[Shootout-list] What people say about shootout

[Dave]

All I can add is that single precision /can/ be faster when you are moving a
lot of data around, but in some cases it won't matter because the important
optimizations are done in 80 bit FP registers anyhow (I'm talking x86
hardware here).

What a single precision test might be good for is pointing out:

1) Issues for implementations that support both SP and DP (where SP actually
runs slower because of alignment issues, for example).

2) Issues where the language doesn't support 32 bit floats. Even if it isn't
obvious from the test, those implementations may run quite a bit slower (for
applications moving a lot of floating point data around where single
precision is good enough precision), which would be good to know. Or, they
could run just as fast which would be even better to know.

- Dave

> -----Original Message-----
> From: Isaac Gouy [mailto:igouy2@yahoo.com]
> Sent: Friday, April 29, 2005 3:01 PM
> To: Sebastien Loisel; Dave
> Cc: shootout-list@lists.alioth.debian.org
> Subject: Re: [Shootout-list] What people say about shootout
>
>
>
> --- Sebastien Loisel <sloisel@gmail.com> wrote:
> > > (...) why are all of
> > > the floating point benchmarks I see out there (not just the
> > Shootout) double
> > > precision when in reality single precision (...)
> >
> > The problem I faced in implicitode is that a large fraction of
> > languages don't even have a single precision data type (e.g.,
> > python.)
>
> (Lua also, actually you choose to build Lua for either double or int.)
>
> Please correct me where I'm wrong, it seems that
> - single precision is faster than double precision
> - some languages only provide double precision
> - we might code the same algorithm using doubles and get the same
> result
> - the practical consequence is that Python would be slower than
> languages that provide distinct single precision data type
>
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>
>