get_global_id(0) cause much latency ?

--- Quote Start ---

You can see the latency values by hovering your mouse over the blocks in the "System Viewer" tab of the report.

--- Quote End ---

Thanks a lot!

I saw that the load in the kernel that compile optimized for NDRange (containing get_globle_id) is in burst-coalesced mode and the single pipeline one is in prefetching mode, and the latency of burst-coalesced mode is huge(159) compare to prefetching(2).

Is there anyway to achieve what I was trying to do in the first place: creating multiple kernel copies by using num_compute_unit and get_globle_id and not enduring the huge latency caused by the burst-coalesced mode?

Beside from launch them separately under different queues of cause:)