[FFmpeg-devel] [PATCH] ALAC Encoder

Michael Niedermayer michaelni
Mon Aug 18 22:38:32 CEST 2008


On Tue, Aug 19, 2008 at 01:09:07AM +0530, Jai Menon wrote:
> Hi,
> 
> On Monday 18 Aug 2008 3:46:23 am Michael Niedermayer wrote:
[...]
> > > +
> > > +        sign_modifier = 0;
> > > +        if(x > 0xFFFF)
> > > +            history = 0xFFFF;
> > > +
> > > +        if((history < 128) && (i < s->avctx->frame_size)) {
> > > +            unsigned int block_size = 0;
> > > +
> > >
> > > +            sign_modifier = 1;
> >
> > unused
> >
> 
> It is very much used actually. After we have coded a block of zeros, the next 
> scalar is non_zero and this sign modifier is applied to that.

no, not this :)
see:

> +            sign_modifier = 1;
               ^^^^^^^^^^^^^
> +            k = 7 - av_log2(history) + ((history + 16) >> 6);
> +
> +            while((*samples == 0) && (i < s->avctx->frame_size)) {
> +                samples++;
> +                i++;
> +                block_size++;
> +            }
> +            encode_scalar(s, block_size, k, 16);
> +
> +            sign_modifier = (block_size <= 0xFFFF);
               ^^^^^^^^^^^^^



[...]
> Index: libavcodec/alacenc.c
> ===================================================================
> --- libavcodec/alacenc.c	(revision 14824)
> +++ libavcodec/alacenc.c	(working copy)
> @@ -38,6 +38,11 @@
>  #define ALAC_MAX_LPC_PRECISION    9
>  #define ALAC_MAX_LPC_SHIFT        9
>  
> +#define ALAC_CHMODE_LEFT_RIGHT    0
> +#define ALAC_CHMODE_LEFT_SIDE     1
> +#define ALAC_CHMODE_RIGHT_SIDE    2
> +#define ALAC_CHMODE_MID_SIDE      3
> +
>  typedef struct RiceContext {
>      int history_mult;
>      int initial_history;
> @@ -53,9 +58,12 @@
>  
>  typedef struct AlacEncodeContext {
>      int compression_level;
> +    int min_prediction_order;
> +    int max_prediction_order;
>      int max_coded_frame_size;
>      int write_sample_size;
>      int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
> +    int32_t predictor_buf[DEFAULT_FRAME_SIZE];
>      int interlacing_shift;
>      int interlacing_leftweight;
>      PutBitContext pbctx;
> @@ -116,6 +124,20 @@
>      put_bits(&s->pbctx, 32, s->avctx->frame_size);          // No. of samples in the frame
>  }
>  
> +static void calc_predictor_params(AlacEncodeContext *s, int ch)
> +{
> +    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
> +    int shift[MAX_LPC_ORDER];
> +    int opt_order;
> +
> +    opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch], s->avctx->frame_size, DEFAULT_MIN_PRED_ORDER, DEFAULT_MAX_PRED_ORDER,
> +                                   ALAC_MAX_LPC_PRECISION, coefs, shift, 1, ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
> +
> +    s->lpc[ch].lpc_order = opt_order;
> +    s->lpc[ch].lpc_quant = shift[opt_order-1];
> +    memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
> +}
> +
>  static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
>  {
>      int i, best;
> @@ -147,7 +169,163 @@
>              best = i;
>          }
>      }
> +    return best;
> +}
>  
> +static void alac_stereo_decorrelation(AlacEncodeContext *s)
> +{
> +    int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
> +    int i, mode, n = s->avctx->frame_size;
> +    int32_t tmp;
> +
> +    mode = estimate_stereo_mode(left, right, n);
> +
> +    switch(mode)
> +    {
> +        case ALAC_CHMODE_LEFT_RIGHT:
> +            s->interlacing_leftweight = 0;
> +            s->interlacing_shift = 0;
> +            break;
> +
> +        case ALAC_CHMODE_LEFT_SIDE:
> +            for(i=0; i<n; i++) {
> +                right[i] = left[i] - right[i];
> +            }
> +            s->interlacing_leftweight = 1;
> +            s->interlacing_shift = 0;
> +            break;
> +
> +        case ALAC_CHMODE_RIGHT_SIDE:
> +            for(i=0; i<n; i++) {
> +                tmp = right[i];
> +                right[i] = left[i] - right[i];
> +                left[i] = tmp + (right[i] >> 31);
> +            }
> +            s->interlacing_leftweight = 1;
> +            s->interlacing_shift = 31;
> +            break;
> +
> +        default:
> +            for(i=0; i<n; i++) {
> +                tmp = left[i];
> +                left[i] = (tmp + right[i]) >> 1;
> +                right[i] = tmp - right[i];
> +            }
> +            s->interlacing_leftweight = 1;
> +            s->interlacing_shift = 1;
> +            break;
> +    }
> +}

ok


> +
> +static void alac_linear_predictor(AlacEncodeContext *s, int ch)
> +{
> +    int i;
> +    LPCContext lpc = s->lpc[ch];
> +
> +    if(lpc.lpc_order == 31) {
> +        s->predictor_buf[0] = s->sample_buf[ch][0];
> +
> +        for(i=1; i<s->avctx->frame_size; i++)
> +            s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
> +
> +        return;
> +    }
> +
> +    // generalised linear predictor
> +
> +    if(lpc.lpc_order > 0) {
> +        int32_t *samples  = s->sample_buf[ch];
> +        int32_t *residual = s->predictor_buf;
> +
> +        // generate warm-up samples
> +        residual[0] = samples[0];
> +        for(i=1;i<=lpc.lpc_order;i++)
> +            residual[i] = samples[i] - samples[i-1];
> +
> +        // perform lpc on remaining samples
> +        for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {

> +            int sum = 0, res_val, j;
> +
> +            for (j = 0; j < lpc.lpc_order; j++) {
> +                sum += (samples[lpc.lpc_order-j] - samples[0]) *
> +                        lpc.lpc_coeff[j];
> +            }

> +            sum += (1 << (lpc.lpc_quant - 1));

the added value can be use instead of 0 to initialize sum



> +            sum >>= lpc.lpc_quant;
> +            sum += samples[0];
> +            residual[i] = samples[lpc.lpc_order+1] - sum;
> +            res_val = residual[i];
> +

> +            if(res_val) {
> +                int index = lpc.lpc_order - 1;
> +                int neg = (res_val < 0);
> +
> +                while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
> +                    int val = samples[0] - samples[lpc.lpc_order - index];
> +                    int sign = (val ? FFSIGN(val) : 0);
> +
> +                    if(neg)
> +                        sign*=-1;
> +
> +                    lpc.lpc_coeff[index] -= sign;
> +                    val *= sign;
> +                    res_val -= ((val >> lpc.lpc_quant) *
> +                            (lpc.lpc_order - index));
> +                    index--;
> +                }
> +            }

this part is also used in the decoder, thus could be factorized into
a common function
alac_entropy_coder() also has some common code with the decoder but it
is maybe too much interleaved with non-common code to factorize.


[...]
> @@ -226,6 +404,32 @@
>          AV_WB8(alac_extradata+20, s->rc.k_modifier);
>      }
>  
> +    if(avctx->min_prediction_order >= 0) {
> +        if(avctx->min_prediction_order < MIN_LPC_ORDER ||
> +            avctx->min_prediction_order > MAX_LPC_ORDER) {
> +            av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
> +                return -1;
> +        }
> +
> +        s->min_prediction_order = avctx->min_prediction_order;
> +    }
> +
> +    if(avctx->max_prediction_order >= 0) {
> +        if(avctx->max_prediction_order < MIN_LPC_ORDER ||
> +           avctx->max_prediction_order > MAX_LPC_ORDER) {
> +            av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
> +                return -1;
> +        }
> +
> +        s->max_prediction_order = avctx->max_prediction_order;
> +    }
> +
> +    if(s->max_prediction_order < s->min_prediction_order) {
> +        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
> +               s->min_prediction_order, s->max_prediction_order);
> +        return -1;
> +    }
> +
>      avctx->extradata = alac_extradata;
>      avctx->extradata_size = ALAC_EXTRADATA_SIZE;
>  
> @@ -255,6 +459,9 @@
>          return -1;
>      }
>  
> +verbatim:
> +    init_put_bits(pb, frame, buf_size);
> +
>      if((s->compression_level == 0) || verbatim_flag) {
>          // Verbatim mode
>          int16_t *samples = data;

ok

btw, if you want a svn write account, send diego your wanted username
and password (GPG encrypted!)
Assuming that you agree to our svn policy and are willing to maintain your
encoder after SOC.

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

There will always be a question for which you do not know the correct awnser.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080818/71f82d1f/attachment.pgp>



More information about the ffmpeg-devel mailing list