[Bug target/78166] [6/7 Regression] hash.c:1887:1: error: unrecognizable insn

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78166

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-11-01
     Ever confirmed|0                           |1

--- Comment #5 from Jeffrey A. Law <law at redhat dot com> ---
How unpleasant.  ISTM we have a few paths forward here.

1. Add back the patterns which accept non-canonical RTL
2. Hack up reload to know about the canonicalization rule and enforce it when
we reload an address
3. Handle the rewriting by using a secondary reload

#1 and #3 are PA specific fixes.  #2 means some ugly hacking in code we really
don't want to support anymore.  But #2 has the advantage that it'd address this
issue on other reload targets with shadd style insns and scaled memory
indexing.

#1 is easy and John has a patch for that.  I've prototyped #3, but don't really
like it.  I think going with #1 on the trunk and backported to gcc-6 is
probably best.