[PATCH] ifcvt: Handle INSV patterns properly when a temp reg is needed
Wed Apr 5 14:33:00 GMT 2006
On Mon, 3 Apr 2006, Andreas Krebbel wrote:
> > At that point I gave up :-(. If you're interested in tackling
> > the BIS_BIG_ENDIAN issue, I'm happy to help by bootstrapping and
> > regression testing candidate patches on ia64-unknown-linux-gnu
> > and ia64-hp-hpux11.2x if you don't have access to them.
> Thanks for the offer. What do you think about the attached patch?
> With the patch noce_emit_move_insn undoes the adjustment done by
> store_bit_field for BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN machines
> before calling store_bit_field again.
> Could you please test it if you think it looks promising?
Sorry for the delay, establishing a baseline for mainline and then
testing your patch took a while on these machines. The good news
is that your patch bootstraps and regression tests without
problems on both ia64-unknown-linux-gnu and ia64-hp-hpux11.22.
It resolves PR rtl-opt/14261, and doesn't suffer the flaw that
Jim Wilson discovered in my original fix for this problem.
I've Cc'd Jim on this e-mail. If he's also happy with this
solution, it's OK for mainline. We can investigate the paradoxical
subreg issue and what to do about 4.1 as follow-up patches once
this is in the tree.
More information about the Gcc-patches