Author: Bo Persson
Date: 07:20:39 11/01/98
Go up one level in this thread
On October 31, 1998 at 20:57:09, Robert Hyatt wrote: >On October 31, 1998 at 19:45:28, Roberto Waldteufel wrote: > >>>On October 31, 1998 at 07:27:43, Bo Persson wrote: >>>>I thought these instructions were microcoded and extremely slow on anything >>>>newer than a 386! When testing on a Pentium, a table lookup was actually much >>>>faster for me. >>>> >>>> >>>>Bo Persson >>>>bop@malmo.mail.telia.com >>> >>> >>>ditto here. But I only tested this on a P5 (non-mmx) when I started doing this >>>stuff. the instructions were grossly slow. Someone even tried some hand-coded >>>stuff to help and it was still slower than what I do at present (table lookup). >> [snip] >>If bsf and bsr really are microcoded, would I not get much lower counts? Or >>maybe I ought to be getting even higher than I am? I don't know exactly how fast >>is fast in this context. >> >>Best wishes, >>Roberto > > >note the architecture I said I tested on: the original P5 with no mmx, no >fancy tricks or anything. The P6/PII has a totally different cpu core, and >the bsf/bsr might be much faster in them. I have not tested this since the >P6/PII came out... but I will now... I don't have a PII, but an AMD K6/233. So I tested the bsf/bsr again today, and found that their performance has improved since last years Pentium machine, but they are still no improvement. Note the difference though, on a Pentium 133 the bsr/bsf perform much worse that a table lookup. On a K6/233 they just make "no improvement". So now I'm really looking forward to my new PII/400, coming soon. :-) Bo Persson bop@malmo.mail.telia.com
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.