Ian Piumarta wrote:

> The sources for a 3.1a-4164pre1 VM are also in the same places for anyone
> interested.

I compiled the vm and it worked as expected. I ran into some problems
trying
to use self generated source files from an image and a different
internal/external module ratio. Most of it had to do with missing
include
paths in the Makefile (if FilePlugin is external I had to insert
$(gendir)/FilePlugin at various places) or references to other libraries
(eg. -lffi for the external FFIPlugin).
But it is a nice base to play with. Which I did when I thought what
might
be the cheapest way to gain some performance. I thought it might be a
good
idea to use the latest compiler which has a special configuration for my
processor. After it turned out that the vm compiled with the newer
compiler
was slower than the original one I went the other way around and used an
older compiler. So and this are the results of the amazed jury:

'54794520 bytecodes/sec; 1623491 sends/sec' "gcc272 -m486 -O3"
'41994750 bytecodes/sec; 1424725 sends/sec' "gcc-2.95 -march=i686 -O3"
'41157556 bytecodes/sec; 1225389 sends/sec' "gcc-2.95 (default)"
'37058482 bytecodes/sec; 1287986 sends/sec' "gcc-3.00"
'36467236 bytecodes/sec; 1232651 sends/sec' "gcc-3.00 -march=athlon"

It might be nothing new for the natural kernel hacker, but I was
quite surprised by this.


Alex

PS: Athlon 600 (Linux 2.4.5)