I’m trying to port some Matlab code to CUDA using the Matlab plug-in. I’ve been through the various tutorials I’ve found here and there and the toy apps all work fine but when I apply the same principles to my project it all goes horribly wrong. I have a mex function which takes care of the memory management (13 input and 3 output matrices) which then calls the kernel which (should) perform a Monte-Carlo sim. What happens is Matlab locks up and has to be killed. Under Windows it’s worse, the whole system locks up.
I’ve found reference elsewhere that it’s something to do with memory management but nothing concrete. Does anyone with experiance with this plug-in for have any general tips or pit falls to avoid?
Is there a Google Group like there is for cudpp maybe?
Edit
Forgot to mention:
-CUDA 2.3
-Fedora 10 64-bit
-MATLAB Plug-in 1.1