On the effect of branch mispredictions in the Squeak VM
Yoshiki.Ohshima at acm.org
Yoshiki.Ohshima at acm.org
Mon Jul 7 22:20:45 UTC 2003
Hello,
> Hi all (real and wannabe ;-) VM hackers,
Ok, I guess I'm included in the second category ^^;
This is interesting. In terms of optimizing interpreter, I vaguely
remember that the Java guys have done something similar in "Bytecode
Fetch Optimization for a Java Interpreter."
(http://www.trl.ibm.com/projects/jit/pub_int.htm) I can't remember the
detail at all, but at least, they deal with the superinstructions in a
portable interpreter.
I guess someone has done similar thing, but we could begin with
extracting "common basic block patterns" in the image. (Well, I
remember that there was a remnant in
pretty sure Dan and John has done this before...) Then we can see how
"static" approach works first, by either replicating the BB code or
individual bytecodes.
I'll try the code on XScale processor, which does have 2-bit branch
history.
-- Yoshiki
More information about the Squeak-dev
mailing list
|