I went ahead and did it just to see what happened.
Yes it was time consuming to change everything. I was disappointed to find
no noticeable improvement in performance however. My current application
takes input of xyz triplets and regresses a surface from them, showing the
original points with anchors, and a base grid. I've been working with about
500 points, from which I then calculate a 20 X 20 QuadArray grid.
The mesh density of the QuadArray grid is variable, so I suppose I might
notice a greater difference in performance with a higher mesh density, say
100 X 100, but so far it seems the exercise wasn't worth the trouble.
Here's a jpg of the graph if anyone's interested:
-----Original Message-----
From: Tim Bray [mailto:[EMAIL PROTECTED]]
Sent: Monday, November 01, 1999 12:58 PM
To: [EMAIL PROTECTED]
Subject: Re: [JAVA3D] float -vs- double
At 12:46 PM 11/1/99 -0500, Casteel, Don wrote:
>Is it reasonable to assume I will see a significant improvement in speed if
>I take the time to convert everything that's double precision to float
>precision?
You'll suffer a significant *decrease* in the speed of writing your code
because of having to type (float) everywhere and put letter 'f' after every
number and having to recompile everything 3 times because you forgot to
type (float) and 0.5f and things like that... I used doubles everywhere for
just that reason, my time is more expensive than the computer's... Now
clearly
it's going to double your memory burn for big arrays of these things, which
is
a significant performance drag, but will it hurt performance otherwise?
Good question. -T.
===========================================================================
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff JAVA3D-INTEREST". For general help, send email to
[EMAIL PROTECTED] and include in the body of the message "help".
Capture0.jpg
Capture0.jpg