difficult profiling example

Kirsten Chevalier catamorphism at gmail.com
Mon Dec 4 10:47:32 EST 2006


On 12/2/06, Serge D. Mechveliani <mechvel at botik.ru> wrote:
> It looks correct that  searchLemmata  costs  90%.
> I could believe that  splitDisjunctScheme and addEquationsWithReduction
> take together about  8%  of what  searchLemma  costs.
>
> But who  inside the  searchLemmata  loop  takes the rest of (90 - 8)%  ?
>

I didn't look at your code all that carefully, but did you build the
GHC libraries with "-prof -auto-all"? (Not just "-prof".) If you don't
build the libraries with "-auto-all", then cost centres won't get
inserted for library functions, and if it's really a standard library
function that's taking all of that time, the profiling report won't
indicate that.

On the other hand, one thing to watch out for with GHC's cost-centre
profiling is that a lot of optimizations get effectively disabled when
you build with "-auto-all" (or insert your own cost centres
liberally), so if you're profiling optimized codes, the results can
sometimes be misleading (especially if you build libraries with
"-auto-all" too). Using ticky-ticky profiling instead could be
helpful, but I don't think it works in 6.6. (I've gotten a limited
form of it working again and I'm hoping to check it into the HEAD
soon.)

Cheers,
Kirsten

-- 
Kirsten Chevalier* chevalier at alum.wellesley.edu *Often in error, never in doubt


More information about the Glasgow-haskell-users mailing list