DImode loads using the stack pointer

[Clarke, Stephen]

> From: Michael Matz [mailto:matz@suse.de] 
> Sent: Friday, September 06, 2002 5:00 AM
>
> On Thu, 5 Sep 2002, tm wrote:
> >         fmov.s  @r15+,fr4       <- here
> >         fmov.s  @r15+,fr5       <- here
> >
> > This is probably bad, because if any other stack-pointer relative
> > accesses were generated in between
> 
> But there aren't.  Did you observe that behaviour somewhen?  Is it a
> problem of the register allocator, or a problem of a later 
> phase, which
> maybe optimizes the insn stream to not use immediates where possible?

It's a requirement that the stack pointer always points to
the lowest allocated address on the stack.  Anything below the
stack pointer is in danger of being overwritten, e.g. by an
interrupt handler.

Steve.