Atmapuri

Direct execution for CPU device.

Discussion created by Atmapuri on Oct 10, 2011

Hi!

Intel introduced:

CL_QUEUE_IMMEDIATE_EXECUTION_ENABLE_INTEL

which when used with CPU devices causes clEnqueNDKernel to bypass the threading engine and calls the specified kernel function from within the callers thread. Any chance to see that for AMD drivers as well?

This makes it possible to reduce the multithreading overhead to 0 when needed.

Thanks!
Atmapuri

 

Outcomes