Re: Speed Comparison

[Fernando Salas <fsalas@fibertel.com.ar>]

AFAIK, Any good modern compiler wil use case B when compiling the case a
source, because the optimizer detects constants for the loop and it put them
outside it.
I think u are trying to optimize at a level that the compiler will do better
job. Can't u optimize the algorithms instead? Are u sure they arre optimized ,
u will have much more speed increase doing high level optimizing than down
level

Just my 2 cents

Fer


Neil wrote:

> I finally figured out the axes thing I've been working on, and my solution
> makes heavy use of trigonometric functions. (I've also improved my previous
> method of doing projection by creating a plane for each pair of axes and
> using the plane equation to find relative distances, thereby skipping over
> rotation and translation of the universe entirely.) Since I'm storing my
> trig values in a look-up table, I was wondering if I could somehow speed
> the whole thing up. The following is a little simplification. My question
> is: for what values (approx.) of 'someNumber' will case 2 be faster than
> case 1, if any and what other variables (if any) affect the speed, e.g. the
> array dimensions / data type of lut1,2[]?
>
> -- case 1:
> for(i = 0; i < someNumber; i++) {
>         foo = lut1[a] * lut2[b];
> }
>
> -- case 2:
> lut1a = lut1[a];
> lut2b = lut2[b];
> for(i = 0; i < someNumber; i++) {
>         foo = lut1a * lut2b;
> }
>
> Essentially, how many times do I have to dereference (That is what [] does,
> with an addition, too, doesn't it?) a value in an array before it becomes
> faster to assign it's value to another variable, if that would actually
> speed anything up at all. In a related question (for curiosity's sake),
> what would the answer be if I were using *lutPtr (without the offset of a
> [] to add the addition which I get the impression it does). Is A = *pB any
> slower than A = B, and how much so?
>
> -- Neil Edelman
> -- ICQ UIN: 705130
> -- email: mailto:dreaded.neil@phreaker.net?subject=sig
> -- home: http://members.xoom.com/dreadedneil
> -- moop: http://moop.bizland.com
> -- "My air-to-air missile overrides your air traffic control clearance"
> =================================================================
> The GameProgrammer.Com mailing list is for the open discussion
> of any topic related to the art, science, and business of
> programming games. This list is especially tolerant of beginners.
> We were all beginners once
>
> To SUBSCRIBE or UNSUBSCRIBE please visit:
> http://gameprogrammer.com/mailinglist.html

=================================================================
The GameProgrammer.Com mailing list is for the open discussion
of any topic related to the art, science, and business of
programming games. This list is especially tolerant of beginners.
We were all beginners once

To SUBSCRIBE or UNSUBSCRIBE please visit:
http://gameprogrammer.com/mailinglist.html