[Sbcl-devel] Re: [Sbcl-commits] CVS: sbcl/src/compiler/ppc cell.lisp,1.10,1.11

[Christophe R. <cs...@ca...>]

Nathan Froyd <nf...@us...> writes:

> Update of /cvsroot/sbcl/sbcl/src/compiler/ppc
> In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv30066/src/compiler/ppc
>
> Modified Files:
> 	cell.lisp 
>
> -(define-vop (instance-ref slot-ref)
> -  (:variant instance-slots-offset instance-pointer-lowtag)
> -  (:policy :fast-safe)
> -  (:translate %instance-ref)
> -  (:arg-types * (:constant index)))

So it turns out that on this platform, the PPC, and (from reading the
code) the SPARC as well, this VOP is not redundant.  The issue is, I
think, that the other %instance-ref transform, based on
word-index-ref, asserts that its first argument is of primitive type
instance (rather than *).

The last compilation output from elgar's overnight build log was

; file: /home/boinkmark/scratch/autobench/src/code/target-type.lisp
; in: DEFUN LAYOUT-OF
;     'NULL
;
; note: deleting unreachable code
;
; note: deleting unreachable code
unhandled condition (of type BUG):
    full call to %INSTANCE-REF

Now, this is odd, because in LAYOUT-OF the call to %INSTANCE-REF is in
the true branch of a conditional on %INSTANCEP, and I'm pretty sure
that that is enough to tell the compiler that the type is definitely
INSTANCE.  So, I'm not sure what's going on there, but I do observe
that the other platforms' DEFINE-FULL-REFFER for %INSTANCE-REF has a *
type restriction, not instance.

Cheers,

Christophe