[FFmpeg-devel] [PATCHv2 2/3] avcodec/mpegaudio_tablegen: speed up table generation

[Ronald S. Bultje]

Hi,

On Thu, Nov 26, 2015 at 10:23 AM, 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
>
> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
> ---
>  libavcodec/mpegaudio_tablegen.h | 21 +++++++++++++++------
>  1 file changed, 15 insertions(+), 6 deletions(-)
>
> diff --git a/libavcodec/mpegaudio_tablegen.h
> b/libavcodec/mpegaudio_tablegen.h
> index 86b2cd3..6cd5ad6 100644
> --- a/libavcodec/mpegaudio_tablegen.h
> +++ b/libavcodec/mpegaudio_tablegen.h
> @@ -29,9 +29,11 @@
>
>  #define TABLE_4_3_SIZE (8191 + 16)*4
>  #if CONFIG_HARDCODED_TABLES
> +#include "libavutil/tablegen.h"
>  #define mpegaudio_tableinit()
>  #include "libavcodec/mpegaudio_tables.h"
>  #else
> +#include "libavutil/libm.h"
>

I don't think this works yet. Check mpegaudio_tablegen.c (it should be
obvious why it does this):

#include <stdlib.h>
#define CONFIG_HARDCODED_TABLES 0
#include "mpegaudio_tablegen.h"
#include "tableprint.h"

Note that this is the table generation which runs on the host platform. So,
the location where you add this, is important. I would recommend including
tablegen.h inside mpegaudio_tablegen.c, and then including libavutil/libm.h
outside the scope of this header, e.g. in users of this header, if
tablegen.h wasn't included. Make sure make checkheaders passes after your
changes, and the build works with as well as without hardcoded_tables.

 static int8_t   table_4_3_exp[TABLE_4_3_SIZE];
>  static uint32_t table_4_3_value[TABLE_4_3_SIZE];
>  static uint32_t exp_table_fixed[512];
> @@ -45,12 +47,21 @@ static float expval_table_float[512][16];
>  static av_cold void mpegaudio_tableinit(void)
>  {
>      int i, value, exponent;
> +    double pow2_lut[4] = {
>

I think somebody else already mentioned exp2. I'd call this table exp2,
since table[idx] corresponds to exp2(idx * constant).


> +            double f = value * cbrt_lut[value] * pow(2, (exponent - 400)
> * 0.25 + FRAC_BITS + 5) / IMDCT_SCALAR;


exp2 here also. For compatibility, you can #define exp2(x) to pow(2, x) in
libavutil/tablegen.h.

Ronald