Hi Todd;
Made some big changes. In order to try and perk up the performance, I
changed the successive logical "if" tests for built-in methods to be
hash-table based.
It works, but I had to make extensive changes. Basically, decaf needed
state -- I had to make a hash table after most of decaf was initialized,
but before it really started. Most of this happens in d_main.cc. I
also teased apart some of the bigger resolution routines into smaller
ones, until I figured out that doing this the next-to-be-described way
would entail REALLY extensive changes to the code, and then some
algorithmic behavioral changes... I also put most of the char*/UTF8
routines into a new utf8.cc file. I also put all the builtin methods
into one new file jvmbuiltins.cc. There are obviously-named header
files for each. There's a new hashtable template in jjos underlying
this. I also tried to remove some globals and make them instance
variables on the new decafJVM class, which puts them in the heap, which
should make our job of finding the GC roots easier.
I verified both builds, but didn't notice a real performance boost on
the VGA driver screen clearing stuff (although I wasn't smart enough to
save the old binary before I overwrote it (Doh!), so I couldn't do A/B
testing -- Thomas and Hilary, please don't make the same mistake I did).
Eventually, I think maybe we should replace this mechanism with one that
forces the loading and resolution of each class for which there's a
builtin method, and then over-writes the resolved_method data structure
with one that calls a builtin. Then we won't need a separate data
structure, and this would be even faster (probably twice as fast) as the
new scheme.
BTW, I should've incorporated the fixes you asked me to, too.
Whew.
I plan to check these in tomorrow AM US EST...
When George gets the conservative GC
Later!
-jm
--
==== John Morrison ==== [EMAIL PROTECTED] == http://www.mak.com/welcome.html
==== MaK Technologies Inc., 185 Alewife Brook Parkway, Cambridge, MA
02138
==== vox:617-876-8085 x115
==== fax:617-876-9208
_______________________________________________
Kernel maillist - [EMAIL PROTECTED]
http://jos.org/mailman/listinfo/kernel
