Author: Dieter Buerssner
Date: 15:42:17 11/22/05
Go up one level in this thread
On November 22, 2005 at 03:23:46, Gerd Isenberg wrote: >Wow, gcc - that's it. Shift is relative expensive on P4, something like four >cycles (shift alu of the mmx-unit is used with "long" data pathes). Shift is >cheap on amd cpus. Hi Gerd, thanks for all the info and explanations, also in the other posts. In this context I wonder, how shift compares to cmovx? Anyway, I think you have guessed part of my opinion (besides the totally technical discussion) - I would not care (anymore) to try to look whether if/else or ? : produce better code. The other "tricks" you mentioned I might consider differently. Eugene's message was also interesting in this context. No doubt, we can save some cycles, by microoptimizing for some specific environment. And it can be much fun, too. I don't doubt either, that it can be worthwhile (in the sense of productive) for chess engines (many examples you have shown for typical bitbase routines). But not in such "simple" cases, that started this discussion. The next compiler/hardware combination may behave different again. Here, I prefer to just trust the compiler. Of course, when the discussed code is a clear bottleneck (which probably is not the case), more effort is justified. Cheers, Dieter
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.