This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH, take 2] Fix PR target/28946
- From: Uros Bizjak <ubizjak at gmail dot com>
- To: Rask Ingemann Lambertsen <rask at sygehus dot dk>
- Cc: Roger Sayle <roger at eyesopen dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>, "H. J. Lu" <hjl at lucon dot org>
- Date: Sat, 07 Oct 2006 10:39:11 +0200
- Subject: Re: [PATCH, take 2] Fix PR target/28946
- References: <44FDD0D0.email@example.com> <Pine.LNX.firstname.lastname@example.org> <20061007082319.GE25625@sygehus.dk>
Rask Ingemann Lambertsen wrote:
Yes, I know... But I was looking for the safest solution to implement
new patterns. Undoubtly, there are plenty of possibilities for macro use
in i386/*.md, but rest assured that no patches for these issues will be
accepted in stage3.
It's now been a few days and RTH hasn't voiced an objection. Although
the patch is ugly, I think this is just a consequence of the way GCC's
combiner and RTL instruction selection work, and similar idioms are
commonly used on other platforms.
Note that you could likely use code macros, code attributes, mode macros
and mode attributes to reduce the number of patterns in the MD file. I think
the i386 shift/rotate instructions are sufficiently uniform that you can do
so, even with the "add reg,reg" vs. "sal reg,reg" issue.