GitLab

- Fixed bug in memperm scheduling

- Added some naive rematerialisation code, which should work for constants,
  reloads (so that no 2nd spill is created) and should later support stackparms
  once backend leaves them in the graph.

- make liveness computation ignore end node (to avoid unnecessary interferences because of keepalive edges)

- place memperm projs into the same block as the memperm
- schedule all projs of memperm

- Fix spillslot coalescing timer not being displayed

use custom value_interfere function in verifiers (which is slower but doesn't rely on correct liveness information), fix memperm creation

- new spillslot verifier (doesn't catch all errors at the moment, because lifeness analysis ignores non-schedule nodes)

- No need to set strange spillcontexts on spills anymore
- Spillcontexts are now assigned and coalesced in a separate pass after spilling/register allocation of all register classes
- There might be cases where the new code inserts memcopies, these are not implemented in the ia32 backend yet