Author: Koundinya Veluri
Date: 22:11:40 01/15/02
Go up one level in this thread
On January 15, 2002 at 08:41:55, Severi Salminen wrote: >>On my computer, bsf is faster if the 1-bit is less significant and bsr is faster >>if the 1-bit is more significant, so the FirstBit() and LastBit() versions with >>the jumps are running faster on average. Actually I was wrong here. I'm not sure about bsf/bsr on my processer (which is a P4 1.7 by the way). I think it takes the same amount of time regardless of where the bit is. >>How does this test on other processors? >>I'm curious... > >You should check from processor specs, if bsr/bsf takes constant amount of >cycles or not. If not, then jumping is very wise thing to do. Otherwise the >opposite might be true. Then second thing is that adding and removing >instructions might cause misalignements and code will run faster/slower even if >it _should_ be slower/faster. So you should really check how many cycles bsr/bsf >takes on your processor (which is?) Ok I understand now. I believe bsf/bsr are taking the same amount of time (according to testing; or at least the difference is unnoticable) so jumping is then not a wise thing to do here. As Miguel has explained, my test is also not good because it makes the branches very predictable. When I use random numbers, the versions with no branches are the fastest by far. Regards, Koundinya > >Severi
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.