netterfield at astro.utoronto.ca
Wed Jul 16 00:44:07 CEST 2003
I've been gprofing kst to see where things are slow....
kstvector::interpolate() is about 40% if the time (largest single subroutine
obviously) on my P4. If I take out the if statements, this drops to ~10%. A
single if check (always false or always true) pops it back up to 32%. Sounds
like branch prediction on the P4 getting it wrong too often.
-the if's are never true unless there are bugs upstream (I think...)
-is there any way that you know of to make an if check less expensive on a P4?
More information about the Kst