Author: Gian-Carlo Pascutto
Date: 08:10:13 09/27/01
Go up one level in this thread
On September 27, 2001 at 10:41:56, Nolan Denson wrote: >After watching a webcast on the optimization for the P4 .. I am convinced that >the code that we are presently using for our chess programs are not taking >advantage of the P4 special feature ( such as vectoring). After watching the >presentation, I compile my crafty to using the Qvec_report3 option to see where >the code of crafty was not using vectoring. Almost the whole code wasn't using >this advance feature. Of course it isnt. As far as crafty's code is concerned it doesn't care at all what system it is running on. Intels compiler is supposed to handle the vectorization itself. Why won't it? Read below... >Maybe some of you advanced programmers can expand on this. I know most write >programs to run across a wide range of processors which only make sense. >But when i see post about how the P4 is slower that the Amd line i find that >hard to believe when the P4 has so many advance features. The 'advanced features' of the P4 are not more than marketing. I was very impressed with the dual-pumped ALU's when I first read about them, but the bitter truth is that the CPU simply isn't fast enough to feed them. SSE2 and the like are all very nice, but their application is limited. They are mostly worthless for chessprograms. Certainly for non-bitboarders, and mostly for bitboarders too because many bit-twiddling functions are not available on the MMX registers. This is why vectorization of bitboard operations is so hard. Additionally, the MMX unit on the P4 is slow (IIRC). SSE/SSE2 is floating point and not usable in a chessprogram. As people have been finding out, recompiling apps for the P4 brings a perfomance increase that is negligible and doesn't help solve its performance problems at all. The 'advanced features' your CPU is supposed to have are useless for most apps, and what is left of the CPU is utter crap compared to anything else out there. No amount of 'optimization' is going to fix _that_ problem. -- GCP
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.