Comments for Falasol.net - Free Audio Engineering Software

Comment on 2^x optimization for double type by koshik

koshik — Sat, 15 Oct 2011 14:01:38 +0000

That is a really good suggestion!
I like the first option more since it doesn’t introduce any performance overhead.
Second could be slightly more accurate, but would require 2x code execution.

The only drawback I see an ugly fractional part for some ‘simple’ outputs, e.g:
2^-1 => 0,500000000161386
This is only a question of human aesthetics sense, since precision comparison speaks for itself.
Values [-10...10] with step 0.001:

Old version:
Max absolute error: 6,60890577819373E-07
Encountered at: 9,99999999999979
Max relative error: 6,45538647088066E-10
Encountered at: -1,87500000000001

Updated version:
Max absolute error: 3,3037224511645E-07
Encountered at: 9,99999999999979
Max relative error: 3,22766880154427E-10
Encountered at: -1,87500000000001

I’m updating the post.
Thank you!

Comment on 2^x optimization for double type by Slime

Slime — Fri, 14 Oct 2011 08:54:15 +0000

You’re always rounding down to the nearest entry in the table. Maybe I’m missing something, but it seems like you could cut your error in half by using ((double)n + 0.5) in the Init() function.

(Or perhaps it would be more correct to use the average of pow(2,n/precision) and pow(2,(n+1)/precision).)

Comment on SoME 2010 Sep: Harmonica obsession by koshik

koshik — Tue, 08 Mar 2011 23:16:55 +0000

You are welcome fellow soul!
Good to know there are other folks out there trying to explore the uncharted lands on the edge of physics and the human mind :)

Feel free to ping me anytime here or via koshik@falasol.net
(also if you’d like to exchange links with your site, or use this one for posting any stories)

Comment on SoME 2010 Sep: Harmonica obsession by Alex

Alex — Mon, 07 Mar 2011 09:38:46 +0000

This is really cool. Actually was searching just for someone else experimenting with timbre.

thanks for the software

Comment on Win7 Math recognition by koshik

koshik — Tue, 08 Feb 2011 17:51:36 +0000

Yeah, we probably could trick it into being rendered inside some winform host, but then it would look like a pop-up inside a window = terrible user experience. Still there is the ‘Insert’ button which need to be pressed manually. Since there is no shortcut key we would need to use some UI automation to simulate the click.
These are way too many drawbacks and workarounds. We decided that this effort will be better invested in finding an alternative recognition method.

Comment on Win7 Math recognition by koshik

koshik — Tue, 08 Feb 2011 17:41:16 +0000

Thanks :)
I wish this was the only problem with that form…

Comment on Win7 Math recognition by kamil

kamil — Mon, 07 Feb 2011 22:21:29 +0000

Try to host it in a separate process running winform app and then use winapi to take ownership of the window/handle and render it within your app.

Comment on Win7 Math recognition by Victor Hurdugaci

Victor Hurdugaci — Mon, 07 Feb 2011 20:26:58 +0000

Cool stuff!
Only one UI comment: the Cancel and Insert button don’t seem to have the ‘standard’ Windows button size (see Cancel/OK/Apply)

Comment on 2^x optimization for double type by koshik

koshik — Sun, 30 Jan 2011 15:48:53 +0000

That is based on the same idea :)
The guy thought even about the fractional buffer breakdown (in supplement part). I was surprised no one mentioned that in other articles when struggling with over 1MB buffers.

Regarding the efficiency – there’s nothing like a thorough, reproducible benchmark…
“And, on one machine, it was over 9 times faster than standard pow.”

Hard to discuss with that. We’re also dealing with C++ vs. C# code. Can be different system function implementations, code optimizations, many various things.

Comment on 2^x optimization for double type by Dan

Dan — Fri, 28 Jan 2011 19:42:13 +0000

Have you looked at this one?

http://www.hxa.name/articles/content/fast-pow-adjustable_hxa7241_2007.html

The claim was 9x speed improvement over the CRT version of pow, which sounds pretty nice.