This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: loop.c bugfix
- To: law at redhat dot com
- Subject: Re: loop.c bugfix
- From: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Date: Thu, 9 Nov 00 05:53:30 EST
- Cc: gcc-patches at gcc dot gnu dot org
I'm somewhat curious -- off the top of my head I thought we wouldn't
generate a base + index address until combine. So I'm a little
curious how we generated such an address that early in the compiler
I'm not sure. The test case was *huge* and I didn't bother to go back and
look at how it got there. I suspect, due to some things I was seeing, that
the RTL was a shared PLUS between an add insn and a address. If so, it's
probably not valid and is another bug, but I didn't have the patience to
track it down since this change is useful anyway.
Note that this problem originally came up in GCC 2.8.1. I don't remember
whether I was able to reproduce it under the 2.96 sources, though that
wouldn't prove much either way for various reasons.
As has been said, tracking REGNO_POINTER_FLAG through loop wins for other
reasons anyway.