I am trying to invoke a kernel on a portion of a buffer that was computed on the device by a previous kernel. I tried to implement this using clCreateSubBuffer but this is not supported in OpenCL 1.0 and unfortunately I can't upgrade to a later release as the system is in a software-controlled, shared environment.
Has anyone come up with a way of doing this that doesn't ruin performance? The only thing that comes to mind immediately is to copy the portion to a new buffer, compute and copy the results back but I'm hoping there's a better way.
Thanks for your help.
pass as argument to kernel offset from begining and enqueue task with smaller global work size.
you must emulate what is supported on 1.1 under global offset parameter of clEnqueueNDRange().