[FFmpeg-devel] [PATCH 3/3] lavc/wmadec: replace pow by faster functions

[Ganesh Ajjanagadde]

On Sat, Jan 16, 2016 at 12:31 AM, Ganesh Ajjanagadde
<gajjanagadde at gmail.com> wrote:
> Further speedups possible by getting rid of exp2f...
>
> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
> ---
>  libavcodec/wmadec.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/libavcodec/wmadec.c b/libavcodec/wmadec.c
> index da54182..1a84323 100644
> --- a/libavcodec/wmadec.c
> +++ b/libavcodec/wmadec.c
> @@ -35,6 +35,7 @@
>
>  #include "libavutil/attributes.h"
>  #include "libavutil/internal.h"
> +#include "libavutil/libm.h"
>
>  #include "avcodec.h"
>  #include "internal.h"
> @@ -164,7 +165,7 @@ static av_cold void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
>      /* tables for x^-0.25 computation */
>      for (i = 0; i < 256; i++) {
>          e                     = i - 126;
> -        s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
> +        s->lsp_pow_e_table[i] = exp2f(e * -0.25);
>      }
>
>      /* NOTE: these two tables are needed to avoid two operations in
> @@ -173,7 +174,7 @@ static av_cold void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
>      for (i = (1 << LSP_POW_BITS) - 1; i >= 0; i--) {
>          m                      = (1 << LSP_POW_BITS) + i;
>          a                      = (float) m * (0.5 / (1 << LSP_POW_BITS));
> -        a                      = pow(a, -0.25);
> +        a                      = 1/sqrt(sqrt(a));
>          s->lsp_pow_m_table1[i] = 2 * a - b;
>          s->lsp_pow_m_table2[i] = b - a;
>          b                      = a;
> --
> 2.7.0
>

ping, note that second can be made sqrtf(sqrtf(x)) at some error which
I deliberately avoided here.