[Protux-devel] Re: Peak revisited

[Luciano Giordana]

On Tuesday 17 June 2003 05:55 pm, you wrote:
> Hi Luciano,
>
> After a busy week at my parents home (had some kind of concert, so I was
> hard studying my pieces of music ;) ) I looked again at the peak file.
>
> I have a few questions. Is it necesary to store peak info in a peak file?
> With large audio files, it consumes much memory and it isn't really faster.
> The only benefit I see is that there's no hard disk IO when recording
> music.

Yes. See SPECIFICATIONS #5
If the peaks arent prebufferes, fast drawing would be slowed down, since HD 
reads
are much slower than RAM read

>
> Also, zoomlvl 64 is stated twice??

It shouldnt, I will check it


> The first bottleneck I encountered was the "blockPos % zoomStep[hzoom]==0"
> statement. This consumes a lot of CPU power.
> The other bottelneck is seeking through a file. With this new peak file,
> this is done only once, with a stepwidht of 16, reducing the time of
> seeking16 times :). This gives also a good waveform as far as I can see. It
> can also be 8, but this consumes much more CPU power.

You have sent another email talking about a modified Peak class. I answered but
that time we had that problem of denyed IP relay. 

I looked at that Peak.cc and found cool the optimization, but it is not the 
ideal. 
The best way is to find the maximum common divisor for zoom levels and scan the 
audio only once,
jumping all steps smaller than the MCD. This way , we find all peak images at 
once. 
If it is what you have done now, thanks a lot. Please send me so I can patch 
the class

>
> I hope you can do something with it. At least for testing purpose :). It
> has now a speedup of at least 13 times. This is even more for smaller
> files.

Surely I will,  thanks again


>
> Greetz,
>
> Remon

-- 
Best Regards
--
Luciano Giordana - Musician - Certified Java/GNU C++ Developer - Free Software 
Evangelist
Project Mustux - http://www.freesoftware.fsf.org/mustux
-- Once Palladium is up and running , I will become a hacker --