Re: [Sbcl-devel] unexpected optimization notes

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

On Tue, Dec 07, 2004 at 02:59:38PM +0000, Christophe Rhodes wrote:
> Nathan Froyd <fr...@cs...> writes:
> > If there are no objections, I can check in something in the next few
> > days for all platforms.
> 
> I think something like this is unlikely to be wrong.  It's probably
> going to be a little unlikely to get every single case right first
> time, so it's probably worth getting this in sooner this notional
> month rather than later, but unless Alexey chimes in at this point
> with a reason why it's bad and cannot possibly work, I'd say go for
> it.

Fixes for x86 checked in as of 0.8.17.24; partial fixes for other
platforms have also been included.  The final piece of the puzzle
for other platforms will be to rewrite
UB32-STRENGTH-REDUCE-CONSTANT-MULTIPLY in src/compiler/generic/vm-tran
to accept a WIDTH argument (probably 32 or 29; could be generalized
to 64 bits, I suppose), change its logic appropriately, and define
appropriate transforms for * on other platforms.  Currently, I suspect
the tests will fail on #-x86 for lack of a fixnum *-with-constant-arg
transform.
-- 
Nathan | From Man's effeminate slackness it begins.  --Paradise Lost