Re: other/7114: ICE building strcoll.op from glibc-2.2.5

[Geoff Keating <geoffk at geoffk dot org>]

> Date: Tue, 16 Jul 2002 09:20:27 +0930
> From: Alan Modra <amodra@bigpond.net.au>

> On Mon, Jul 15, 2002 at 12:43:02PM -0700, Geoff Keating wrote:
> > > Date: Mon, 15 Jul 2002 18:56:03 +0930
> > > From: Alan Modra <amodra@bigpond.net.au>
> > 
> > > This patch cures the testcase.  The !using_store_multiple code tests
> > > whether regs are live before saving.  We need to do something similar
> > > for using_store_multiple, in case all regs need not be saved.
> > 
> > Those registers are actually saved, whether they need to be or not,
> > correct?
> > 
> > So the RTL generated is an accurate representation of the instruction,
> > and the bug must be elsewhere.
> 
> The testcase saves r30 and r31, but both are marked unused (don't
> appear in regs_ever_live).  Later rtl analysis decides that the
> save instruction can be eliminated, thus the ICE.  The real bug is
> that r30 is not marked used when current_function_needs_context.
> This is also the reason for PR5967.
> 
> The code that I copied from the !using_store_multiple case just
> papers over this bug.  So the above patch merely makes -mmultiple
> and -mno-multiple consistently wrong.

No, the code below causes unused registers to actually not be saved,
which is correct (it does sometimes happen that all uses of a register
are eliminated after reload).  This can be done when individual loads
and stores are being used, you just don't emit that store.  It can't
be done when store-multiple is being used.

-- 
- Geoffrey Keating <geoffk@geoffk.org> <geoffk@redhat.com>