[Halld-offline] observations after updating to the latest trunk version and hangup problem

[David Lawrence]

Hi Kei,

     How much older is the JANA version you were using? The current 
trunk should work with JANA 0.6.2 and 0.6.3.

     After some local discussion following Matt's e-mail earlier today, 
we agreed that the Makefile should be changed to disable SIMD by default 
and only turn it on when explicitly done so by the user. (e.g. user will 
have to type "make USE_SIMD=yes"). Simon was going to put that change in 
sometime today or tomorrow. We plan to discuss it at the offline meeting 
next week.

     The crashing problem is still under study. The current workaround 
treats it symptomatically and does not address the underlying cause so 
we've been reticent to implement it in the make system. As a side note, 
one should be able to implement it without editing a Makefile by simply 
invoking "make OPTIMIZE=0" after touching the 
DTrackCandidate_factory_CDC.cc file.

     Admittedly, the crashing problem has not been worked on for the 
last week due to focus on other high priority work. It is still listed 
as a problem of "major" severity in Mantis 
(https://halldnew.jlab.org/mantisbt/view.php?id=38).

Regards,
-David

On 3/2/11 4:02 PM, Kei Moriya wrote:
> Dear all,
>
> After three days of struggling, I have finally
> been able to install the current trunk code and
> run my analysis programs. Most of my trouble seems
> to have stemmed from
> 1. an older jana version being used,
> 2. omitting the DISABLE_SIMD=1 flag for make
> For 2., it seems setting this in my .tcshrc does
> not have an effect, and I need to type
>   >  make DISABLE_SIMD=1
> on the command line.
>
> One more thing I noticed is that the "workaround"
> for the hangup/infinite loop problem still needs to be
> implemented. I tried running my analysis program
> without the workaround, and it silently stopped
> processing after ~2k events. After implementing
> the workaround, the code successfully ran on 10k
> events. Although I do not expect any, I will run
> several million events on the clusters tonight to see
> if there are further problems.
>
> Just to be sure, the "workaround" is setting the
> optimization flag to 0 in Makefile.common and then
> compiling DTrackCandidate_factory_CDC.cc
> After that, the flag is returned to 2 to compile
> other programs.
>
> 	Kei
>
> _______________________________________________
> Halld-offline mailing list
> Halld-offline at jlab.org
> https://mailman.jlab.org/mailman/listinfo/halld-offline