2 * The simplest mpeg audio layer 2 encoder
3 * Copyright (c) 2000, 2001 Fabrice Bellard.
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include "mpegaudio.h"
22 /* currently, cannot change these constants (need to modify
23 quantization stage) */
26 #define MUL(a,b) (((INT64)(a) * (INT64)(b)) >> FRAC_BITS)
27 #define FIX(a) ((int)((a) * (1 << FRAC_BITS)))
29 #define SAMPLES_BUF_SIZE 4096
31 typedef struct MpegAudioContext
{
35 int lsf
; /* 1 if mpeg2 low bitrate selected */
36 int bitrate_index
; /* bit rate */
38 int frame_size
; /* frame size, in bits, without padding */
39 INT64 nb_samples
; /* total number of samples encoded */
40 /* padding computation */
41 int frame_frac
, frame_frac_incr
, do_padding
;
42 short samples_buf
[MPA_MAX_CHANNELS
][SAMPLES_BUF_SIZE
]; /* buffer for filter */
43 int samples_offset
[MPA_MAX_CHANNELS
]; /* offset in samples_buf */
44 int sb_samples
[MPA_MAX_CHANNELS
][3][12][SBLIMIT
];
45 unsigned char scale_factors
[MPA_MAX_CHANNELS
][SBLIMIT
][3]; /* scale factors */
46 /* code to group 3 scale factors */
47 unsigned char scale_code
[MPA_MAX_CHANNELS
][SBLIMIT
];
48 int sblimit
; /* number of used subbands */
49 const unsigned char *alloc_table
;
52 /* define it to use floats in quantization (I don't like floats !) */
55 #include "mpegaudiotab.h"
57 int MPA_encode_init(AVCodecContext
*avctx
)
59 MpegAudioContext
*s
= avctx
->priv_data
;
60 int freq
= avctx
->sample_rate
;
61 int bitrate
= avctx
->bit_rate
;
62 int channels
= avctx
->channels
;
68 bitrate
= bitrate
/ 1000;
69 s
->nb_channels
= channels
;
71 s
->bit_rate
= bitrate
* 1000;
72 avctx
->frame_size
= MPA_FRAME_SIZE
;
73 avctx
->key_frame
= 1; /* always key frame */
78 if (mpa_freq_tab
[i
] == freq
)
80 if ((mpa_freq_tab
[i
] / 2) == freq
) {
89 /* encoding bitrate & frequency */
91 if (mpa_bitrate_tab
[s
->lsf
][1][i
] == bitrate
)
98 /* compute total header size & pad bit */
100 a
= (float)(bitrate
* 1000 * MPA_FRAME_SIZE
) / (freq
* 8.0);
101 s
->frame_size
= ((int)a
) * 8;
103 /* frame fractional size to compute padding */
105 s
->frame_frac_incr
= (int)((a
- floor(a
)) * 65536.0);
107 /* select the right allocation table */
108 table
= l2_select_table(bitrate
, s
->nb_channels
, freq
, s
->lsf
);
110 /* number of used subbands */
111 s
->sblimit
= sblimit_table
[table
];
112 s
->alloc_table
= alloc_tables
[table
];
115 printf("%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
116 bitrate
, freq
, s
->frame_size
, table
, s
->frame_frac_incr
);
119 for(i
=0;i
<s
->nb_channels
;i
++)
120 s
->samples_offset
[i
] = 0;
126 v
= (v
+ (1 << (16 - WFRAC_BITS
- 1))) >> (16 - WFRAC_BITS
);
132 filter_bank
[512 - i
] = v
;
136 v
= (int)(pow(2.0, (3 - i
) / 3.0) * (1 << 20));
139 scale_factor_table
[i
] = v
;
141 scale_factor_inv_table
[i
] = pow(2.0, -(3 - i
) / 3.0) / (float)(1 << 20);
144 scale_factor_shift
[i
] = 21 - P
- (i
/ 3);
145 scale_factor_mult
[i
] = (1 << P
) * pow(2.0, (i
% 3) / 3.0);
160 scale_diff_table
[i
] = v
;
169 total_quant_bits
[i
] = 12 * v
;
175 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
176 static void idct32(int *out
, int *tab
)
180 const int *xp
= costab32
;
182 for(j
=31;j
>=3;j
-=2) tab
[j
] += tab
[j
- 2];
221 x3
= MUL(t
[16], FIX(SQRT2
*0.5));
225 x2
= MUL(-(t
[24] + t
[8]), FIX(SQRT2
*0.5));
226 x1
= MUL((t
[8] - x2
), xp
[0]);
227 x2
= MUL((t
[8] + x2
), xp
[1]);
240 xr
= MUL(t
[28],xp
[0]);
244 xr
= MUL(t
[4],xp
[1]);
245 t
[ 4] = (t
[24] - xr
);
246 t
[24] = (t
[24] + xr
);
248 xr
= MUL(t
[20],xp
[2]);
252 xr
= MUL(t
[12],xp
[3]);
253 t
[12] = (t
[16] - xr
);
254 t
[16] = (t
[16] + xr
);
259 for (i
= 0; i
< 4; i
++) {
260 xr
= MUL(tab
[30-i
*4],xp
[0]);
261 tab
[30-i
*4] = (tab
[i
*4] - xr
);
262 tab
[ i
*4] = (tab
[i
*4] + xr
);
264 xr
= MUL(tab
[ 2+i
*4],xp
[1]);
265 tab
[ 2+i
*4] = (tab
[28-i
*4] - xr
);
266 tab
[28-i
*4] = (tab
[28-i
*4] + xr
);
268 xr
= MUL(tab
[31-i
*4],xp
[0]);
269 tab
[31-i
*4] = (tab
[1+i
*4] - xr
);
270 tab
[ 1+i
*4] = (tab
[1+i
*4] + xr
);
272 xr
= MUL(tab
[ 3+i
*4],xp
[1]);
273 tab
[ 3+i
*4] = (tab
[29-i
*4] - xr
);
274 tab
[29-i
*4] = (tab
[29-i
*4] + xr
);
282 xr
= MUL(t1
[0], *xp
);
291 out
[i
] = tab
[bitinv32
[i
]];
295 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS)
297 static void filter(MpegAudioContext
*s
, int ch
, short *samples
, int incr
)
300 int sum
, offset
, i
, j
;
305 // print_pow1(samples, 1152);
307 offset
= s
->samples_offset
[ch
];
308 out
= &s
->sb_samples
[ch
][0][0][0];
310 /* 32 samples at once */
312 s
->samples_buf
[ch
][offset
+ (31 - i
)] = samples
[0];
317 p
= s
->samples_buf
[ch
] + offset
;
321 sum
= p
[0*64] * q
[0*64];
322 sum
+= p
[1*64] * q
[1*64];
323 sum
+= p
[2*64] * q
[2*64];
324 sum
+= p
[3*64] * q
[3*64];
325 sum
+= p
[4*64] * q
[4*64];
326 sum
+= p
[5*64] * q
[5*64];
327 sum
+= p
[6*64] * q
[6*64];
328 sum
+= p
[7*64] * q
[7*64];
333 tmp1
[0] = tmp
[16] >> WSHIFT
;
334 for( i
=1; i
<=16; i
++ ) tmp1
[i
] = (tmp
[i
+16]+tmp
[16-i
]) >> WSHIFT
;
335 for( i
=17; i
<=31; i
++ ) tmp1
[i
] = (tmp
[i
+16]-tmp
[80-i
]) >> WSHIFT
;
339 /* advance of 32 samples */
342 /* handle the wrap around */
344 memmove(s
->samples_buf
[ch
] + SAMPLES_BUF_SIZE
- (512 - 32),
345 s
->samples_buf
[ch
], (512 - 32) * 2);
346 offset
= SAMPLES_BUF_SIZE
- 512;
349 s
->samples_offset
[ch
] = offset
;
351 // print_pow(s->sb_samples, 1152);
354 static void compute_scale_factors(unsigned char scale_code
[SBLIMIT
],
355 unsigned char scale_factors
[SBLIMIT
][3],
356 int sb_samples
[3][12][SBLIMIT
],
359 int *p
, vmax
, v
, n
, i
, j
, k
, code
;
361 unsigned char *sf
= &scale_factors
[0][0];
363 for(j
=0;j
<sblimit
;j
++) {
365 /* find the max absolute value */
366 p
= &sb_samples
[i
][0][j
];
374 /* compute the scale factor index using log 2 computations */
377 /* n is the position of the MSB of vmax. now
378 use at most 2 compares to find the index */
379 index
= (21 - n
) * 3 - 3;
381 while (vmax
<= scale_factor_table
[index
+1])
384 index
= 0; /* very unlikely case of overflow */
387 index
= 62; /* value 63 is not allowed */
391 printf("%2d:%d in=%x %x %d\n",
392 j
, i
, vmax
, scale_factor_table
[index
], index
);
394 /* store the scale factor */
395 assert(index
>=0 && index
<= 63);
399 /* compute the transmission factor : look if the scale factors
400 are close enough to each other */
401 d1
= scale_diff_table
[sf
[0] - sf
[1] + 64];
402 d2
= scale_diff_table
[sf
[1] - sf
[2] + 64];
404 /* handle the 25 cases */
405 switch(d1
* 5 + d2
) {
437 sf
[1] = sf
[2] = sf
[0];
442 sf
[0] = sf
[1] = sf
[2];
448 sf
[0] = sf
[2] = sf
[1];
454 sf
[1] = sf
[2] = sf
[0];
461 printf("%d: %2d %2d %2d %d %d -> %d\n", j
,
462 sf
[0], sf
[1], sf
[2], d1
, d2
, code
);
464 scale_code
[j
] = code
;
469 /* The most important function : psycho acoustic module. In this
470 encoder there is basically none, so this is the worst you can do,
471 but also this is the simpler. */
472 static void psycho_acoustic_model(MpegAudioContext
*s
, short smr
[SBLIMIT
])
476 for(i
=0;i
<s
->sblimit
;i
++) {
477 smr
[i
] = (int)(fixed_smr
[i
] * 10);
482 #define SB_NOTALLOCATED 0
483 #define SB_ALLOCATED 1
486 /* Try to maximize the smr while using a number of bits inferior to
487 the frame size. I tried to make the code simpler, faster and
488 smaller than other encoders :-) */
489 static void compute_bit_allocation(MpegAudioContext
*s
,
490 short smr1
[MPA_MAX_CHANNELS
][SBLIMIT
],
491 unsigned char bit_alloc
[MPA_MAX_CHANNELS
][SBLIMIT
],
494 int i
, ch
, b
, max_smr
, max_ch
, max_sb
, current_frame_size
, max_frame_size
;
496 short smr
[MPA_MAX_CHANNELS
][SBLIMIT
];
497 unsigned char subband_status
[MPA_MAX_CHANNELS
][SBLIMIT
];
498 const unsigned char *alloc
;
500 memcpy(smr
, smr1
, s
->nb_channels
* sizeof(short) * SBLIMIT
);
501 memset(subband_status
, SB_NOTALLOCATED
, s
->nb_channels
* SBLIMIT
);
502 memset(bit_alloc
, 0, s
->nb_channels
* SBLIMIT
);
504 /* compute frame size and padding */
505 max_frame_size
= s
->frame_size
;
506 s
->frame_frac
+= s
->frame_frac_incr
;
507 if (s
->frame_frac
>= 65536) {
508 s
->frame_frac
-= 65536;
515 /* compute the header + bit alloc size */
516 current_frame_size
= 32;
517 alloc
= s
->alloc_table
;
518 for(i
=0;i
<s
->sblimit
;i
++) {
520 current_frame_size
+= incr
* s
->nb_channels
;
524 /* look for the subband with the largest signal to mask ratio */
527 max_smr
= 0x80000000;
528 for(ch
=0;ch
<s
->nb_channels
;ch
++) {
529 for(i
=0;i
<s
->sblimit
;i
++) {
530 if (smr
[ch
][i
] > max_smr
&& subband_status
[ch
][i
] != SB_NOMORE
) {
531 max_smr
= smr
[ch
][i
];
538 printf("current=%d max=%d max_sb=%d alloc=%d\n",
539 current_frame_size
, max_frame_size
, max_sb
,
545 /* find alloc table entry (XXX: not optimal, should use
547 alloc
= s
->alloc_table
;
548 for(i
=0;i
<max_sb
;i
++) {
549 alloc
+= 1 << alloc
[0];
552 if (subband_status
[max_ch
][max_sb
] == SB_NOTALLOCATED
) {
553 /* nothing was coded for this band: add the necessary bits */
554 incr
= 2 + nb_scale_factors
[s
->scale_code
[max_ch
][max_sb
]] * 6;
555 incr
+= total_quant_bits
[alloc
[1]];
557 /* increments bit allocation */
558 b
= bit_alloc
[max_ch
][max_sb
];
559 incr
= total_quant_bits
[alloc
[b
+ 1]] -
560 total_quant_bits
[alloc
[b
]];
563 if (current_frame_size
+ incr
<= max_frame_size
) {
564 /* can increase size */
565 b
= ++bit_alloc
[max_ch
][max_sb
];
566 current_frame_size
+= incr
;
567 /* decrease smr by the resolution we added */
568 smr
[max_ch
][max_sb
] = smr1
[max_ch
][max_sb
] - quant_snr
[alloc
[b
]];
569 /* max allocation size reached ? */
570 if (b
== ((1 << alloc
[0]) - 1))
571 subband_status
[max_ch
][max_sb
] = SB_NOMORE
;
573 subband_status
[max_ch
][max_sb
] = SB_ALLOCATED
;
575 /* cannot increase the size of this subband */
576 subband_status
[max_ch
][max_sb
] = SB_NOMORE
;
579 *padding
= max_frame_size
- current_frame_size
;
580 assert(*padding
>= 0);
583 for(i
=0;i
<s
->sblimit
;i
++) {
584 printf("%d ", bit_alloc
[i
]);
591 * Output the mpeg audio layer 2 frame. Note how the code is small
592 * compared to other encoders :-)
594 static void encode_frame(MpegAudioContext
*s
,
595 unsigned char bit_alloc
[MPA_MAX_CHANNELS
][SBLIMIT
],
598 int i
, j
, k
, l
, bit_alloc_bits
, b
, ch
;
601 PutBitContext
*p
= &s
->pb
;
605 put_bits(p
, 12, 0xfff);
606 put_bits(p
, 1, 1 - s
->lsf
); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
607 put_bits(p
, 2, 4-2); /* layer 2 */
608 put_bits(p
, 1, 1); /* no error protection */
609 put_bits(p
, 4, s
->bitrate_index
);
610 put_bits(p
, 2, s
->freq_index
);
611 put_bits(p
, 1, s
->do_padding
); /* use padding */
612 put_bits(p
, 1, 0); /* private_bit */
613 put_bits(p
, 2, s
->nb_channels
== 2 ? MPA_STEREO
: MPA_MONO
);
614 put_bits(p
, 2, 0); /* mode_ext */
615 put_bits(p
, 1, 0); /* no copyright */
616 put_bits(p
, 1, 1); /* original */
617 put_bits(p
, 2, 0); /* no emphasis */
621 for(i
=0;i
<s
->sblimit
;i
++) {
622 bit_alloc_bits
= s
->alloc_table
[j
];
623 for(ch
=0;ch
<s
->nb_channels
;ch
++) {
624 put_bits(p
, bit_alloc_bits
, bit_alloc
[ch
][i
]);
626 j
+= 1 << bit_alloc_bits
;
630 for(i
=0;i
<s
->sblimit
;i
++) {
631 for(ch
=0;ch
<s
->nb_channels
;ch
++) {
632 if (bit_alloc
[ch
][i
])
633 put_bits(p
, 2, s
->scale_code
[ch
][i
]);
638 for(i
=0;i
<s
->sblimit
;i
++) {
639 for(ch
=0;ch
<s
->nb_channels
;ch
++) {
640 if (bit_alloc
[ch
][i
]) {
641 sf
= &s
->scale_factors
[ch
][i
][0];
642 switch(s
->scale_code
[ch
][i
]) {
644 put_bits(p
, 6, sf
[0]);
645 put_bits(p
, 6, sf
[1]);
646 put_bits(p
, 6, sf
[2]);
650 put_bits(p
, 6, sf
[0]);
651 put_bits(p
, 6, sf
[2]);
654 put_bits(p
, 6, sf
[0]);
661 /* quantization & write sub band samples */
666 for(i
=0;i
<s
->sblimit
;i
++) {
667 bit_alloc_bits
= s
->alloc_table
[j
];
668 for(ch
=0;ch
<s
->nb_channels
;ch
++) {
669 b
= bit_alloc
[ch
][i
];
671 int qindex
, steps
, m
, sample
, bits
;
672 /* we encode 3 sub band samples of the same sub band at a time */
673 qindex
= s
->alloc_table
[j
+b
];
674 steps
= quant_steps
[qindex
];
676 sample
= s
->sb_samples
[ch
][k
][l
+ m
][i
];
677 /* divide by scale factor */
681 a
= (float)sample
* scale_factor_inv_table
[s
->scale_factors
[ch
][i
][k
]];
682 q
[m
] = (int)((a
+ 1.0) * steps
* 0.5);
686 int q1
, e
, shift
, mult
;
687 e
= s
->scale_factors
[ch
][i
][k
];
688 shift
= scale_factor_shift
[e
];
689 mult
= scale_factor_mult
[e
];
691 /* normalize to P bits */
693 q1
= sample
<< (-shift
);
695 q1
= sample
>> shift
;
696 q1
= (q1
* mult
) >> P
;
697 q
[m
] = ((q1
+ (1 << P
)) * steps
) >> (P
+ 1);
702 assert(q
[m
] >= 0 && q
[m
] < steps
);
704 bits
= quant_bits
[qindex
];
706 /* group the 3 values to save bits */
708 q
[0] + steps
* (q
[1] + steps
* q
[2]));
710 printf("%d: gr1 %d\n",
711 i
, q
[0] + steps
* (q
[1] + steps
* q
[2]));
715 printf("%d: gr3 %d %d %d\n",
716 i
, q
[0], q
[1], 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 int MPA_encode_frame(AVCodecContext
*avctx
,
739 unsigned char *frame
, int buf_size
, void *data
)
741 MpegAudioContext
*s
= avctx
->priv_data
;
742 short *samples
= data
;
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
->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 init_put_bits(&s
->pb
, frame
, MPA_MAX_CODED_FRAME_SIZE
, NULL
, NULL
);
762 encode_frame(s
, bit_alloc
, padding
);
764 s
->nb_samples
+= MPA_FRAME_SIZE
;
765 return pbBufPtr(&s
->pb
) - s
->pb
.buf
;
769 AVCodec mp2_encoder
= {
773 sizeof(MpegAudioContext
),