[FFmpeg-devel] [PATCH] avcodec/nellymoserenc: avoid wasteful pow

[Ganesh Ajjanagadde]

On Tue, Dec 15, 2015 at 5:25 PM, Ganesh Ajjanagadde <gajjanag at mit.edu> wrote:
> On Tue, Dec 15, 2015 at 2:23 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> On Wed, Dec 09, 2015 at 06:55:25PM -0500, Ganesh Ajjanagadde wrote:
[...]
>>>
>>> diff --git a/libavcodec/nellymoserenc.c b/libavcodec/nellymoserenc.c
>>> index d998dba..e6023e3 100644
>>> --- a/libavcodec/nellymoserenc.c
>>> +++ b/libavcodec/nellymoserenc.c
>>> @@ -179,8 +179,15 @@ static av_cold int encode_init(AVCodecContext *avctx)
>>>
>>>      /* Generate overlap window */
>>>      ff_init_ff_sine_windows(7);
>>> -    for (i = 0; i < POW_TABLE_SIZE; i++)
>>> -        pow_table[i] = pow(2, -i / 2048.0 - 3.0 + POW_TABLE_OFFSET);
>>> +    pow_table[0] = 1;
>>> +    pow_table[1024] = M_SQRT1_2;
>>> +    for (i = 1; i < 513; i++) {
>>> +        double tmp = exp2(-i / 2048.0);
>>> +        pow_table[i] = tmp;
>>> +        pow_table[1024-i] = M_SQRT1_2 / tmp;
>>> +        pow_table[1024+i] = tmp * M_SQRT1_2;
>>> +        pow_table[2048-i] = 0.5 / tmp;
>>
>> how much overall init time is gained by this ?
>> that is time in ffmpeg main() from start to finish when just opening
>> the file with no decoding aka ./ffmpeg -i somefile
>
> Don't know, all I know is cycles are unnecessarily wasted. Will put in
> cycle numbers.
>

Here they are:
proposed: 424160 decicycles in pow_table,     512 runs,      0 skips
exp2 only: 1262093 decicycles in pow_table,     512 runs,      0 skips
old: 2849085 decicycles in pow_table,     512 runs,      0 skips

Thus old to exp2 is roughly 2.25x speedup, exp2 to proposed roughly 3x
speedup, net ~ 6.7x speedup.