Author: Robert Hyatt
Date: 10:07:41 10/24/03
Go up one level in this thread
On October 24, 2003 at 09:13:31, martin fierz wrote: >On October 23, 2003 at 17:14:13, Daniel Clausen wrote: > >>On October 23, 2003 at 14:25:07, Matthew McKnight wrote: >> >>>I was aiming to speed up my over-all nodes per second. Previously I had one >>>object that did make/unmake move, evaluation, and some general board >>>maintenance. I broke the object into three, one for each respective portion. >>>The new version seems to be about 3% slower. The original object was huge, so I >>>assumed that breaking it down would make it quicker, not slower, especially with >>>relevant functions grouped together. Am I incorrect to assume that smaller >>>objects, when logically organized, are faster? Or should everything be together >>>like before? >> >>If you had the stuff in one file and now have it split in multiple files (which >>lead to multiple object files) you can sometimes gain some percentages by >>reordering the object files on the linker line. > >is there any way to guess what a good ordering of these files would be? if you >have e.g. 10 files you have a bit too many ways of ordering them to just test >all of them :-) > >cheers > martin Put the most used object first. The next-most used object next. This gets code that calls other code into the same memory page, using the same TLB entry, sharing some cache lines, etc... I did this in Crafty by looking at the profile tree to see who calls who. > >>Or you make a file which somehow includes everything and compile just that. >>Whether the few percentages are worth that is up to you. >> >>HTH >> >>Sargon
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.