Kernel is not being invoked

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

Am developing a kernel that executes Cholesky factorization on the GPU using cuda…

After tracing the execution of my code, it turned out that the kernel is not being invoked at all… Placing a “printf” in the kernel and executing in the emulation mode I get nothing on the screen… Moreover, the matrix passed to the kernel remained unchanged…

I made sure that the execution configuration is correct…
Any idea about what is going on???

Thanks in advance…

2

I added some statements to check the error that is generated when invoking the kernel and I got:

[font=“Arial Black”]CUDA version is insufficient for CUDART version.[/font]

I searched previous posts and they say that the toolkit and the sdk versions are not compatible…

Any other ideas?

3

No other ideas, because that is the problem. It means the driver you are using is too old for the toolkit you have installed. What Toolkit and driver version are you using?

4

I am running in emulation mode, using cuda toolkit and sdk both version 3.0…

I was able to compile and execute cuda programs in the emulation mode successfully.

Now am doing cholesky factorization and i downloaded the CLAPACK library to use its subroutines for performance comparison purposes and to make sure that my kernel is producing correct results…

I placed the CLAPACK folders that contain its libraries and files under the C/common folder and modified the common.mk file (which i refer to in my MakeFile) to include and link to them…

Do you think that this caused the problem, taking in consideration that I was able to execute cuda programs normally before doing that?

thanks

5

I have got the same situation once, at that time, I forgot that I use the total thread I declared was bigger than it was allowed (512 for X, Y, Z block thread, actually, 512 for X,Y and 64 for Z) so I changed it. It worked fine. I dont’ know if you get the same my mistake.

6

Thanks for your replies…

The problem was not with my execution configuration, actually as I reinstalled the sdk all over again and downloaded CLAPACK files in another folder out of the NVIDIA_SDK folder everything went ok… It seems that downloading these files in the NVIDIA_SDK folder caused the problem. I wonder why!!!

Thanks again …