PGI User Forum
 SearchSearch   MemberlistMemberlist     RegisterRegister   ProfileProfile    Log inLog in 

CUDA-x86.

MPICH linking failing
Goto page Previous  1, 2, 3
 
Post new topic   Reply to topic    PGI User Forum Forum Index -> Programming and Compiling
View previous topic :: View next topic  
Author Message
mkcolg



Joined: 30 Jun 2004
Posts: 6134
Location: The Portland Group Inc.

PostPosted: Mon Oct 21, 2013 9:00 am    Post subject: Reply with quote

Hi R,

I should have looked at the symbol names a bit closer. CUDA X86 only supports the CUDA runtime API not the CUDA Driver API. Hence, the "cu.." routines are not supported.

For the "dl.." symbols, try adding the "-ldl" library.

The rest of the Runtime API routines, i.e. "cuda", are all newer CUDA 5.0 calls. CUDA X86 only fully supports CUDA 3.2 routines. Currently, there are no plans to update CUDA x86 to newer versions of CUDA, though that may change given that we are now part of NVIDIA.

- Mat
Back to top
View user's profile
rvasquez



Joined: 24 Sep 2013
Posts: 7

PostPosted: Wed Oct 23, 2013 2:29 am    Post subject: Reply with quote

Hello Mat

Thanks for your help. After writing several conditions for compilation we're able to run the project, the only pending issue is that we can not use va_start due to the undefined __builtin_va_start error. I've added the -nobuild flag (or something like this) to compile but it's still throwing the error.

Another question: The performance of our code, which is a simple finite difference algorithm is really bad. Which optimizations regarding thread block size and streams do you recommend?

Thanks in advance.
Back to top
View user's profile
mkcolg



Joined: 30 Jun 2004
Posts: 6134
Location: The Portland Group Inc.

PostPosted: Fri Oct 25, 2013 8:54 am    Post subject: Reply with quote

For the va_start error I'm not sure why it's occurring but what I'd start with is to remove all the "-I" paths from your compile. My best guess is that you're picking up the wrong stdarg.h file which is defining va_start to be a bulitin type.

Quote:
performance of our code, which is a simple finite difference algorithm is really bad.
Compared to what? If you're comparing to a GPU, then this would be expected. CUDA x86 runs in emulation on the host using OpenMP style threading. It's made for portability not performance.

- Mat
Back to top
View user's profile
Display posts from previous:   
Post new topic   Reply to topic    PGI User Forum Forum Index -> Programming and Compiling All times are GMT - 7 Hours
Goto page Previous  1, 2, 3
Page 3 of 3

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © phpBB Group