"+m" constraints bogus?

[Jan Hubicka]

> On 7/24/07, Jan Hubicka <jh@suse.cz> wrote:
> >I will try to find the original Jason's problem, does anyone recall?
> 
> well the gimplifier no longer allows "+m" to pass to expand, we always
> expoand the constraints.

Yep, I noticed we decompose it now, however it still does not quite
answer whether we can accept +m now or not... (ie whether there exists
testcases where reload desynchronize input and output constraint for
some good reason).

Honza