Yes thanks it worked out exactly like you said. I first made the mistake to constrain the match_scratch 'rX' and then wondering why I am still seeing a scratch after reload. I realized you were talking about a real alternative 'r,X'. Regards, Hendrik Greving
On Wed, Sep 18, 2013 at 12:26 AM, Richard Sandiford <rdsandif...@googlemail.com> wrote: > Hendrik Greving <hendrik.greving.in...@gmail.com> writes: >> One follow-up question here. Is it of any disadvantage to always >> generate a clobber (scratch) during expansion? Meaning that in the >> case we don't need one (we won't split, which I am checking with >> memory_operand(myoutput_operand), that register won't get allocated >> and adds to register pressure, does it? > > Yeah, the idea is that if the match_scratch isn't needed for a particular > alternative, the constraint for that alternative should be "X" rather than > a register class. That way the (scratch) stays a (scratch) even after > register allocation. > > Thanks, > Richard
