On Sat, 2005-01-01 at 11:31 -0500, Daniel Veillard wrote: > We should double-check where time is actually spent. Using (k)cachegrind > is very useful to make such an analysis. I'd recommend starting with sysprof instead - kcachegrind only counts number of instructions, not how long they take, which can often be very misleading about the big picture. If sysprof says that most of the time is in a CPU-bound function, that's when I like to break out kcachegrind to optimize that function. I mention this because kcachegrind sent me down a few wild goose chases until I got out sysprof also ... Havoc
