Author: Gerd Isenberg
Date: 15:07:26 12/03/02
Go up one level in this thread
On December 03, 2002 at 16:14:57, Jeremiah Penery wrote: >On December 03, 2002 at 15:50:15, Russell Reagan wrote: > >>Is Athlon's bsf instruction significantly slower than that on a P3? I tried this >>code against a bsf version on my P3 733 and bsf kills it, big time. I tried all >>of these without the feature of resetting the bit. I'm still amazed by the bsf >>time though. >> >>bsf: 4.265 seconds >>LSB_64: 42.968 seconds >>LastOne(): 39.906 seconds >>FirstOne(): 55.125 seconds >> >>These were posted by Eugene Nalimov. I found them in the archives. > >BSF/BSR on a P3 are like 1-2 cycle operations. On Athlon, they're more than 10 >cycles, plus they block other execution resources. I wonder how the x87, 3DNow! or SSE integer to floating point conversion performs in micro-ops. Isn't it necessesary to have a kind of popcount (eg. counting leading ones or zeros in hardware) as micro-opcode instruction to perform this conversion by determing the biased exponent? Hmm, may be Athlons bsf/bsr-implemention occured on a monday ;-) Which hopefully is not true for Hammer. Gerd
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.