Gross inefficiency in kmail

[Josef Weidendorfer]

On Sunday 21 March 2004 16:15, David Faure wrote:
> On Sunday 21 March 2004 01:14, Josef Weidendorfer wrote:
> > Am Saturday 20 March 2004 07:01 schrieb George Staikos:
> > > sync with at most 5-10 new messages) under calltree takes 1.5 hours to
> > > complete.  It takes many seconds (over a minute or two?) to complete in
> >
> > Just a tip: Did you run calltree with option "--simulate-cache=no" ?
> > I suppose that detailed cache simulation is not needed here, but the
> > number of instructions executed and call counts are enough.
> > This should cut down simulation times in halve.
>
> Interesting!
> Does calltree support reading options from e.g. $CALLTREE_OPTS?

Currently not. But you could modify the startup script (i.e. calltree) to 
include CALLTREE_OPTS into VALGRIND_OPTS.
Neverless, newer valgrind (2.1.x) supports skin/tool options in VALGRIND_OPTS 
like this: --calltree:simulate-cache=no.

> Sounds good to me (since I don't know how to read the cache simulation
> anyway :)

Even if you know, often it's difficult to make modifications to the better ;-(
General guideline for caches: Try to improve access locality, which means
for the data layout: put data together which are used together, and don't 
interleave sparely used data with often used data.

Josef