This removes the 2GB process limitation I ran into.
(With the EH frame format we are using the code offset is specified as a 32bit signed offset)
I checked if aligning the code speeds it up but I did not notice any perf change.