[FFmpeg-devel] [PATCH 2/4] avcodec/mpegaudio_tablegen: speed up dynamic table creation

[Ganesh Ajjanagadde]

On Sat, Nov 28, 2015 at 12:03 PM, Ganesh Ajjanagadde
<gajjanagadde at gmail.com> wrote:
> This does some miscellaneous stuff mainly avoiding the usage of pow to
> achieve significant speedups. This is not speed critical, but is
> unnecessary latency and cycles wasted for a user.
>
> All tables tested and are identical to the old ones
> (bit-exact even in floating point case).
>
> Sample benchmark (x86-64, Haswell, GNU/Linux):
> old:
> 102329530 decicycles in mpegaudio_tableinit,       1 runs,      0 skips
>
> new:
> 34111900 decicycles in mpegaudio_tableinit,       1 runs,      0 skips
>
> Reviewed-by: Ronald S. Bultje <rsbultje at gmail.com>
> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
> ---
>  libavcodec/mpegaudio_tablegen.c |  1 +
>  libavcodec/mpegaudio_tablegen.h | 19 +++++++++++++------
>  2 files changed, 14 insertions(+), 6 deletions(-)
>
> diff --git a/libavcodec/mpegaudio_tablegen.c b/libavcodec/mpegaudio_tablegen.c
> index 90c9de4..9a9bb4d 100644
> --- a/libavcodec/mpegaudio_tablegen.c
> +++ b/libavcodec/mpegaudio_tablegen.c
> @@ -23,6 +23,7 @@
>  #include <stdlib.h>
>  #define CONFIG_HARDCODED_TABLES 0
>  #include "mpegaudio_tablegen.h"
> +#include "libavutil/tablegen.h"
>  #include "tableprint.h"
>
>  int main(void)
> diff --git a/libavcodec/mpegaudio_tablegen.h b/libavcodec/mpegaudio_tablegen.h
> index 86b2cd3..dd67a09 100644
> --- a/libavcodec/mpegaudio_tablegen.h
> +++ b/libavcodec/mpegaudio_tablegen.h
> @@ -45,12 +45,21 @@ static float expval_table_float[512][16];
>  static av_cold void mpegaudio_tableinit(void)
>  {
>      int i, value, exponent;
> +    double exp2_lut[4] = {
> +        1.00000000000000000000, /* 2 ^ (0 * 0.25) */
> +        1.18920711500272106672, /* 2 ^ (1 * 0.25) */
> +        M_SQRT2               , /* 2 ^ (2 * 0.25) */
> +        1.68179283050742908606, /* 2 ^ (3 * 0.25) */
> +    };
> +    double cbrt_lut[16];
> +    for (i = 0; i < 16; ++i)
> +        cbrt_lut[i] = cbrt(i);
> +
>      for (i = 1; i < TABLE_4_3_SIZE; i++) {
>          double value = i / 4;
>          double f, fm;
>          int e, m;
> -        /* cbrtf() isn't available on all systems, so we use powf(). */
> -        f  = value / IMDCT_SCALAR * pow(value, 1.0 / 3.0) * pow(2, (i & 3) * 0.25);
> +        f  = value / IMDCT_SCALAR * cbrt(value) * exp2_lut[i & 3];
>          fm = frexp(f, &e);
>          m  = (uint32_t)(fm * (1LL << 31) + 0.5);
>          e += FRAC_BITS - 31 + 5 - 100;
> @@ -61,10 +70,8 @@ static av_cold void mpegaudio_tableinit(void)
>      }
>      for (exponent = 0; exponent < 512; exponent++) {
>          for (value = 0; value < 16; value++) {
> -            /* cbrtf() isn't available on all systems, so we use powf(). */
> -            double f = (double)value * pow(value, 1.0 / 3.0) * pow(2, (exponent - 400) * 0.25 + FRAC_BITS + 5) / IMDCT_SCALAR;
> -            /* llrint() isn't always available, so round and cast manually. */
> -            expval_table_fixed[exponent][value] = (long long int) (f < 0xFFFFFFFF ? floor(f + 0.5) : 0xFFFFFFFF);
> +            double f = value * cbrt_lut[value] * pow(2, (exponent - 400) * 0.25 + FRAC_BITS + 5) / IMDCT_SCALAR;
> +            expval_table_fixed[exponent][value] = (f < 0xFFFFFFFF ? llrint(f) : 0xFFFFFFFF);
>              expval_table_float[exponent][value] = f;
>          }
>          exp_table_fixed[exponent] = expval_table_fixed[exponent][1];
> --
> 2.6.2
>

pushed