[rfc] multi-word subreg lowering pass

[Björn Haase]

Ian Lance Taylor wrote on Donnerstag, 1. Juni 2006 20:26 :
> I've also been working on this.  I'm currently running a version of
> RTH's lower-subreg pass twice, once early and once after splitting
> instructions. 
Would it be possible to sketch with a few lines what you needed to do in order 
to get the lowering working after splitting (i.e. despite the newly inserted 
pseudos)? 

That would be very helpful for me for evaluating the possible benefit for the 
AVR target. Right now, I am maybe 30% through with converting the AVR backend 
to CCmode and with using splitting patterns instead of the lowering at text 
output. Evaluating what could come out after smashing all the QImode and 
HImode subregs that I get after the splitting would be very interesting. 
Right now, I could only make guesses since gcc will crash as soon as I try to 
make RTH's original pass run after split1. I think this is due to the 
inserted new pseudos.

Bjoern.