float vs double

Roberta Eklund eklund at best.com
Wed Mar 24 17:38:42 EST 1999



>
> Also, interestingly, I find double precision calculation to be quite a bit
> slower than single precision, even though the FPU internally performs all
> calculation at double precision or greater.  I suspect the slow
> down is due
> to the fact that twice as much data must be shuffled around from memory to
> the FPU and back.
yup.  once the data is in the cache, clock hits should be the same.  It's
just doing 2x
memory loads before that.  32-bit alignment is how memory accesses are made
on IA.
In the Pentium Data book, most cycle counts are the same for double and
float.

Speaking of which...does anybody here like to read rough drafts of papers
for suggestions
to improve?  I'm looking for a volunteer, but don't want to protrude on
anyone.  ;)
>
>
> Chris
>
>
> dupswapdrop: the music-dsp mailing list and website
> http://shoko.calarts.edu/~glmrboy/musicdsp/music-dsp.html
>
>


dupswapdrop: the music-dsp mailing list and website
http://shoko.calarts.edu/~glmrboy/musicdsp/music-dsp.html




More information about the music-dsp mailing list