Author: Eugene Nalimov
Date: 13:27:34 10/22/01
Go up one level in this thread
There is population count on IA-64, but no find first/last non-zero bits. And when used for Crafty purposes popcnt has latency of 4 on the Itanium, so the way it's currently implemented in Crafty is probably not worse. You cannot modify the program source when you are calculating CPU2000, nor you can use asembly routines or inline instructions. As a result, reported Crafty score measured performance of the "pure" C implementation. Athlon shines when running code compiled for other CPUs (AMD had no choice, as it's doubtful that any major vendor will compile its software for the Athlon). For Pentium IV some changes in Crafty sources are necessary, otherwise it hits too many microarchitecture problems. Vendors who submitted Itanium results to Spec used Intel's compiler. 64-bit MS Visual C produces faster Crafty for IA-64. Eugene On October 22, 2001 at 15:45:20, Dann Corbit wrote: >On October 22, 2001 at 14:45:05, Brian Richardson wrote: > >>On October 22, 2001 at 13:58:31, Dann Corbit wrote: >> >>>On October 22, 2001 at 07:20:07, Gian-Carlo Pascutto wrote: >>> >>>>On October 22, 2001 at 05:33:50, Ralf Elvsén wrote: >>>> >>>>>Has anyone seen how Crafty does on an Itanium processor? >>>>> >>>>>Ralf >>>> >>>>www.spec.org >>>> >>>>Short answer: pathetic >>>> >>>>Itanium 800: 355 >>> >>>I suspect that this measurement is largely a misnomer. >>>The Itanium chip has a popcount instruction. >> >>There is no bit-level Itanium popcount instruction. There is one for bytes >>(non-zero bytes, not bits per byte). There were some posts about this about a >>year ago. Merced performance is very poor. McKinley may be acceptable. > >What's this then: >http://www.calvin.edu/academic/rit/webBook/computer/design/hardware/cpu/itanium/population.html
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.