Author: Robert Hyatt
Date: 18:13:21 02/18/03
Go up one level in this thread
On February 18, 2003 at 20:43:13, Sean Mintz wrote:
>Using the latest MingW32 available (gcc 3.2, for windows) I compiled tscp 1.73.
>A long time ago I had compiled 1.73 using intel c 5 with all optimizations and
>profiling and made the fastest executable for amd processors there was (you can
>even check the tscp benchmark site
>http://home.attbi.com/~tckerrigan/bench.html).
>
>Here's the results:
>
>Compiler MIPS
>Intel C[1] 1452.4
>GCC[2] 1598.1
>
>1 - (Profile directed) icl *.c -O2 -Qipo -QxiM -G6 -Qunroll -Qprof_use
>-Qprof_dir c:\profiles
You might try this: It is the fastest I have found for Intel and Crafty:
CFLAGS='$(CFLAGS) -D_REENTRANT -O2 -march=pentiumiii \
-mcpu=pentiumpro -prof_gen -prof_dir ./profdir \
-fno-alias -tpp6' \
The above will compile for a pentium 3 box. Pentium4 needs a change. But
I would avoid the unroll options and so forth and let the compiler do what
it wants...
replace prof_gen with prof_use after running a _reasonble_ set of test
cases to get all the branch probabilities...
>
>2 - -O3 -march=athlon-mp -mcpu=athlon-mp -mfpmath=sse -mmmx -fomit-frame-pointer
>-funroll-loops -finline-functions -fprefetch-loop-arrays
>
>Those same options get 215knps with tscp 1.8. Intel gets 193knps...
>
>I'm on a dual athlon mp 1.2 ghz on a tyan tiger s2460 with 512 mb memory.
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.