Nios Instruction Execution Count

It ought to be possible to write a slave for the 'tightly coupled instruction memory' (that just accesses some M9K memory) and count the number of code fetches.

Or are you really trying to benchmark the compiler! In which case analysing the generated code might be better.

However, overall performance is significantly changed by memory delays, execution stalls, and mis-predicted branches.

So any attempt to benchmark the processor should really take these into account.

Unless, of course, you are trying to deterimine the number of lost cycles (ie comparing the clock count to the instruction count).