Re: register renaming causes many ARM testsuite failures

[Richard Earnshaw <rearnsha at arm dot com>]

> On Wed, Jan 03, 2001 at 07:57:51PM +0000, Philip Blundell wrote:
> > >The function build_def_use uses constraints to determine which registers can
> > >be used for renaming.  The relevant bit of code is this:
> > 
> > If I'm reading that right, it just notes which class an operand is in so
> > that it can later choose another register in the same class.  The problem
> > at hand here is that the actual registers in use, not just their classes,
> > matter in an ldmsi insn.  Am I missing something about the code you quoted?
> 
> He's right Bernd.  We're going to need some way to squash renaming
> in this case.  Probably through a new constraint token indicating
> that we shouldn't consider this operand for renaming.
> 
> 
> r~

Before you go into to much re-engineering of this, I think I can fix the 
ARM back-end for this case.  The problem here is that we have an operand 
with the predicate:

	(match_operand:SI n "s_register_operand" "r")

but we pre-allocate a specific hard register to this, and don't expect it 
to change.  The work-around is to change this operand to match the hard 
reg we have pre-allocated, since there are other places that also expect 
this to remain unchanged.

I'm currently bootstrapping a patch which does exactly that.

R.