On Sat, Sep 07, 2013 at 09:06:08AM +0930, Alan Modra wrote: > The testcase gives me > > .L.foo: > lis 9,0x4000 > sldi 9,9,32 > addis 3,2,x@toc@ha > addi 3,3,x@toc@l > add 3,3,9 > blr > > How did you manage to get an unrecognizable insn? I can't see how we > generate the pattern except in combine.
Never mind. I updated and rebuilt from a clean tree and now see the failure too. "tocrefdi" is where combine is still munging together the large offset. -- Alan Modra Australia Development Lab, IBM
