2013.1: Add possibility to profile GenericScanKernel based kernels

[petRUShka]

It doesn't return an Event even if I enable profiling in queue. It would be great if call method of GenericScanKernel based kernels return event if profiling is enabled in kernel.

[inducer]

It can't return a single event, because a scan usually consists of three kernel calls. (fine, coarse, postprocess) What it could do at that point is return marker events before and after, but you can do that just as well from the outside. And I'm against exposing events from the internal kernel calls, because how exactly scan is implemented is just that, implementation detail.

[petRUShka]

As I see, you can make series of events A (fine), B (coarse), C (postprocess) where B depends on A, C depends on B and return C. Markers it a solutions but no so native as returned events.

[inducer]

What's your intended use case? Out-of-order queues or profiling? For OOQ, I would agree that the scan would probably have to do what you say. For profiling, the string-of-events is mostly useless...

[petRUShka]

First of all OOQ. I want to create complex scheme of kernels with complex dependencies between them. Profiling is no more that nice addition...

[inducer]

Fixed by 6046ccd.