Wei correctly pointed out in the discussion of D29436 that we cannot properly deal with control flow even with the fixes proposed there. This is also hard to fix as we do not track liveness for reserved registers (we only have the defs but no full live ranges available). So instead of trying to fix things for control flow we should simply abort:
joinReservedPhysReg() can only deal with a liverange in a single basic
block when copying from a vreg into a physreg.
The testcase is to check COPY to reserved physregs instead of from.