Re: Need help with a very scary error
Re: Need help with a very scary error
- Subject: Re: Need help with a very scary error
- From: Jean-Daniel Dupas <email@hidden>
- Date: Sun, 1 Nov 2009 15:46:00 +0100
Le 1 nov. 2009 à 14:05, Jean-Daniel Dupas a écrit : Le 1 nov. 2009 à 13:23, Christian Demmer a écrit : Jean-Daniel Dupas wrote: I managed to reliably reproduce this issues in a complex project, but
didn't managed to reproduce it in a simple test case.
I don't know if this is relevant, but some of these call (especially
the one that fails) are perform asynchronously in a serial
dispatch_queue, and this code is in a shared library.
I am no expoert on this, but I thought a bit about this problem. My question is: is this on i386 or ppc and which optimization level -O0 versus -Os do you use. Can you try if this makes a difference?
This is i386. -Os, -O0, same result (except that test1 and test2 are both invalid with -Os, probably because gcc detect that test1 and test2 are equal and optimize the function). Is the error always with test or is it mixed with test2 and direct
printf call calculation?
This is always the first float-int conversion that failed. My feeling is that there maybe some problem with x87 fpu mode and float
-> int conversion.
That's my thought too. And after deeper investigation, it look like it append after my code spend some times in heavily optimized assembly code. I will continue to investigate to narrow down this problem. This code is maybe messing the fpu mode.
I'm progressing. I managed to identity the problem. This problem is due to some erroneous assembly in libavcodec (a part of ffmpeg) that does not correctly reset the ftag x87 register (using the emms instruction).
|
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Xcode-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden