SCOUG-Programming Mailing List Archives

>> Next Message >>

Date:	Sun, 11 Jan 2004 21:50:56 PST8
From:	Peter Skye <pskye@peterskye.com >
Reply-To:	scoug-programming@scoug.com
To:	scoug-programming@scoug.com
Subject:	SCOUG-Programming: Optimization techniques (was QA equals testing, Part One:Detection)

Content Type: text/plain

Steven Levine wrote way back on January 4:
>
> Back in the 8086 and 80286 days when you did a lot of
> your original code, you really had no choice or you
> would still be waiting for results.
>
> However, over time the performance of FP hardware has
> increased dramatically to the point where often there's
> little if any performance difference between floating
> point and fixed point operations on a given system.
>
> Have you done any recent studies to ensure that your
> fixed point methods still outperform floating point
> methods on today's typical PC?

Hiya Steven,

I'm b-a-a-c-k.

No, I haven't done those studies because (as you say) the hardware has
improved dramatically and my old software optimization is no longer
necessary.

If I designed ("spec'd") the system today I would do it much
differently. The original design used a custom-written LIM (lurkers:
DOS memory management) driver front-end to cut down on the amount of
page swapping (which takes a lot of time). The pages were only 16K and
I had to split the arrays into non-obvious vectors to do the
calculations.

I was going to move towards Java but, because I'm calculation-intensive,
I was worried about "how long before I get an answer?" Java has also
improved dramatically so this is also less of a concern that it once
was.

Finally, being able to _understand_ the answer was much more of a
problem then. But I've gone from 80x25 and 640x480 monochrome graphics
to 1600x1200 color which gives much better visualization. The display
hardware is far superior now and that's another dramatic improvement.

Amusingly, current hardware can throw answers at me faster than I can
absorb them, and I take advantage of that by making my matrices larger
so the answers take longer.

A few years ago I calculated that my entire first two years of research
using almost four dozen machines could now be done in two weeks on one
machine. And that was based on hardware speed improvements only -- no
change in the software. When I wrote my programs, an IEEE 80-bit number
had to be calculated in software (I don't remember if an 80287 could
handle them); today that's not a concern.

Ah yes, your question: I don't have the slightest idea how my
fixed-point vectored matrices might compare to simpler floating point
non-vectored matrix calculations.

- Peter

=====================================================

To unsubscribe from this list, send an email message
to "steward@scoug.com". In the body of the message,
put the command "unsubscribe scoug-programming".

For problems, contact the list owner at
"rollin@scoug.com".

=====================================================

Return to [ 11 | January | 2004 ]

SCOUG, Warp Expo West, and Warpfest are trademarks of the Southern California OS/2 User Group. OS/2, Workplace Shell, and IBM are registered trademarks of International Business Machines Corporation. All other trademarks remain the property of their respective owners.