Author: Matt Taylor
Date: 20:28:27 12/24/02
http://65.186.75.165/~mtaylor/bitscan/bitscan.html The posted source is outdated, but as my school conveniently decided to turn off all the power without warning me, I can't update the source. (My source highlighter is on the school servers.) I think the older tests were not completely valid. While you can bitscan in a loop, I was caching data in registers, and that is not very likely to happen in a general loop. I moved the loop constructs outside of the timed code, and I also eliminated the caching that I was doing in registers. I think these results will be a little more fair. As a natural consequence of my changes, all timings will be inflated (biased by timing code latency), but they will be biased equally. It is not worthwhile to compare timings between processors, but it is valid to compare timings for different pattersn on the same processor. Also, I'm quite sure that the MMX routines could be written better. If someone wants to look over them and make some suggestions, I would invite them to do so. One last thing... merry Christmas! -Matt
This page took 0.01 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.