RFC: using clz for comparing to zero on the tree level.

[Michael Meissner]

On Thu, May 08, 2003 at 09:27:04AM -0700, Richard Henderson wrote:
> On Thu, May 08, 2003 at 10:22:23AM -0400, David Edelsohn wrote:
> > 	Shouldn't this be based on the rtx_cost of CLZ?
> 
> Possibly, but at the moment I'm not aware of *any* other
> target for which this would be profitable.

The machine I'm just starting to work on has a single cycle CLZ amongst other
features.

I did attempt to do this about 3-4 years ago when I was involved with the
PowerPC, but we couldn't get it stabalized properly.  Now that CLZ is part of
the RTL, it might be more feasible, assuming the costs of CLZ is less than
other methods.

-- 
Michael Meissner
email: gnu@the-meissners.org
http://www.the-meissners.org