AMDGPU: Split flat offsets that don't fit in DAG

We handle it this way for some other address spaces.

Since r349196, SILoadStoreOptimizer has been trying to do this. This
is after SIFoldOperands runs, which can change the addressing
patterns. It's simpler to just split this earlier.

git-svn-id: 91177308-0d34-0410-b5e6-96231b3b80d8
12 files changed