Author: J. Wesley Cleveland
Date: 18:15:37 05/22/02
Go up one level in this thread
On May 22, 2002 at 18:43:09, Dan Andersson wrote: >If the function is fast to compute, faster than a memory access. And the LUT >can't be guaranteed to be in the cache. Then it will be a bad idea. YMMV. > If the code is executed enough to matter, the LUT is almost surely in the cache. You need to compute the probability of it being in each level of cache or memory times the cost (in cycles) for each to have an estimate of the speed of the routine. Then you will find that adding some code or a variable can dramatically change the real speed of your routine or it is very different on a different x86 processor. X86 optimization is lots of "fun", and is to a large extent avoiding cache misses.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.