develooper Front page | perl.perl6.internals | Postings from August 2001

Re: Opcode Dispatch

Thread Previous | Thread Next
Dan Sugalski
August 6, 2001 06:39
Re: Opcode Dispatch
Message ID:
At 09:31 AM 8/6/2001 -0400, Bryan C. Warnock wrote:
>On Monday 06 August 2001 09:16 am, Dan Sugalski wrote:
> > It's probably the name causing problems. Unless things change, everything
> > in the executable bits of parrot's bytecode stream will be 32 bits. We
> > might drop that to 16 bits, but that puts a heavy burden on branches I'd
> > as soon not have to deal with if we can avoid it.
>Okay, I may be slow, but I make mistakes.  I'll probably rework it tonight.

Cool. I really, *really* appreciate the work you're doing on this. It's 
definitely cemented the idea that the op dispatch loop will be generated 
dynamically, and will be different on various platforms. The wins you saw 
with different schemes on the x86 and SPARC are too big to ignore. That 
should be a win for ultimately TILing the code, since we'll need to make 
sure all the opcodes can be either stuffed bodily into the switch, or 
called as functions. Cross-platform tuning is always so much fun... :)

FWIW, I should have a preliminary assembly language document done by this 
afternoon some time, with at least a good chunk of the core instruction set 
spec'd out.


--------------------------------------"it's like this"-------------------
Dan Sugalski                          even samurai                         have teddy bears and even
                                      teddy bears get drunk

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About