[vox-tech] float vs double, int vs short

[Micah J. Cowan]

On Thu, Feb 16, 2006 at 11:21:11AM -0500, Peter Jay Salzman wrote:
> I've been googling for the past 15 minutes and have read that:
> 
>    * computations with doubles are always faster than floats
>    * computations with doubles are often faster than floats
>    * computations with floats are often faster than doubles
>    * computations with floats are always faster than doubles
> 
> The arguments I've read for all four cases seem reasonable.  I trust the
> collective brains of lugod more than any internet forum.  What's the real
> scoop on the relative speed of basic arithmetic operations on floats and
> doubles?
> 
> The platforms I'm mostly concerned with are 32 and 64 bit x86 running
> Windows and Linux using Java and C++.

This is outside of my specific experience, but...

I'm fairly certain that floats vs doubles is an architecture-independant
problem. However, on a 32- or 64- bit system with MPU, I'm also
fairly certain that doubles are faster than floats.

> I'm also interested in the relative speed of shorts vs ints.  I've
> heard that ints are almost always faster (with the "usual" width of an
> int in C, since an int and short may have the same width on a given
> platform) due to word alignment issues.

I think this is true in general: however, x86 has a large selection of
instructions tailored to a variety of widths; I think the difference
would be negligible.

However, on some architectures, a short might need to be "expanded" into
an int before it can be worked with. In fact, C actually /requires/ that
this be done (or that the implementation behave "as if" it were being
done); for that reason, some poorly optimized compilers (or indeed, gcc
with optimizations off--I'd have to look at the output) may convert it
to int and then convert the results back to short before storage.

> I did my own test of taking averages of a large number of numbers in C
> with shorts and ints on Cygwin on WinXP.  The results were pretty
> inconclusive.

Which should mean that the differences are negligible. Although, you
should try it with optimizations and see if it makes a difference.

Remember, I may just be talking out my ass on this--it's all AIUI.

-- 
Micah J. Cowan
micah at cowan.name