This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH, arm] Fix gcc.c-torture/execute/930628-1.c -O3 for Thumb
- From: Hans-Peter Nilsson <hp at bitrange dot com>
- To: Richard dot Earnshaw at arm dot com
- Cc: Adam Nemet <anemet at Lnxw dot COM>, <gcc-patches at gcc dot gnu dot org>, Nick Clifton <nickc at redhat dot com>, <rearnsha at arm dot com>
- Date: Wed, 30 Oct 2002 12:46:08 -0500 (EST)
- Subject: Re: [PATCH, arm] Fix gcc.c-torture/execute/930628-1.c -O3 for Thumb
On Wed, 30 Oct 2002, Richard Earnshaw wrote:
> > On Wed, 30 Oct 2002, Richard Earnshaw wrote:
> > > LEGITIMIZE_RELOAD_ADDRESS is only supposed to be used to optimize cases
> > > where reload generates inefficient code. It shouldn't be needed to get
> > > correct code, so this change can't be right even if it fixes the cases
> > > you've found.
> >
> > Since that seems like a general remark, let me just say that in
> > the general case (not arm/thumb-specific) that is not true.
> > LEGITIMIZE_RELOAD_ADDRESS is sometimes needed for necessity. I
> > think I've said this before. [Checking...] right; summary at
> > <URL:http://gcc.gnu.org/ml/gcc-patches/2002-07/msg00054.html>.
> > Or alternatively, those target-specific cases need to catered
> > for by reload or *another* target mechanism. Or alternatively,
> > some addressing modes on some targets can't be used.
>
> No, it just means you are papering over a deficiency elsewhere in reload.
That's what I wrote, I think, except for the valued "papering
over". Put it another way, I'm papering over that reload breaks
addresses that it needs to spill registers for, at the mem:s.
I can live with that; it seems better than changing reload.
brgds, H-P