John,
Thanks for the pointer. Rather than backreving to an earlier compiler, I hacked the 'sqUnixConfig.h' file to make it look like the one generated with gcc 2.95.3 (i.e., I defined DOUBLE_WORD_ORDER). Things worked...
...except, that now I'm seeing the "Comanche data dribble problem". I'll wait a few days and try this again, per your earlier post where you said:
Yes this issue is being fixed, I"m peering at a broken freebsd VM right now.... Certainly the issue of dribbles of data has been fixed, so wait a day or so
Thanks again,
Nevin
*************************************
John M McIntosh wrote:
- I quit my existing Squeak image, and replace the old 'squeak' VM
with the new one, and try to restart my image. I get a walkback. The relevant part of the stack walkback is the following:
I think this is the 3.0.3 gcc issue
See Ned Konz notes in the archives At 3:55 PM -0800 3/7/02, Ned Konz wrote:
Looking further, it may be a config issue; the sqUnixConfig.h file generated for the two differed as follows:
$ diff -c ../sqUnix* *** ../sqUnixConfig.h.2.95.3 Thu Mar 7 15:34:11 2002 --- ../sqUnixConfig.h.3.03 Thu Mar 7 15:33:25 2002
*** 111,117 ****
/* #undef WORDS_BIGENDIAN */ /* #undef DOUBLE_WORD_ALIGNMENT */ ! #define DOUBLE_WORD_ORDER 1
/* damage containment */
--- 111,117 ----
/* #undef WORDS_BIGENDIAN */ /* #undef DOUBLE_WORD_ALIGNMENT */ ! /* #undef DOUBLE_WORD_ORDER */
/* damage containment */
So I tried doing the config for GCC 3.0.3 and then edited the sqUnixConfig.h file, defining DOUBLE_WORD_ORDER.
And guess what? It ran!
So I decided to run the tinyBenchmarks:
With 2.95.3: 0 tinyBenchmarks '42553191 bytecodes/sec; 1283021 sends/sec'
With 3.0.3: '38811400 bytecodes/sec; 1253071 sends/sec'
So then I changed the -O2 in the Makefile to -O3 and tried compiling again. It got a bit faster in the sends/sec but slower in the bytecodes/sec...
With 3.0.3, -O3: '36802760 bytecodes/sec; 1285996 sends/sec' '37383177 bytecodes/sec; 1285003 sends/sec'
Anyway, thought you might be amused.
At 4:36 PM -0800 3/7/02, Ned Konz wrote:
Oh come on now; it probably just needs someone to tweak the knobs on the back of the set... (twiddle twiddle)..
How's this 3.03 result:
0 tinyBenchmarks '47442550 bytecodes/sec; 1434550 sends/sec' '47761194 bytecodes/sec; 1415033 sends/sec'
For reference, 2.95.3 was: '42553191 bytecodes/sec; 1283021 sends/sec' '42356055 bytecodes/sec; 1282033 sends/sec'
I used CFLAGS= -I/usr/X11R6/include -g -O3 -DLSB_FIRST=1 -funroll-loops -fno-gcse
Adding the -fno-gcse made quite a bit of difference (27% more bytecodes/second, 11% more sends/sec).
So now it's about 12% faster than the 2.95.3 was. Of course, I don't know what'll happen if you use "-fno-gcse" with the 2.95.3 ...