 sf[0] = sf[1] = sf[2];
462 sf[0] = sf[2] = sf[1];
468 sf[1] = sf[2] = sf[0];
471 assert(0); //cannot happen
472 code = 0; /* kill warning */
475 av_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j,
476 sf[0], sf[1], sf[2], d1, d2, code);
477 scale_code[j] = code;
482 /* The most important function : psycho acoustic module. In this
483 encoder there is basically none, so this is the worst you can do,
484 but also this is the simpler. */
485 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
489 for(i=0;i<s->sblimit;i++) {
490 smr[i] = (int)(fixed_smr[i] * 10);
495 #define SB_NOTALLOCATED 0
496 #define SB_ALLOCATED 1
499 /* Try to maximize the smr while using a number of bits inferior to
500 the frame size. I tried to make the code simpler, faster and
501 smaller than other encoders :-) */
502 static void compute_bit_allocation(MpegAudioContext *s,
503 short smr1[MPA_MAX_CHANNELS][SBLIMIT],
504 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
507 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
509 short smr[MPA_MAX_CHANNELS][SBLIMIT];
510 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
511 const unsigned char *alloc;
513 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
514 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
515 memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
517 /* compute frame size and padding */
518 max_frame_size = s->frame_size;
519 s->frame_frac += s->frame_frac_incr;
520 if (s->frame_frac >= 65536) {
521 s->frame_frac -= 65536;
528 /* compute the header + bit alloc size */
529 current_frame_size = 32;
530 alloc = s->alloc_table;
531 for(i=0;i<s->sblimit;i++) {
533 current_frame_size += incr * s->nb_channels;
537 /* look for the subband with the largest signal to mask ratio */
541 for(ch=0;ch<s->nb_channels;ch++) {
542 for(i=0;i<s->sblimit;i++) {
543 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
544 max_smr = smr[ch][i];
552 av_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n",
553 current_frame_size, max_frame_size, max_sb, max_ch,
554 bit_alloc[max_ch][max_sb]);
556 /* find alloc table entry (XXX: not optimal, should use
558 alloc = s->alloc_table;
559 for(i=0;i<max_sb;i++) {
560 alloc += 1 << alloc[0];
563 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
564 /* nothing was coded for this band: add the necessary bits */
565 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
566 incr += total_quant_bits[alloc[1]];
568 /* increments bit allocation */
569 b = bit_alloc[max_ch][max_sb];
570 incr = total_quant_bits[alloc[b + 1]] -
571 total_quant_bits[alloc[b]];
574 if (current_frame_size + incr <= max_frame_size) {
575 /* can increase size */
576 b = ++bit_alloc[max_ch][max_sb];
577 current_frame_size += incr;
578 /* decrease smr by the resolution we added */
579 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
580 /* max allocation size reached ? */
581 if (b == ((1 << alloc[0]) - 1))
582 subband_status[max_ch][max_sb] = SB_NOMORE;
584 subband_status[max_ch][max_sb] = SB_ALLOCATED;
586 /* cannot increase the size of this subband */
587 subband_status[max_ch][max_sb] = SB_NOMORE;
590 *padding = max_frame_size - current_frame_size;
591 assert(*padding >= 0);
595 * Output the mpeg audio layer 2 frame. Note how the code is small
596 * compared to other encoders :-)
598 static void encode_frame(MpegAudioContext *s,
599 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
602 int i, j, k, l, bit_alloc_bits, b, ch;
605 PutBitContext *p = &s->pb;
609 put_bits(p, 12, 0xfff);
610 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
611 put_bits(p, 2, 4-2); /* layer 2 */
612 put_bits(p, 1, 1); /* no error protection */
613 put_bits(p, 4, s->bitrate_index);
614 put_bits(p, 2, s->freq_index);
615 put_bits(p, 1, s->do_padding); /* use padding */
616 put_bits(p, 1, 0); /* private_bit */
617 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
618 put_bits(p, 2, 0); /* mode_ext */
619 put_bits(p, 1, 0); /* no copyright */
620 put_bits(p, 1, 1); /* original */
621 put_bits(p, 2, 0); /* no emphasis */
625 for(i=0;i<s->sblimit;i++) {
626 bit_alloc_bits = s->alloc_table[j];
627 for(ch=0;ch<s->nb_channels;ch++) {
628 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
630 j += 1 << bit_alloc_bits;
634 for(i=0;i<s->sblimit;i++) {
635 for(ch=0;ch<s->nb_channels;ch++) {
636 if (bit_alloc[ch][i])
637 put_bits(p, 2, s->scale_code[ch][i]);
642 for(i=0;i<s->sblimit;i++) {
643 for(ch=0;ch<s->nb_channels;ch++) {
644 if (bit_alloc[ch][i]) {
645 sf = &s->scale_factors[ch][i][0];
646 switch(s->scale_code[ch][i]) {
648 put_bits(p, 6, sf[0]);
649 put_bits(p, 6, sf[1]);
650 put_bits(p, 6, sf[2]);
654 put_bits(p, 6, sf[0]);
655 put_bits(p, 6, sf[2]);
658 put_bits(p, 6, sf[0]);
665 /* quantization & write sub band samples */
670 for(i=0;i<s->sblimit;i++) {
671 bit_alloc_bits = s->alloc_table[j];
672 for(ch=0;ch<s->nb_channels;ch++) {
673 b = bit_alloc[ch][i];
675 int qindex, steps, m, sample, bits;
676 /* we encode 3 sub band samples of the same sub band at a time */
677 qindex = s->alloc_table[j+b];
678 steps = ff_mpa_quant_steps[qindex];
680 sample = s->sb_samples[ch][k][l + m][i];
681 /* divide by scale factor */
685 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]];
686 q[m] = (int)((a + 1.0) * steps * 0.5);
690 int q1, e, shift, mult;
691 e = s->scale_factors[ch][i][k];
692 shift = scale_factor_shift[e];
693 mult = scale_factor_mult[e];
695 /* normalize to P bits */
697 q1 = sample << (-shift);
699 q1 = sample >> shift;
700 q1 = (q1 * mult) >> P;
701 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1);
706 assert(q[m] >= 0 && q[m] < steps);
708 bits = ff_mpa_quant_bits[qindex];
710 /* group the 3 values to save bits */
712 q[0] + steps * (q[1] + steps * q[2]));
714 put_bits(p, bits, q[0]);
715 put_bits(p, bits, q[1]);
716 put_bits(p, bits, q[2]);
720 /* next subband in alloc table */
721 j += 1 << bit_alloc_bits;
727 for(i=0;i<padding;i++)
734 static int MPA_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
735 const AVFrame *frame, int *got_packet_ptr)
737 MpegAudioContext *s = avctx->priv_data;
738 const int16_t *samples = (const int16_t *)frame->data[0];
739 short smr[MPA_MAX_CHANNELS][SBLIMIT];
740 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
743 for(i=0;i<s->nb_channels;i++) {
744 filter(s, i, samples + i, s->nb_channels);
747 for(i=0;i<s->nb_channels;i++) {
748 compute_scale_factors(s->scale_code[i], s->scale_factors[i],
749 s->sb_samples[i], s->sblimit);
751 for(i=0;i<s->nb_channels;i++) {
752 psycho_acoustic_model(s, smr[i]);
754 compute_bit_allocation(s, smr, bit_alloc, &padding);
756 if ((ret = ff_alloc_packet(avpkt, MPA_MAX_CODED_FRAME_SIZE))) {
757 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
761 init_put_bits(&s->pb, avpkt->data, avpkt->size);
763 encode_frame(s, bit_alloc, padding);
765 if (frame->pts != AV_NOPTS_VALUE)
766 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);
768 avpkt->size = put_bits_count(&s->pb) / 8;
773 static av_cold int MPA_encode_close(AVCodecContext *avctx)
775 #if FF_API_OLD_ENCODE_AUDIO
776 av_freep(&avctx->coded_frame);
781 static const AVCodecDefault mp2_defaults[] = {
786 AVCodec ff_mp2_encoder = {
788 .type = AVMEDIA_TYPE_AUDIO,
789 .id = AV_CODEC_ID_MP2,
790 .priv_data_size = sizeof(MpegAudioContext),
791 .init = MPA_encode_init,
792 .encode2 = MPA_encode_frame,
793 .close = MPA_encode_close,
794 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
795 AV_SAMPLE_FMT_NONE },
796 .supported_samplerates = (const int[]){
797 44100, 48000, 32000, 22050, 24000, 16000, 0
799 .channel_layouts = (const uint64_t[]){ AV_CH_LAYOUT_MONO,
802 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
803 .defaults = mp2_defaults,