Committed, Fix part of PR target/19065; an autoincrement-free mem constraint for CRIS.

[Mark Mitchell]

Hans-Peter Nilsson wrote:
> This is necessary for PR target/19065.  Actually, 'Q' was used
> all along in cris.md to only mean (mem reg) and never (reg), so
> the work there was mostly formatting adjustment and review to
> make sure, which found the miss in "cmphi".

I'm content for backend maintainers to use their best judgement, so long 
as they act in keeping with the idea behind the regression-only rule; 
namely, that only conservative patches that fix very important problems 
are applied.  Regressions are used as a proxy for that because we assume 
that people were using the previous compiler with whatever bugs it had. 
  However, if there's a bug that's so severe that you think it's vital 
it be fixed, and the fix is entirely within target-dependent code, I'm 
happy for you to use your best judgement.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304