Re: performance measurement and caches

Mary Fernandez <mff@research.att.com> writes:
> programs. Personally, I want to measure and to read about real
> elapsed-times, not simulated results.


Each has their place. Elapsed performance measurements for some
optimizations simply tell you that something is "slower" or "faster";
they don't often tell you why, unless you use performance monitoring
packages.


For example, an optimization may cause memory layout to change enough
that TLB thrashing occurs. Only knowing that something runs slower
doesn't help deduce the influence of the TLB - a TLB simulation would.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.