(In reply to ubizjak from comment #15)
> (In reply to Dave Airlie from comment #14)
> > I think Roland's first patch is correct, just call fold_alu_op2 if we get
> > back 2 sources from fold_assoc. return true is for when we've finished all
> > folding on that instruction, so I don't think that's correct without calling
> > fold_alu_op2
> 
> LGTM then (it also fixes the failure for me), but can we please drop the
> "else" from the patch, it is really confusing.

Sure that can be done. (I was actually also wondering if the test for
n.src.size() < 3 at the beginning of the function is actually necessary, are we
really supposed to have already fewer sources when this is called? But I
couldn't quite answer that. This code is a bit confusing to me.)