RFD: fixup_fallthru_exit_predecessor bug (Was: Re: [PATCH/RFA] PR target/13250)

[Joern Rennecke]

P.S.: Conceivably, we might have fall-through edges all the way from
the the entry block to the exit block, and then some extra blocks
that are being jumped to in exceptional circumstances.
And there is nothing actually wrong with that.  I think we should
allow the predecessor of the exit block to be in the middle of the
insn stream.
As far as I can see now, this basically requires that the epilogue is
emitted at the end of the block with the fallthrough edge to the exit
block, or in a new block after the former block, if there are other
edges to the exit block.
Any code that assumes that the physical end of the function is also
the logical end after cfg optimizations have started would have to
be fixed up.