Author: Dezhi Zhao
Date: 06:37:49 09/01/99
Go up one level in this thread
On August 31, 1999 at 14:55:54, Eugene Nalimov wrote: >Please notice - I compared virtual method call not with the if statement, but >with indirect function call. And once again - performance in those cases is >comparable. And yes, you are absolutely right, for small amount of >alternatives(less than ~5 possible cases) if's are better. > >Eugene > I think ifs are better provided alternatives are less than 3 for P6 cpus. call near generates 4 micro-ops call [mem] (near) generates 7 micro-ops too many alternatives would offset the branch prediction effect. dzhao >On August 31, 1999 at 13:03:43, Dan Andersson wrote: > >>On August 31, 1999 at 08:16:31, Dan Andersson wrote: >> >>>The extra instructions could cause cashe overwrites and such, thus making it a >>>memory bound penalty. I know that vtables are a very bad deal on x86 machines, >>>much better to use switches and inlined functions. >>I meant building binary desicision trees w. if statements, and inlined >>functions. On x86 you can do aprox. four or five conditional jumps at the same >>performance cost as one vtable lookup. If it is possible to know the >>probabilities of method calls it is a win to make Huffman codes and construct an >>unbalanced tree corresponding to them. >>> C++ is only about 30% faster >>>than natively compiled OCAML, a much better prototyping and developement >>>language. >>> >>>Regards DAn
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.