On 10/9/12 10:57 , Oliver Bock wrote: >> Do you have proper "critical section" protection around kernel launches > > No, and from our app's ignorant perspective it shouldn't be necessary > since it's robust enough to not produce invalid results because of such > action. However, it might be useful to add these anyway to "protect" the > potentially fragile GPU runtime/driver ecosystem from harm.
Maybe the friend of yours was referring to this part? > http://boinc.berkeley.edu/trac/wiki/AppCoprocessor#Cleanuponprematureexit Oh well, we obviously missed that story since it was added way after we launched our CUDA app. As mentioned above, we'll add critical sections around the CUDA parts of the code. The question then is how large can they be...? I don't think that a very fine-grained control on the per-kernel-call level is reasonable, depending of course on the individual app's runtime behavior. If there's time to do a proper app/CUDA shutdown I'd prefer to do that, rather than just exit. @David/Rom: I presume that an app that doesn't react on a quit/abort request gets terminated/killed. If so, what's the time interval for that? But still, while the current problem might be related in the sense that the task keeps running, the BOINC client fails to detect that or doesn't react accordingly. Best, Oliver _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
