Floating point performance (decimal vs double)

20 Jan 2006 2:35 PM

Jeff

Hi-

I am developing a WinForms app with the dotnet 2.0 framework that
manipulates GPS data. Some of the algoritms are pretty math intensive,
with Sin, Cos, Log, etc.

I originally developed the app using the type "double" for all my data
points, but as an experiment, I changed the types to "decimal" and was
surprised to see that the performance improved. Is this expected?
FYI, I am running this app on a Turion ML-34 (1.8Ghz 1MB cache), so my
fp unit should be pretty good.

Are there any good resources on tweaking floating point performance in
dotnet? I was very surprised to see decimal out perform double, so I'm
wondering if I'm missing something else. For example, what about
exetended floating point types? I imagine that this may be even faster
since it would use the SSE unit of the chip.

Any input is appreciated.

-jeff

24 Jan 2006 8:37 AM

Dmytro Lapshyn [MVP]

Hi,

Some food for thought - decimal is actually a *fixed* point data type (as
opposed to *floating* point). And the former has been known as a faster one
since times of ID Software's Quake 1 (which used fixed point math for 3D
calculations all the way).

Given you are working with coordinates, which are unlikely to have a large
number of digits after the decimal point (I might be wrong though), you
might be better off using the decimal data type (and having a benefit of
predicted precision of all your calculations).

Show quote

"Jeff" <j***@kalikstein.com> wrote in message
news:1137767722.560247.87520@g43g2000cwa.googlegroups.com...
> Hi-
>
> I am developing a WinForms app with the dotnet 2.0 framework that
> manipulates GPS data. Some of the algoritms are pretty math intensive,
> with Sin, Cos, Log, etc.
>
> I originally developed the app using the type "double" for all my data
> points, but as an experiment, I changed the types to "decimal" and was
> surprised to see that the performance improved. Is this expected?
> FYI, I am running this app on a Turion ML-34 (1.8Ghz 1MB cache), so my
> fp unit should be pretty good.
>
> Are there any good resources on tweaking floating point performance in
> dotnet? I was very surprised to see decimal out perform double, so I'm
> wondering if I'm missing something else. For example, what about
> exetended floating point types? I imagine that this may be even faster
> since it would use the SSE unit of the chip.
>
> Any input is appreciated.
>
> -jeff
>