profiling

1997-06-15 Thread Marc van Dongen

Hi all,

I have noticed that profiling in ghc-2.03 in combination with
-O or -O2 is not possible because of some problems at the
linking stage. Also it turns out that profiling combined with
tracing results in programs which don't output any trace at all.

Are the problems related to profiling an optimization resolved
in ghc-2.04? I have programs which havilly depend of
optimization. Without it, I have to say nasty things like
 +RTS -H100M -K1M -hC -RTS. Whereas normally a heap of 16M
suffices.

I have also noticed (no bug) that programs with bigger heaps
need more stack. Is this due to garbage collection?

Regards,


Marc



-fshow-specialisations

1997-06-15 Thread Marc van Dongen

This does not work in ghc-2.03.
  ghc-2.03: unrecognised option: -fshow-specialisations

Hope this helps.


Regards,


Marc van Dongen



Re: profiling

1997-06-15 Thread Sigbjorn Finne


[redirected to g-h-u, as other GHC users may have
 a story or two to tell re: 2.04  profiling ]

Marc van Dongen writes:
 Hi all,
 
 I have noticed that profiling in ghc-2.03 in combination with
 -O or -O2 is not possible because of some problems at the
 linking stage. Also it turns out that profiling combined with
 tracing results in programs which don't output any trace at all.
 
 Are the problems related to profiling an optimization resolved
 in ghc-2.04? I have programs which havilly depend of
 optimization. Without it, I have to say nasty things like
  +RTS -H100M -K1M -hC -RTS. Whereas normally a heap of 16M
 suffices.
 

Hi,

the problems you ran into with 2.03 have been resolved, although not
much testing of these bits was done before 2.04 went out the door.

Re: tracing when profiling, this is unknown to me - do you have an
example program that reproduces the behaviour?

--Sigbjorn