Re: [Sbcl-devel] improvements for x86 that should be applied to x86-64

[Nathan F. <fr...@cs...>]

On Wed, Mar 14, 2007 at 11:48:20PM -0400, James Y Knight wrote:
> Hi, I notice that sometimes changes get made to sbcl/x86, that also  
> should be changed on sbcl/x86-64. It'd be really nice if people  
> making such changes could add a note to the TODO file (or somewhere  
> else if there's a more appropriate place) that it should be applied  
> to the x86-64 backend too. Even nicer if both are done at the same  
> time, but making a note that it should be ported is almost as good.
> 
> Two that prompted me to send this email (if I'm wrong about these  
> being applicable to x86-64, my apologies):
> 1.0.2.1: DATA-VECTOR-{REF,SET}-WITH-OFFSET
> 1.0.2.4: Improve inlined array heap allocation on x86

The other two that I can recall offhand are:

- Shorter raw slot access code (the NEWS file states that this was
  applied to x86 and x86-64, but I believe it never made it to x86-64);
- The x86's support for improved LOGBITP and LOGTEST (maybe one or
  two others?).  Christophe, Alastair, and Juho all had comments
  about how to improve the interface for doing this sort of general
  optimization (letting the backend decide how to implement a particular
  function call), so I was waiting until I had time to rewrite the
  interface before "porting" the improvements to x86-64.

Alastair's recent sub-byte-sized data vector ref improved could also
be ported over easily.

(If anybody reading this wants a quick introduction to SBCL internals
and a small project to experiment with, the "improved inlined array
heap allocation" and the "shorter raw slot access code" bits are pretty
mechanical.  The DATA-VECTOR-{REF,SET}-WITH-OFFSET changes are also
fairly mechanical, but there's a lot more tedium.)
-- 
Nathan | From Man's effeminate slackness it begins.  --Paradise Lost

The last good thing written in C was Franz Schubert's Symphony Number 9.
                                               --Erwin Dieterich