[FFmpeg-devel] PATCH: COOK audio decode infastructure to support fixpoint optimization

[Benjamin Larsson]

Marc Hoffman wrote:
> This patch changes the way the data flows by introducing a vec_t data
> type which is just unsigned char. This way we can abstract the low
> level data type itself keeping the logic the same.  I'm using a char
> type so we can walk through the data this is not possible with voids
> to my knowledge. I have chosen to use vec_t so that we can easily
> locate the vectors which represent the input and output signals of the
> system.

Ok, I don't know if this is the best way to do it, but I'll assume it is
for now.

> 
> I will then add a set of overloaded primitives to the existing
> structure that handle the lowlevel stuff like quantization, decouple,
> satruate, interpolate, and imlt etc that needs to know if your working
> with float or shorts.  I want to keep all the logic the same so I will
> also introduce a fixpoint fft which I'm in the process of developing
> under the ffmpeg license as GPL code.

No chance for a LGPL license ? And if you need a reference fp fft to
look at use this:

http://svn.rockbox.org/viewvc.cgi/trunk/apps/codecs/libwma/wmadeci.c?revision=13859&view=markup

> 
> Please review:
> 
> 
> 
> ------------------------------------------------------------------------
> 
> Index: libavcodec/cook.c
> ===================================================================
> --- libavcodec/cook.c	(revision 9441)
> +++ libavcodec/cook.c	(working copy)
> @@ -61,6 +61,8 @@
>  #define JOINT_STEREO    0x1000003
>  #define MC_COOK         0x2000000   //multichannel Cook, not supported
>  
> +#define vec_t uint8_t
> +
>  #define SUBBAND_SIZE    20
>  //#define COOKDEBUG
>  
> @@ -69,7 +71,43 @@
>      int *previous;
>  } cook_gains;
>  
> -typedef struct {
> +
> +
> +typedef struct cook {
> +
> +    /*
> +      joint_decode: creates mlt_buffer1, mlt_buffer2
> +
> +	mono_decode: creates decode_buffer
> +	     ... category handling all integer and bitstream related...
> +
> +	     decode_vectors: creates mlt_buffer from categories.
> +		 scalar_dequant: for each block of 20 elements ie SUBBAND_SIZE. 32bit-floats
> +		    (quant_centroid_tab[index][subband_coef_index[i]],
> +		     dither_tab[index],
> +		     rootpow2tab[quant_index+63])
> +
> +	     mlt_compensate_output:
> +		 imlt_gain: inverse modified dct + windowing all dones as 32bit-floats

dones

> +		    (mlt_window)
> +		 saturate to 16bit integers
> +
> +	then decode_buffer is separated into mlt_buffer1, mlt_buffer2 via cplscale multipliers.
> +    */
> +    void (* scalar_dequant)(struct cook *q, int index, int quant_index,
> +			    int* subband_coef_index, int* subband_coef_sign,
> +			    vec_t* mlt_p);
> +
> +    void (* decouple) (struct cook *q, int *decouple_tab,
> +		       vec_t *decode_buffer, vec_t *mlt_buffer1, vec_t *mlt_buffer2);
> +
> +    void (* interpolate) (struct cook *q, vec_t* buffer, int gain_index, int gain_index_next);
> +
> +    void (* saturate_output) (struct cook *q, int chan, int16_t *out);
> +
> +
> +    int f_sample_size;
>
> [...]
>
> +
> +static void joint_decode(COOKContext *q, vec_t* mlt_buffer1,
> +                         vec_t* mlt_buffer2) {
> +    int decouple_tab[SUBBAND_SIZE];
> +    vec_t *decode_buffer;
> +
> +    decode_buffer = av_malloc (1060*q->f_sample_size);

Don't allocate memory with malloc during decode. Use a fixed max size.

> +    memset(decouple_tab, 0, sizeof(decouple_tab));
> +    memset(decode_buffer, 0, 1060*q->f_sample_size);
> +
> +    /* Make sure the buffers are zeroed out. */
> +    memset(mlt_buffer1,0, 1024*q->f_sample_size);
> +    memset(mlt_buffer2,0, 1024*q->f_sample_size);
> +    decouple_info(q, decouple_tab);
> +    mono_decode(q, decode_buffer);
> +    q->decouple (q, decouple_tab, decode_buffer, mlt_buffer1, mlt_buffer2);
> +    av_free (decode_buffer);
> +}
> +

To me it looks ok but are you sure the f_sample_size thing is needed ?
It should be 4 for both float and fp and if it is set to that lots of
logic is simplified.

MvH
Benjamin Larsson