🚀 go-pugleaf

"Klaus J. Koch" <klaus@mdi.mr.he.schule.de> skrev i melding
news:7k26wEf91hB@klaus.mr.he.schule.de...
> Eddy Vluggen wrote:
>
> The guys at Borland who write the
> compilers know what code to generate to obtain fastest executables.

Well, almost... Look for a thread in this NG a few months ago on a small
design flaw in TList.Get method in versions 2-4 that made it take 200% longer
than needed. But generally, I agree.

> The other day I spent my time to convert 3 lines of pascal code into
> more than 50 lines of highly optimized assembler code avoiding slow
> commands as loops, jumps, calls, and pushs, and it ended up in
> taking 5 per cent _more_ cpu time than the compiler generated code.

You're good at it. I typically ended up with some 10-20%...;-)

> A hint that hasn't been given yet might be helpful:
> sometimes you often have to call a function that takes a lot of time.
> If there are only a small amount of values that are passed to that
> function then you can precalculate all the results and store them
> in an array to fetch them when needed.
> That's what a pentium does internally when executing a DIV instruction.

Before anyone can give some useful advice, get a profiler, and locate the
stuff that spends 90% of the time. Then post snippets of this code, and
someone may provide help.
--
Bjoerge Saether
Consultant / Developer
Asker, Norway
bsaether.removethis@online.no (remove the obvious)