Author: Robert Hyatt
Date: 20:07:26 08/06/03
Go up one level in this thread
On August 06, 2003 at 11:59:04, Vincent Diepeveen wrote: >On August 05, 2003 at 10:45:41, Ricardo Gibert wrote: > >But did you try gcc 3.3 already with PGO ? > >(profile guided optimizations) > >for example at supercomputer i first compile with: > >CC = gcc3 >#CPP = g++ >CPPFLAGS = -D_CONSOLE >CFLAGS = -O2 -fprofile-arcs -DUNIXPII -DIA64 -DCONFIG_SMP >#CFLAGS = -O2 -fbranch-probabilities -DUNIXPII -DIA64 -DCONFIG_SMP >#CFLAGS = -O2 -DUNIXPII -DIA64 -DCONFIG_SMP > >then i run diep SINGLE CPU for 1 hour at openingsposition and let it search and >search. That's the _wrong_ way to profile. The openings position is _not_ typical of the entire game. I run a set of 24 positions, some opening, some middlegame, some endgame positions. I then run a couple of positions that hit heavily on tablebases, and finally I run a position using parallel search. That makes sure _nearly all_ branches are hit, and representative probabilities are computed for all of them. Just running the starting position won't hit much endgame code at all, nor any tablebase positions, etc... > >quit diep and remove *.o files. then second compile i enable: > >CFLAGS = -O2 -fbranch-probabilities -DUNIXPII -DIA64 -DCONFIG_SMP > >then recompile. > >In which case gcc 3.3 is faster than msvc and intel c++. >except at itanium. intel c++ is really a lot faster there which >may not come as a suprise to you i bet. ICC is faster for me on PIII _and_ PIV for crafty. And for other applications that I have tested. I haven't found any cases where GCC has beat it yet, on anything we do here. That includes chess, protein simulation, mechanical engineering (simulated annealing, etc.) and so forth. > >>On August 05, 2003 at 09:53:26, Robert Hyatt wrote: >> >>>On August 05, 2003 at 09:24:07, ERIQ wrote: >>> >>>>. >>> >>> >>>No. >> >>I have to disagree. I have performed a number of experiments and found msvc >>faster some of the time and gcc faster some of the time. It all depends on a >>number of factors that are not completely predictable. >> >>The experiments consisted of simulataneously dropping CD copies of each compiler >>from the roof of my home and observing which CD impacted the ground first. They >>never seemed to impact the ground at exactly the same time, but there was no >>clear favorite either.
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.