atg@leenooks ~ $ squeak -version 3.7b-5 #1 Wed Oct 20 07:10:59 EST 2004 gcc 3.3.4 Squeak3.7 of '4 September 2004' [latest update: #5989] Linux leenooks 2.4.27 #2 SMP Wed Aug 18 05:52:55 EST 2004 i686 AMD Athlon(tm) Processor AuthenticAMD GNU/Linux default plugin location: /usr/local/lib/squeak/3.7b-5/*.so atg@leenooks ~ $
---------------------------------------------- 1 tinyBenchmarks. '122 137 404 bytecodes/sec; 4 133 412 sends/sec' // WORKHORSE ##################
atg@leenooks ~ $ source/Squeak-3.7b-5/bld/squeak -version 3.7b-5 #1 Wed Jun 22 21:06:31 EST 2005 gcc 3.4.4 Squeak3.7beta of '1 April 2004' [latest update: #5868]
[[[ note, I can't generate a working interp.c with my 3.8 image, using the default sources. ]]]]
Linux leenooks 2.6.12 #1 SMP Sat Jun 18 16:46:08 EST 2005 i686 AMD Athlon(tm) Processor AuthenticAMD GNU/Linux default plugin location: /usr/local/lib/squeak/3.7b-5/*.so atg@leenooks ~ $
----------------------------------------- ' 69 189 189 bytecodes/sec; 3 338 403 sends/sec' // -march=athlon-mp -O3 ' 68 412 613 bytecodes/sec; 3 356 407 sends/sec' // DEFAULT 3.7 beta #####################
Comment:
Gcc 3.1.0 had problems with its optimizer such that it inserted worthless register loads into the instruction stream -- maybe that bug is back.....
With gcc 3.1, the performance saw a substantial improvment when it was compiled with -O0 -- that was not the case this time...
With -O1, however, there was a substantial improvment, though still much worse than my workhorse VM...
' 72 072 072 bytecodes/sec; 3 465 884 sends/sec. //-march=athlon-mp -O1 ' 49 306 625 bytecodes/sec; 1 690 230 sends/sec' // -march=athlon-mp -O0
File size comparison: ###################### atg@leenooks ~ $ ls -l source/Squeak-3.7b-5/bld/squeak -rwxr-xr-x 1 atg users 936148 Jun 22 21:06 source/Squeak-3.7b-5/bld/squeak atg@leenooks ~ $ ls -l /usr/local/lib/squeak/3.7b-5/squeak -rwxr-xr-x 1 root root 866604 Oct 20 2004 /usr/local/lib/squeak/3.7b-5/squeak atg@leenooks ~ $