int nnz!?!

General CULA Sparse support and troubleshooting. Use this forum if you are having a general problem or have encountered a bug.

int nnz!?!

Postby blgolden » Sat Jul 05, 2014 10:13 am

I know no one at EMP is reading this but...

Really!?! Number of non-zero values is passed as an int!?! :shock:

Can't we get that changed to an unsigned int?

Sure, no single GPU supports problems large enough to go beyond that count... yet. But what about when using the Host routines?

UPDATE: A quick note to anyone who might be reading this. After some extensive testing, I'm pretty sure this is a big problem with cula sparse. I have equations system that I can scale easily so I am able to make runs of varying NNZ easily. I also have written my own PCG solver. As I scale problems up I go from solutions that match my solver to geting "solver has broken down" and diverging solutions from cula sparse (both float and double versions) while my solver runs fine. Having an int data type makes no sense here. Even unsigned is not good for passing more then a few hundred million NNZ from host to GPU. E.g., cudamemcpy with sizeof(double)*large int will not return something sensible. I suspect something like hat is going on in the library?
blgolden
 
Posts: 4
Joined: Wed Mar 24, 2010 8:42 am

Return to CULA Sparse Support

Who is online

Users browsing this forum: No registered users and 2 guests

cron