Re: profiling experience

On Wed, Dec 06, 2006 at 03:21:23PM +0000, Kirsten Chevalier wrote:

On 12/6/06, Serge D. Mechveliani wrote:

...

What may consitute this strange CAF cost of 96% ?

Kirsten Chevalier wrote

...

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.

I made ghc-6.6 from the official source in a standard way:

./configure ...; make; make install

Does this presume that this also generates the .p.o GHC library versions for -prof -auto-all ?

No; you must have built the profiling libraries (i.e., building the libraries with -prof), otherwise you wouldn't have been able to compile your code for profiling. But, building the profiling libraries in the standard way doesn't add "-auto-all" to the compile flags. So if you want to build the libraries with "-auto-all", do: make EXTRA_HC_OPTS=-auto-all

Thank you. I am trying this. One could ask, why the GHC installation does not apply -auto-all by default?

but be careful! As I said in my previous message, adding cost centres disables some optimizations, so the profiling results you get from this may not be accurate with respect to what would happen if you ran your code after building it with -O and no profiling.

In my example, -O makes the run 2 times faster, and the profiling time per-cents are shown the same as for -Onot. I suspect that this particular invilisible time eater will present independently on -O flag and on the cost centers. ----------------- Serge Mechveliani mechvel@botik.ru