This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Bug with copyprop_hardreg_forward and shared RTX
- From: Richard Henderson <rth at redhat dot com>
- To: Ulrich Weigand <Ulrich dot Weigand at de dot ibm dot com>
- Cc: gcc at gcc dot gnu dot org
- Date: Mon, 4 Mar 2002 12:46:32 -0800
- Subject: Re: Bug with copyprop_hardreg_forward and shared RTX
- References: <OF5642ED36.98F5EB04-ONC1256B72.005002A2@de.ibm.com>
On Mon, Mar 04, 2002 at 03:44:43PM +0100, Ulrich Weigand wrote:
> However, this appears to be done intentionally by reload,
> if the comment at that place is to be believed:
>
> The REG-rtx's for the pseudos are modified in place,
> so all insns that used to refer to them now refer to memory.
>
> So if this is the bug, do you have suggestions how to best fix it?
Hum. True, I'd forgotten about that sharing. I'd have thought in
general such substitutions would end up using the stack pointer or
the frame pointer, and so wouldn't end up being optimized.
I guess we could run an unshare_all_rtl_again pass after reload is
done...
r~