Author: Eugene Nalimov
Date: 17:11:00 11/27/99
Go up one level in this thread
On November 27, 1999 at 18:15:27, Vincent Lejeune wrote: >On November 27, 1999 at 17:04:11, Eugene Nalimov wrote: > >>On November 27, 1999 at 08:59:51, Vincent Diepeveen wrote: >> >>>On November 27, 1999 at 04:16:53, blass uri wrote: >>> >>>>Does someone know about a book about the fastest C program to do simple tasks? >>>>(for example to find if a number is a prime) >>> >>>>I think that knowing this information can help to think about ideas how to do >>>>chess programs faster. >>> >>>Let's please distinguish two things >>> - used algorithm >>> - used implementation >>> >>>You can make a very good optimized program A to generate primes, >>>but a silly implemented other program B might generate at the same >>>hardware a lot faster primes using a better algorithm. >>> >>>In general people then say that program A sucks and that a program B is >>>wonderful... ...yet from programming viewpoint this is rather hard to do. >>> >>>>I understand that +- operations are faster than */ operations but I do not know >>>>how much faster and if it is dependent on the computer. >>> >>>That depends upon architecture. >>>Multiply is in general not slower than incrementing with a register. >>>I'm using a lot of multiples in my DIEP's eval. >>> >>>At the P5 it sure is very slow. At the PRO/PII/PIII/Xeon it isn't. >>>Don't know about the K7, but i bet it's not slower there either, as >>>DIEP runs 15.5% faster on a K7 than on a PIII at the same Mhz. >>> >>>>Where can I get information like this information? >>> >>>www.intel.com >>> >>>Go to the technical references of the CPUs. >>>K7 is harder. >>> >>>>I can find it by doing a simple program but I prefer not to check everything >>>>about speed by testing because I may have many question like this question. >>>>Uri >>> >>>Rather hard to find it by doing a simple program, measuring things >>>accurately is also an art in itselve. >>> >>>What is however slowing down programs most is not the clocks needed to >>>do an integer multiplication. The real problem are branches. At the >>>Merced this should all be solved, so i wonder why the first vague reports >>>about the merced don't say it to be a lot faster than the 21264 cpu. >> >>Yes, but at IA-64 multiplication is *very* slow. >> >>Eugene > >How is it possible and why intel did that ?? I assure you that it's possible to implement a slow multiplication; more, it's much easier to do it this way. Why it was done I don't know - maybe because they don't want for programs to use it, so they later will not have to implement single-cycle (or even 2-cycles) multiplication on the 2GHz machine in the future. Eugene
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.