[FFmpeg-devel] [PATCH] Limit Rice parameter for progressive decoding in ALS

[Justin Ruggles]

Thilo Borgmann wrote:

> Hi,
> 
> as in $subject, the reference encoder limits the Rice parameter during
> progressively coding the first few samples of a random access block.
> 
> Yet another "feature" not specified in 14496-3...

Fun. Nice catch.


> Regards,
> Thilo
> 
> 

> Index: libavcodec/alsdec.c
> ===================================================================
> --- libavcodec/alsdec.c	(Revision 21799)
> +++ libavcodec/alsdec.c	(Arbeitskopie)
> @@ -192,6 +192,7 @@
>      unsigned int frame_id;          ///< the frame ID / number of the current frame
>      unsigned int js_switch;         ///< if true, joint-stereo decoding is enforced
>      unsigned int num_blocks;        ///< number of blocks used in the current frame
> +    unsigned int s_max;             ///< maximum Rice parameter allowed in entropy coding
>      uint8_t *bgmc_lut;              ///< pointer at lookup tables used for BGMC
>      unsigned int *bgmc_lut_status;  ///< pointer at lookup table status flags used for BGMC
>      int ltp_lag_length;             ///< number of bits used for ltp lag value
> @@ -720,9 +721,9 @@
>          if (opt_order)
>              bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
>          if (opt_order > 1)
> -            bd->raw_samples[1] = decode_rice(gb, s[0] + 3);
> +            bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max));
>          if (opt_order > 2)
> -            bd->raw_samples[2] = decode_rice(gb, s[0] + 1);
> +            bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max));
>  
>          start = FFMIN(opt_order, 3);
>      }
> @@ -1508,6 +1509,11 @@
>          avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
>      }
>  
> +    // set maximum Rice parameter for progressive decoding based on resolution
> +    // This is not specified in 14496-3 but actually done by the reference
> +    // codec RM22 revision 2.
> +    ctx->s_max = (16 << (sconf->resolution > 1)) - 1;
> +

It seems simpler to me to just do:
ctx->s_max = sconf->resolution > 1 ? 31 : 15;


-Justin