[FFmpeg-devel] [PATCH 2/5] Add dot_product function for use by the G.723.1 decoder

[Vitor Sessak]

On 03/17/2011 11:56 PM, banan at ludd.ltu.se wrote:
> From: Mohamed Naufal Basheer<naufal11 at gmail.com>
>
> ---
>   libavcodec/celp_math.c |   12 ++++++++++++
>   libavcodec/celp_math.h |   11 +++++++++++
>   2 files changed, 23 insertions(+), 0 deletions(-)
>
> diff --git a/libavcodec/celp_math.c b/libavcodec/celp_math.c
> index 09111da..b78edd1 100644
> --- a/libavcodec/celp_math.c
> +++ b/libavcodec/celp_math.c
> @@ -25,6 +25,7 @@
>   #include<assert.h>
>
>   #include "avcodec.h"
> +#include "mathops.h"
>   #include "celp_math.h"
>
>   #ifdef G729_BITEXACT
> @@ -196,6 +197,17 @@ int ff_log2(uint32_t value)
>       return (power_int<<  15) + value;
>   }
>
> +int ff_dot_product(const int16_t *a, const int16_t *b, int length, int shift)
> +{
> +    int i, sum = 0;
> +
> +    for (i = 0; i<  length; i++) {
> +        int64_t prod = av_clipl_int32(MUL64(a[i], b[i])<<  shift);
> +        sum = av_clipl_int32(sum + prod);
> +    }
> +    return sum;
> +}

Hmm... I don't understand why this is not implemented like

int64_t ff_dot_product(const int16_t *a, const int16_t *b, int length)
{
     int i;
     int64_t sum = 0;

     for (i = 0; i < length; i++)
         sum += MUL16(a[i], b[i]);

     return sum;
}

and then

av_clipl_int32(ff_dot_product(a, b, len) << shift);
?

Is the reason that clipping as in your code necessary for the decoder to 
be bitexact? IMHO, if it is so, I think this function is too much G723.1 
specific to be in common code...

-Vitor