Agreed,

The simplest method to calculate throughput is to toggle an IO in and measure on a scope / analiser. This is the only way you can be realy sure how long your code takes (not forgetting that there are a few cycles taken up toggling the IO pin).

However you can calculate the throughput by hand (I am yet to find a reliable simulator that can also do this) counting fetch execute cycles for each line of assembler..... but at the moment that isn't automated :-(

With referance to compiler v hand optimised. My experiance suggests that compilers today make a damn good job of it and only a real good assembly programmer spending a lot of time can shave a few more cycles off the execution time, AND keep the code the same size or smaller. Of cause if you want to unrole ALL your itteration....

Andy