Re: New assembly language instructions to support OO languages?

Ray Dillinger <bear@sonic.net>
Sat, 13 Dec 2008 09:36:17 -0800

          From comp.compilers

Related articles
[32 earlier articles]
Re: New assembly language instructions to support OO languages? gah@ugcs.caltech.edu (Glen Herrmannsfeldt) (2008-12-09)
Re: New assembly language instructions to support OO languages? kamalpr@hp.com (kamal) (2008-12-10)
Re: New assembly language instructions to support OO languages? jasen@xnet.co.nz (Jasen Betts) (2008-12-11)
Re: New assembly language instructions to support OO languages? first@last.name (Morten Reistad) (2008-12-12)
Re: New assembly language instructions to support OO languages? torbenm@pc-003.diku.dk (2008-12-12)
Re: New assembly language instructions to support OO languages? johnzabroski@gmail.com (John \Z-Bo\Zabroski) (2008-12-13)
Re: New assembly language instructions to support OO languages? bear@sonic.net (Ray Dillinger) (2008-12-13)
Re: New assembly language instructions to support OO languages? jasen@xnet.co.nz (Jasen Betts) (2008-12-14)
Re: New assembly language instructions to support OO languages? gavin@allegro.com (2008-12-16)
| List of all articles for this month |
From: Ray Dillinger <bear@sonic.net>
Newsgroups: comp.compilers,comp.arch
Date: Sat, 13 Dec 2008 09:36:17 -0800
Organization: Organized? Seems unlikely.
References: 08-12-014
Keywords: architecture, OOP
Posted-Date: 13 Dec 2008 18:18:39 EST

Tony wrote:


> To me, it seems like "reducing everything to a function" may be a bit
> dated given that OO languages are the thing nowadays. Can anyone
> imagine any new potential assembly language instructions that would
> make implemention of OO languages easier? (Not just necessarily the
> function thing, but anything).


If I had my druthers, I'd add a number of VERY wide registers
replacing on the chip die the "implicit" registers known as cache
lines and make the management of these explicit with its own
instructions rather than implicit and heuristic.


The payoff for the extra instruction bandwidth would be several fairly
simple instructions to operate on the data in the wide registers in
parallel. Aside from generally useful things like supporting parallel
mathematical operations for tight loops and just plain wide-data
operations like multiple coordinate transforms in parallel for
graphics, I'd add a few instructions specifically to support OO.
These instructions are:


WMASK WIDEREG1 REG WIDEREG2 Load each word in wide register 2 with
                                                          the result of AND'ing the corresponding
                                                          word in Wide register 1 with the contents
                                                          of Reg.


WHASH WIDEREG REG Load Reg with the results of hashing
                                                          widereg.


WHERE WIDEREG REG1 REG2 Load Reg2 with the address within the
                                                          wide register of the word having the
                                                          same value as Reg1.




Out of these instructions I'd manufacture a fast multiple dispatch for
a dynamic method table for languages with type tagged values:


Load the arguments of a function call into widereg 1,
Load a "method table index" block into widereg 2,
mask the type tags from widereg 1 into widereg 3,
hash widereg 3 into reg A,
Find the address in widereg 2 that has the same value as reg A,
Add that address to the address of the base of the method table,
Jump to the result.


                                                                Bear


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.