[Haskell-cafe] Interpreting profiling results

[Daniel Fischer]

Am Dienstag 16 Februar 2010 15:45:38 schrieb Jean-Marie Gaillourdet:
> Warning: speculation ahead, but is based on my knowledge on other
> profilers.
>
> Many profilers work statistically, they interrupt a program at more less
> random (or equal) intervals and inspect the stack, whick is of course
> quite difficult in Haskell as far as I understand it. I have interpreted
> the entries column as an indication for the amount of "profile
> interrupts" which happened when a function f was on top of the stack,
> whatever that means in Haskell.
>
> The manual of GHC 6.10.4, chapter 5 states:
> >The actual meaning of the various columns in the output is:
> >
> >entries
> >
> >    The number of times this particular point in the call graph was
> >    entered.
>
> So for me the question remains open, is "entries" a precisely counted
> value or a statistically determined one?

I have one observation that supports "precisely counted", namely, while the 
time spent in each cost centre (number of ticks) varies between profiling 
runs of the same code, the number of bytes allocated and the number of 
entries remain the same.

It's far from conclusive, though. Anybody willing to dig into the profiler 
code? 8-)

>
>
> Best regards,
> Jean

Cheers,
Daniel