> This means that ruby will give us the memory accesses and the number of
> cycles assuming that computational units all have single cycles latency and
> no resource constraint.
>
Yes
> If we have a code like this:
> for( some condition)
> {
> memory access
> FP Add
> FP Add
> FP Mul
> }
>
> then the number of cycles elapsed between two consecutive memory accesses
> will be three as given by ruby though it will depends a lot on the
> functional unit latencies. Is that correct? If yes, I think I would be
> better off using Opal with ruby.
Yes, assuming SIMICS_RUBY_MULTIPLIER is set as 1, the memory access will
take as long as Ruby determines whereas the FP operations take a single
cycle.
> the number of cycles which we get as the output of dump stats are simics
> cycles right? which will be twice that of ruby cycles if
> SIMICS_RUBY_MULTIPLIER if set to 2.
>
The output in the Ruby stats file are in terms of Ruby cycles.
--Mike
|