Hi ,
I am a graduate student at Purdue University. I am doing performance testing on EVMK2H -OpenCL runtime and I would like to know how to do the following.
1> Limit the number of DSP cores to which WorkGroups are offloaded to less than 8 using a configurable parameter from within OpenCL runtime / from Linux or may be using a system call of Sysbios.
2> View L2 Usage of DSP when OpenCL offloads WorkItems to DSP cores.
3> View Thread ID of Sysbios corresponding to WorkItem Thread ID of OPenCL. Is the Thread ID same or different for WorkItems executing within a workgroup ?
4> Is there a way to view DSP registers and other parameters using CCS JTAG while the OpenCL Host offloads WorkItems to DSP Cores to gather Scheduling timeline per core ? ps - I have fully licsensed CCS version 6.0 .
These are very critical questions and it would be helpful to many in the OpenCL on ARM-DSP developers community for improving their code.
Thanks to get back to me at the earliest.
Ravi Gupta