correct kernel lauch

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear cuda-developers,

how to launch a kernel correctly, when I have to pass a deviceID and streamID and the size of shared manory I would like to allacoate?

I know I can call a as follow:

MyKernel<<< numBlocks, threadsPerBlock, deviceID, streamID>>>(args);

and

MyKenrel<<< numBlocks, threadsPerBlock, sharedMemorySize, streamID >>> ( args );

But how to do both in one?

thank you in advance.

cheers greg

2

Oh I see,

I can use cudaSetDevice() instead of setting deviceID in the kernel call, right?

regards greg

3

Yes …you are right
Once you use cudaSetDevice(), all subsequent cuda calls are particular to the current device.
Make sure you use cudaResetDevice() once you are finished with a particular device.