can single work items kernels run in parallel on same device

Can single work item kernels run in parallel on the same device (i.e. on the same board).

I've been trying to get a very simple example of task parallelism working but have not been able to

to get more than one kernel to run at the same time on the same board.

The kernel computes part of summing equations - let's say it sums numbers from "start" to "end".

In a .cl file there are multiple identical kernels that do this - let's say there at 12 of them.

Single work items kernels have been used to insure that the equation can be pipelined.

The host code creates multiple kernels and multiple contexts in an effort to run more than one in parallel.

After trying many, many things, I've yet to get them to run in parallel. Initially I used just the time profile

to see how much time they take to run. Each kernel takes about the same time (e.g. 25 ms). If 12 kernels

are started, the time is 300 ms.

There are four identical boards in the system. If 12 kernels are used and three are used on each of the four

boards then each one takes 25 ms but each board can run them in parallel so the total time is only 75 ms.

What else is needed to get the kernels to run in parallel on the same board. I've been able to turn on

profiling and can see that each one is started - one after the other.

Everything seems to work (i.e. the correct answer is produced) but the kernels don't run at the same time on

a single board.

Do I need to use NDR range kernels?

Any suggestion would be greatly appreciated! (this should be so hard?!?)