[Beignet] clEnqueueNDRangeKernel and kernel completion

[Zou, Nanhai]

>>Hi Nanhai,

>>I have a quick look at this patch. And have one question here, the original bug should be that it forgets to set the queue's last batch's value, so it always do nothing when call clFinish().
>>So IMHO, the easiest fix is to add code to set the queue's last_batch maybe.

>>Your fix seems to not only fix that bug, but also move the whole things from hardware independent layer to the intel driver layer. Right? Is there any specific reason to do so

Batch buffer handle and waiting code etc. is hardware specific.
Put batch buffer code in frontend looks not reasonable for me, the hardware sync logic may change between platforms.

Thanks
Zou Nanhai