This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: fold_const.c/tree_swap_operands_p change from Oct 11 breaks m68k
On Mon, Oct 27, 2003 at 08:46:14AM -0800, Richard Henderson wrote:
> On Mon, Oct 27, 2003 at 04:40:50PM +0100, Gunther Nikl wrote:
> > With VALUE >= 128 that doesn't happen and a 20031008 doesn't save the
> > register either. I would like to know where I should look to findout
> > why GCC saves %d2?
>
> You need to study the debugging dumps. I suspect that it was used in
> one dump, then eliminated.
The DFI_greg dumps list it as a used hardreg and reload1.c/find_reg
says it used reg 2 (%d2) for to reload a spilled instruction. Later
in the dump the reload says it uses reload_reg_rtx of %d1 for the
reload. %d2 is marked as a used spill register in used_spill_reg and
thus regs_ever_live for it is set to 1 in finish_spills().
I don't know what I am really looking for.
Anyway, are you going to commit the proposed patch?
Gunther