Zebra 2.2 VST vs AU CPU spikes

[pgmvdm]

So this is what loginwindow does...

http://www.peachpit.com/guides/content. ... &seqNum=59 reads:

Once the user session is running, loginwindow monitors the session and user applications in the following ways:

• ...

• Writes any standard error (stderr) output to /Library/Logs/Console/user/console.log (where user is the short name), which is then used as input by the Console application

By default stdlog goes to stderr which the logingwindow process writes to file. I do have some huge log files living in /Library/Logs/Console/501/ full of "DelayNaN4: LimitCoeff".

Pieter.

[Urs]

Thanks! Highly interesting!

Normally none of these messages should turn up but d'oh... d'oh...

Thanks again, I'll check it out... I'm on a mission now in respect of bug elimination, and today I had big success in some weird cases.

Hope to have something for you soon!

Urs

[SaganTech]

Very intesresting. I did some analysis using Apple's excellent shark utility and it looks like perhaps this is due to logging, which does file writes. Apparent the routine apdelay::msRatio does a crap load of printfs.

Jeremy

[Urs]

Very intesresting. I did some analysis using Apple's excellent shark utility and it looks like perhaps this is due to logging, which does file writes. Apparent the routine apdelay::msRatio does a crap load of printfs.

Jeremy

Ooooohh!!! Jeremy, you're a life saver!

#define OutputDebugString //

#define OutputDebugString printf

I guess I can just recompile

(I'm still startled about possible NaNs in the delay though)

Urs

P.S.: Where do printfs go in Leopard? I can't just open Console.app?!?

[jeremydb]

In Leopard, printfs typically go to /var/log/system.log (which is what you see in Console.app)

[mkelly]

Thanks for the effort here guys. While I'm not seeing the problem anymore, it's always good to see bugs being tracked down!

Cheers,

Maurice

[SaganTech]

...I'm not seeing the problem anymore, ...

Well I am still seeing it.

Urs, if you have a newer version of the VST I would be glad to test this to verify it fixed.

Thanks,
Jeremy

[Urs]

Thing is, I'm still not sure under which circumstances a NaN would still happen, and the delay algorithm is naturally reused in all sorts of modules. Hence it's easily possible that the preset, the number of instances and the usage of the allocated memory prior to the instance plays a role, i.e. if the NaN is a result of memory supposedly being zeroed when in fact it's maybe random values. I think the last possible NaN I fixed was in one of the filter modules where the first sample could have a random value. Hmmm...

Nevertheless, the same algorithm that cures denormals also cures NaNs, hence it's no big deal. I've thrown out the printf and will compile something for testing tomorrow and hopefully post it here after a short (expected) nod from beta team.

Urs

[Urs]

Yo, please check out this new beta version:

http://www.kvraudio.com/forum/viewtopic.php?t=207873

This should remove any cpu spikes related to logging...

Cheers,

Urs