Hi folks,
I measured how many u += a % b operations my Celeron 366
is able to perform in one sec.
(an example of the % operation would be to use it in boundary clipping
code)
about 9million ops/sec , disappointing,
( it uses the idivl operation).
when I use & (and) instead of % , the speed is boosted
to 150million ops/sec ! , almost factor 16 !
But to use & instead of % you need that in the
a % b operation b is a power of 2.
( in this case a %b is replaced with a & bminus1
bminus1=b - 1 :-) )
In the audio field we could use this optimization,
when the buffersize is a power of 2.
I just grepped through the ALSA driver code and there
are some % frag_size operations.
I think on almost all cards frag_size (the fragment size) is a power
of two, therfore we could replace the % with the &.
ALSA does a few % operations per every played fragment.
But even if & is 16 times faster than % , using very small fragments
like 1msec fragments, and assuming ALSA does 5 % operations per fragment.
we would save at maximum:
1000 (frags/sec) * 5 (% ops per fragment) = 5000 ops per sec
every % operation takes 0.11 usecs ( 1 / 9000000)
5000*0.11 = 550 usecs saved (assumed that & has zero cost) per second.
that is about 550 usec / 1000000usecs = an insignificant 0.55 % performance
gain.
not worth the trouble IMHO
:-)
do you agree ?
PS: getting too paranoid sometimes , eh ? :-)
Benno.
