Author: Roberto Nerici
Date: 13:16:58 02/20/04
Go up one level in this thread
On February 20, 2004 at 13:42:25, Russell Reagan wrote: >On February 20, 2004 at 07:42:58, Roberto Nerici wrote: > >>This was an interesting test that you did. However, my results are a bit >>different. On a P3-733 and using MSVC6, I found the "Gerd" test the fastest by a >>fair bit, followed by "table16" and "eugene2", with "eugene" (the one you're >>using) slower still. >> >>I had to make some very minor changes to get it to compile under VC6, but >>nothing that should make a difference. However, I did then make one further >>change: I declared the scan routines as __inline and this speeded them all up >>compared to letting the compiler inline what it wanted. > >It probably varies depending upon the CPU you use. I don't trust the test >results of the bitscans that use 16-bit tables however, because in this test the >16-bit lookup table will usually get stuffed into the cache. When the chess >program is running, the entire 16-bit table will rarely be in the cache, so I >like the ones that use smaller lookup tables better for an actual chess program. I agree the test may be overly kind to the larger lookup table ones. Also Bob's comment, which you quoted in the original article, that in a real program some things will inline better is also important. Note that the "Gerd" one which was fastest for me in the test only used a 64 element array, so I'm happy with that one either way... Roberto/.
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.