I'm currently using an EVMK2HX evaluation board to do some algorithm and performance prototyping. For my prototype I'm testing with a qsort of 512MB of TeraByte sort style data. So far I've run the algorithm on my desktop PC (Dell T5600 running Ubuntu 14.04), a single ARM core (boot Linux, scp my executable, execute from command line), and a single DSP core (download and run through CCS and JTAG). Before I evaluate the results I want to make sure that I have the Keystome set up to get the optimal performance, and that I'm not set up in a configuration that's intentionally running at less than 100%.
Ignoring the details of my specific algorithm, I'm seeing performance numbers similar to this for the same algorithm:
my PC - 3.9 sec
ARM - 8.1 sec
DSP - 19.5 sec
When I load CCS, and before I run the benchmark, I'm loading the
/opt/ti/mcsdk_bios_3_00_04_18/tools/program_evm/gel/xtcievmk2x.gel GEL file, then doing a Scripts | TI66X Functions | Global_Default_Setup
Is there something more that I need to do to ensure that I'm getting the best performance out of the EVM ?
Thanks,
Brian