view g726.c @ 3995:b00c06477dff libavcodec

write cabac low and range variables as early as possible to prevent stalls from reading them before they where written, the P4 is said to disslike that alot, on P3 its 2% faster (START/STOP_TIMER over decode_residual)

/*
 * G.726 ADPCM audio codec
 * Copyright (c) 2004 Roman Shaposhnik.
 *
 * This is a very straightforward rendition of the G.726
 * Section 4 "Computational Details".
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
#include <limits.h>
#include "avcodec.h"
#include "common.h"
#include "bitstream.h"

/**
 * G.726 11bit float.
 * G.726 Standard uses rather odd 11bit floating point arithmentic for
 * numerous occasions. It's a mistery to me why they did it this way
 * instead of simply using 32bit integer arithmetic.
 */
typedef struct Float11 {
        int sign;   /**< 1bit sign */
        int exp;    /**< 4bit exponent */
        int mant;   /**< 6bit mantissa */
} Float11;

static inline Float11* i2f(int16_t i, Float11* f)
{
        f->sign = (i < 0);
        if (f->sign)
                i = -i;
        f->exp = av_log2_16bit(i) + !!i;
        f->mant = i? (i<<6) >> f->exp :
                         1<<5;
        return f;
}

static inline int16_t mult(Float11* f1, Float11* f2)
{
        int res, exp;

        exp = f1->exp + f2->exp;
        res = (((f1->mant * f2->mant) + 0x30) >> 4) << 7;
        res = exp > 26 ? res << (exp - 26) : res >> (26 - exp);
        return (f1->sign ^ f2->sign) ? -res : res;
}

static inline int sgn(int value)
{
   return (value < 0) ? -1 : 1;
}

typedef struct G726Tables {
        int  bits;            /**< bits per sample */
        int* quant;           /**< quantization table */
        int* iquant;          /**< inverse quantization table */
        int* W;               /**< special table #1 ;-) */
        int* F;               /**< special table #2 */
} G726Tables;

typedef struct G726Context {
         G726Tables* tbls;    /**< static tables needed for computation */

         Float11 sr[2];       /**< prev. reconstructed samples */
         Float11 dq[6];       /**< prev. difference */
         int a[2];            /**< second order predictor coeffs */
         int b[6];            /**< sixth order predictor coeffs */
         int pk[2];           /**< signs of prev. 2 sez + dq */

         int ap;              /**< scale factor control */
         int yu;              /**< fast scale factor */
         int yl;              /**< slow scale factor */
         int dms;             /**< short average magnitude of F[i] */
         int dml;             /**< long average magnitude of F[i] */
         int td;              /**< tone detect */

         int se;              /**< estimated signal for the next iteration */
         int sez;             /**< estimated second order prediction */
         int y;               /**< quantizer scaling factor for the next iteration */
} G726Context;

static int quant_tbl16[] =                       /**< 16kbit/s 2bits per sample */
           { 260, INT_MAX };
static int iquant_tbl16[] =
           { 116, 365, 365, 116 };
static int W_tbl16[] =
           { -22, 439, 439, -22 };
static int F_tbl16[] =
           { 0, 7, 7, 0 };

static int quant_tbl24[] =                       /**< 24kbit/s 3bits per sample */
           {  7, 217, 330, INT_MAX };
static int iquant_tbl24[] =
           { INT_MIN, 135, 273, 373, 373, 273, 135, INT_MIN };
static int W_tbl24[] =
           { -4,  30, 137, 582, 582, 137,  30, -4 };
static int F_tbl24[] =
           { 0, 1, 2, 7, 7, 2, 1, 0 };

static int quant_tbl32[] =                       /**< 32kbit/s 4bits per sample */
           { -125,  79, 177, 245, 299, 348, 399, INT_MAX };
static int iquant_tbl32[] =
           { INT_MIN,   4, 135, 213, 273, 323, 373, 425,
                 425, 373, 323, 273, 213, 135,   4, INT_MIN };
static int W_tbl32[] =
           { -12,  18,  41,  64, 112, 198, 355, 1122,
            1122, 355, 198, 112,  64,  41,  18, -12};
static int F_tbl32[] =
           { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };

static int quant_tbl40[] =                      /**< 40kbit/s 5bits per sample */
           { -122, -16,  67, 138, 197, 249, 297, 338,
              377, 412, 444, 474, 501, 527, 552, INT_MAX };
static int iquant_tbl40[] =
           { INT_MIN, -66,  28, 104, 169, 224, 274, 318,
                 358, 395, 429, 459, 488, 514, 539, 566,
                 566, 539, 514, 488, 459, 429, 395, 358,
                 318, 274, 224, 169, 104,  28, -66, INT_MIN };
static int W_tbl40[] =
           {   14,  14,  24,  39,  40,  41,   58,  100,
              141, 179, 219, 280, 358, 440,  529,  696,
              696, 529, 440, 358, 280, 219,  179,  141,
              100,  58,  41,  40,  39,  24,   14,   14 };
static int F_tbl40[] =
           { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
             6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };

static G726Tables G726Tables_pool[] =
           {{ 2, quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
            { 3, quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
            { 4, quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
            { 5, quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};


/**
 * Para 4.2.2 page 18: Adaptive quantizer.
 */
static inline uint8_t quant(G726Context* c, int d)
{
   int sign, exp, i, dln;

   sign = i = 0;
   if (d < 0) {
       sign = 1;
       d = -d;
   }
   exp = av_log2_16bit(d);
   dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);

   while (c->tbls->quant[i] < INT_MAX && c->tbls->quant[i] < dln)
        ++i;

   if (sign)
       i = ~i;
   if (c->tbls->bits != 2 && i == 0) /* I'm not sure this is a good idea */
       i = 0xff;

   return i;
}

/**
 * Para 4.2.3 page 22: Inverse adaptive quantizer.
 */
static inline int16_t inverse_quant(G726Context* c, int i)
{
    int dql, dex, dqt;

    dql = c->tbls->iquant[i] + (c->y >> 2);
    dex = (dql>>7) & 0xf;        /* 4bit exponent */
    dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
    return (dql < 0) ? 0 : ((dqt<<7) >> (14-dex));
}

static inline int16_t g726_iterate(G726Context* c, int16_t I)
{
    int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
    Float11 f;

    dq = inverse_quant(c, I);
    if (I >> (c->tbls->bits - 1))  /* get the sign */
        dq = -dq;
    re_signal = c->se + dq;

    /* Transition detect */
    ylint = (c->yl >> 15);
    ylfrac = (c->yl >> 10) & 0x1f;
    thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
    if (c->td == 1 && abs(dq) > ((thr2+(thr2>>1))>>1))
        tr = 1;
    else
        tr = 0;

    /* Update second order predictor coefficient A2 and A1 */
    pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
    dq0 = dq ? sgn(dq) : 0;
    if (tr) {
        c->a[0] = 0;
        c->a[1] = 0;
        for (i=0; i<6; i++)
           c->b[i] = 0;
    } else {
        /* This is a bit crazy, but it really is +255 not +256 */
        fa1 = clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);

        c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
        c->a[1] = clip(c->a[1], -12288, 12288);
        c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
        c->a[0] = clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);

        for (i=0; i<6; i++)
             c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
    }

    /* Update Dq and Sr and Pk */
    c->pk[1] = c->pk[0];
    c->pk[0] = pk0 ? pk0 : 1;
    c->sr[1] = c->sr[0];
    i2f(re_signal, &c->sr[0]);
    for (i=5; i>0; i--)
       c->dq[i] = c->dq[i-1];
    i2f(dq, &c->dq[0]);
    c->dq[0].sign = I >> (c->tbls->bits - 1); /* Isn't it crazy ?!?! */

    /* Update tone detect [I'm not sure 'tr == 0' is really needed] */
    c->td = (tr == 0 && c->a[1] < -11776);

    /* Update Ap */
    c->dms += ((c->tbls->F[I]<<9) - c->dms) >> 5;
    c->dml += ((c->tbls->F[I]<<11) - c->dml) >> 7;
    if (tr)
       c->ap = 256;
    else if (c->y > 1535 && !c->td && (abs((c->dms << 2) - c->dml) < (c->dml >> 3)))
       c->ap += (-c->ap) >> 4;
    else
       c->ap += (0x200 - c->ap) >> 4;

    /* Update Yu and Yl */
    c->yu = clip(c->y + (((c->tbls->W[I] << 5) - c->y) >> 5), 544, 5120);
    c->yl += c->yu + ((-c->yl)>>6);

    /* Next iteration for Y */
    al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
    c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;

    /* Next iteration for SE and SEZ */
    c->se = 0;
    for (i=0; i<6; i++)
       c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
    c->sez = c->se >> 1;
    for (i=0; i<2; i++)
       c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
    c->se >>= 1;

    return clip(re_signal << 2, -0xffff, 0xffff);
}

static int g726_reset(G726Context* c, int bit_rate)
{
    int i;

    c->tbls = &G726Tables_pool[bit_rate/8000 - 2];
    for (i=0; i<2; i++) {
       i2f(0, &c->sr[i]);
       c->a[i] = 0;
       c->pk[i] = 1;
    }
    for (i=0; i<6; i++) {
       i2f(0, &c->dq[i]);
       c->b[i] = 0;
    }
    c->ap = 0;
    c->dms = 0;
    c->dml = 0;
    c->yu = 544;
    c->yl = 34816;
    c->td = 0;

    c->se = 0;
    c->sez = 0;
    c->y = 544;

    return 0;
}

static int16_t g726_decode(G726Context* c, int16_t i)
{
    return g726_iterate(c, i);
}

#ifdef CONFIG_ENCODERS
static int16_t g726_encode(G726Context* c, int16_t sig)
{
   uint8_t i;

   i = quant(c, sig/4 - c->se) & ((1<<c->tbls->bits) - 1);
   g726_iterate(c, i);
   return i;
}
#endif

/* Interfacing to the libavcodec */

typedef struct AVG726Context {
   G726Context c;
   int bits_left;
   int bit_buffer;
   int code_size;
} AVG726Context;

static int g726_init(AVCodecContext * avctx)
{
    AVG726Context* c = (AVG726Context*)avctx->priv_data;

    if (avctx->channels != 1 ||
        (avctx->bit_rate != 16000 && avctx->bit_rate != 24000 &&
         avctx->bit_rate != 32000 && avctx->bit_rate != 40000)) {
        av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
        return -1;
    }
    if (avctx->sample_rate != 8000 && avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL) {
        av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
        return -1;
    }
    g726_reset(&c->c, avctx->bit_rate);
    c->code_size = c->c.tbls->bits;
    c->bit_buffer = 0;
    c->bits_left = 0;

    avctx->coded_frame = avcodec_alloc_frame();
    if (!avctx->coded_frame)
        return -ENOMEM;
    avctx->coded_frame->key_frame = 1;

    return 0;
}

static int g726_close(AVCodecContext *avctx)
{
    av_freep(&avctx->coded_frame);
    return 0;
}

#ifdef CONFIG_ENCODERS
static int g726_encode_frame(AVCodecContext *avctx,
                            uint8_t *dst, int buf_size, void *data)
{
    AVG726Context *c = avctx->priv_data;
    short *samples = data;
    PutBitContext pb;

    init_put_bits(&pb, dst, 1024*1024);

    for (; buf_size; buf_size--)
       put_bits(&pb, c->code_size, g726_encode(&c->c, *samples++));

    flush_put_bits(&pb);

    return put_bits_count(&pb)>>3;
}
#endif

static int g726_decode_frame(AVCodecContext *avctx,
                             void *data, int *data_size,
                             uint8_t *buf, int buf_size)
{
    AVG726Context *c = avctx->priv_data;
    short *samples = data;
    uint8_t code;
    uint8_t mask;
    GetBitContext gb;

    if (!buf_size)
        goto out;

    mask = (1<<c->code_size) - 1;
    init_get_bits(&gb, buf, buf_size * 8);
    if (c->bits_left) {
        int s = c->code_size - c->bits_left;;
        code = (c->bit_buffer << s) | get_bits(&gb, s);
        *samples++ = g726_decode(&c->c, code & mask);
    }

    while (get_bits_count(&gb) + c->code_size <= buf_size*8)
        *samples++ = g726_decode(&c->c, get_bits(&gb, c->code_size) & mask);

    c->bits_left = buf_size*8 - get_bits_count(&gb);
    c->bit_buffer = get_bits(&gb, c->bits_left);

out:
    *data_size = (uint8_t*)samples - (uint8_t*)data;
    return buf_size;
}

#ifdef CONFIG_ENCODERS
AVCodec adpcm_g726_encoder = {
    "g726",
    CODEC_TYPE_AUDIO,
    CODEC_ID_ADPCM_G726,
    sizeof(AVG726Context),
    g726_init,
    g726_encode_frame,
    g726_close,
    NULL,
};
#endif //CONFIG_ENCODERS

AVCodec adpcm_g726_decoder = {
    "g726",
    CODEC_TYPE_AUDIO,
    CODEC_ID_ADPCM_G726,
    sizeof(AVG726Context),
    g726_init,
    NULL,
    g726_close,
    g726_decode_frame,
};