Author: Vincent Diepeveen
Date: 07:29:13 08/30/03
Go up one level in this thread
On August 29, 2003 at 23:48:12, Matthew Hull wrote: >On August 29, 2003 at 10:04:26, Rick Bischoff wrote: > >>Hi all, >> >>As you may or may not know, IBM has released their C/C++/Fortran compiler for >>Mac OS X. Here are some results from running said compiler against GCC with >>various optimizations-- Tom's simple chess program (TSCP) is used (i.e., the >>bench command). The machine was my G4 tower (dual 867 w 1 GB Ram): >> >>GCC - "gcc *.c -o tscp_gcc0" produces >>Trial NPS Score >>1 44101 0.181 >>2 43977 0.181 >>3 45197 0.186 >> >>IBM - "xlc *.c -o tscp_ibm0": >>4 54253 0.223 >>5 54387 0.224 >>6 54424 0.224 >> >>GCC - "gcc -O *.c -o tscp_gcc1": >>7 113632 0.467 >>8 113609 0.467 >>9 113797 0.468 >> >>IBM - "xlc -O *.c -o tscp_ibm1" >>10 138769 0.571 >>11 138769 0.571 >>12 138560 0.570 >> >>GCC - "gcc -O2 *.c -o tscp_gcc2" >>13 121156 0.498 >>14 120811 0.497 >>15 121076 0.498 >> >>IBM -O2 not tested as man page says it is equivalent to -O >> >>GCC - "gcc -O3 *.c -o tscp_gcc3" >>16 119448 0.491 >>17 119734 0.492 >>18 120414 0.495 >> >>IBM - "xlc -O3 *.c -o tscp_ibm3" >>19 150939 0.621 >>20 150815 0.620 >>21 149994 0.617 >> >>IBM - "xlc -O4 *.c -o tscp_ibm4" >>22 167613 0.689 >>23 167868 0.690 >>24 168330 0.692 >> >>IBM - "xlc -O5 *.c -o tscp_ibm5" >>25 168691 0.694 >>26 168451 0.693 >>27 168536 0.693 >> >>Executable size >>gcc0 47076 >>gcc1 38948 >>gcc2 38948 >>gcc3 38948 >>ibm0 47164 >>ibm1 43068 >>ibm3 51260 >>ibm4 71468 >>ibm5 104500 >> >>(I ran diff on the gcc executables of the same size and it assures me they are >>different! That is somewhat unusual?) >> >>Anyway, it looks like the IBM compiler is a sure win over GCC if you are running >>TSCP :-) I haven't been able to get Crafty to compile cleanly wih it yet, when I >>do I will post an update. If you know of any other compiler flags for GCC or >>the IBM compiler let me know and I will test those too. > >What about the -mcpu or -march switches. Are those set automatically? and compiling using PGO also and options like: # gebruiken om profile info te verzamelen voor betere optimalisatie op k7 CFLAGS = -fprofile-arcs -O3 -march=athlon -mcpu=athlon -frename-registers -DUN IXPII -fno-gcse -foptimize-register-move -fif-conversion # te gebruiken om profile informatie te gebruiken bij optimalisatie k7 #CFLAGS = -O3 -march=athlon -mcpu=athlon -fbranch-probabilities -frename-regi sters -DUNIXPII -fno-gcse -foptimize-register-move -fif-conversion Most persons do not know a thing about compiling with GCC. doesn't take away that it is likely that the ibm compiler is going to be faster at G5 than gcc. >MH > > >> >>Rick
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.