the device being used is no longer there (MATLAB + GPU)

hi Folks: This bothers me: I have 2 GPU cards on my machine. I designate the 2nd one to use in my MATLBA code, but I got following error:

The device being used is no longer current.

Error in deriv_optcontMxy_cuda (line 18)
    gpuderivMxy.GridSize        = [min(1024,tmp.MaxGridSize(1)) 1]; % min(1024, maxblocks)

Error in optcontMxyHBG (line 72)
        drfn = deriv_optcontMxy_cuda(rfn,gamgdtn,xx,auxa,auxb,a,b,rfmask);

I am not sure why; Here is my code:

persistent gpuderivMxy
persistent drfrt drfit;

Nt   = size(rf,1);
Ns   = size(xx,1);
Ndim = size(xx,2);

if isempty(gpuderivMxy)
    disp 'Initializing GPU drf calculation...';
   gpuderivMxy = parallel.gpu.CUDAKernel('deriv_optcontMxy_cuda.ptx','deriv_optcontMxy_cuda.cu','deriv_optcontMxy_cuda');
    tmp = gpuDevice(2);
    gpuderivMxy.GridSize        = [min(1024,tmp.MaxGridSize(1)) 1]; % min(1024, maxblocks)
    gpuderivMxy.ThreadBlockSize = [min(128, tmp.MaxThreadsPerBlock) 1 1]; % min(128, maxthreadsperblock)
    drfrt = gpuArray(zeros(Ns, Nt)); 
    drfit = gpuArray(zeros(Ns, Nt));
end;

If I do not use tmp = gpuDevice(2), only tmp = gpuDevice(); it has no mistake like above;

Any one has any idea why this is happening? I appreciate it.

Answer 1

[Nick Tsui]

I think if you put

tmp = gpuDevice(2)

before

gpuderivMxy = parallel.gpu.CUDAKernel('deriv_optcontMxy_cuda.ptx','deriv_optcontMxy_cuda.cu','deriv_optcontMxy_cuda');

thsn you will be fine.

Thank you.