[FFmpeg-devel] [PATCH] Add x86-optimized function ac3_or_abs_int16() and use in log2_tab().

[Måns Rullgård]

Justin Ruggles <justin.ruggles at gmail.com> writes:

> diff --git a/libavcodec/ac3dsp.h b/libavcodec/ac3dsp.h
> index 7f13b11..a6a9af4 100644
> --- a/libavcodec/ac3dsp.h
> +++ b/libavcodec/ac3dsp.h
> @@ -35,6 +35,17 @@ typedef struct AC3DSPContext {
>       * @param nb_coefs  number of frequency coefficients.
>       */
>      void (*ac3_exponent_min)(uint8_t *exp, int num_reuse_blocks, int nb_coefs);
> +
> +    /**
> +     * Calculate the logical 'or' of the absolute value of each element in an
> +     * array of int16_t.
> +     * @param src input array
> +     *            constraints: align 16
> +     * @param len number of values in the array
> +     *            constraints: multiple of 16 greater than 0
> +     * @return    abs(src[0]) | abs(src[1]) | ... | abs(src[len-1])
> +     */
> +    int (*ac3_or_abs_int16)(const int16_t *src, int len);
>  } AC3DSPContext;

What is the range of the input values?  Specifically, is INT16_MIN allowed?
Whatever the answer, please mention it in the comment.

>  void ff_ac3dsp_init    (AC3DSPContext *c);
> diff --git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c
> index 0db41df..d1f15ac 100644
> --- a/libavcodec/ac3enc_fixed.c
> +++ b/libavcodec/ac3enc_fixed.c
> @@ -270,14 +270,9 @@ static void apply_window(DSPContext *dsp, int16_t *output, const int16_t *input,
>   * @param n   number of values in the array
>   * @return    log2(max(abs(tab[])))
>   */
> -static int log2_tab(int16_t *tab, int n)
> +static int log2_tab(AC3EncodeContext *s, int16_t *tab, int n)
>  {
> -    int i, v;
> -
> -    v = 0;
> -    for (i = 0; i < n; i++)
> -        v |= abs(tab[i]);
> -
> +    int v = s->ac3dsp.ac3_or_abs_int16(tab, n);
>      return av_log2(v);
>  }
>
> @@ -308,7 +303,7 @@ static void lshift_tab(int16_t *tab, int n, unsigned int lshift)
>   */
>  static int normalize_samples(AC3EncodeContext *s)
>  {
> -    int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE);
> +    int v = 14 - log2_tab(s, s->windowed_samples, AC3_WINDOW_SIZE);
>      lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v);
>      return v - 9;
>  }

C parts look OK.  I'm leaving x86 bits to someone else.

-- 
M?ns Rullg?rd
mans at mansr.com