Different kernels of same algorithm give different throughputs

Hi HRZ,

I've checked the clock frequency as you've mentioned and it seems to be like: 314MHz, 311MHz, 330MHz, 302MHz, 315MHz, 315MHz, 294MHz, 324MHz. As it's clear, there seems to be no correlation between the frequency and the throughput.

From what you've mentioned regarding scheduling threads into the pipeline. I have some simple question: I have understood initiation interval, which is the way for the design to fire iteration after iteration after II clock cycles, which helps improving performance of the loop execution. Now you have mentioned about scheduling threads into loop pipeline. What exactly does that mean, and how it's being correlated with initiation interval? How can I understand how many threads can be scheduled at most into the pipeline? Could you please elaborate on this a little bit? I'm trying to figure out whether it can demonstrate the performance gap I see or not.

Thanks