GitLab

This puts the special logic for cfops into the common code instead of
repeating it in all schedulers.

I don't see a reason why this would be bad, and I have such a case now
so I removed the assert.

We now add missing keeps exactly before register allocation where we
actually need them (so that unused outputs which still need registers
assigned are not missed by the register allocation).

This removes functions from the public API which are pointless outside
localopt rules, and also avoids another struct in the public API.

The start block is not special block without any "real" instructions
anymore. In particular this means:

- There is no X_initial_exec Proj from the Start node anymore, "normal"
  cfopts are used for jumping out of the start block.
- The backend never had the special convention of the start block not
  containing actual code and used to fixup this discrpancy, no more
  fixup code needed now.
- The part_block() code got a bit more complicated because splitting the
  start block is tricky (previously part_block() simply failed in this
  case but was never used in the Start block because of its special
  convention)

- It was only available in the ia32 backend and broken there. As the
  feature is rarely used, I remove it for now instead of going through
  the trouble of fixing it.

This should fix the 186.crafty problem.

The previous approach put the irg link into the block attributes, so
while saving memory an extra indirection was necessary. Some testing
with valgrind massif+callgrind revealed that a direct link increses peak
memory usage by ~1% while also increasing runtime by ~1%. More
importantly the code gets a bit simpler and some strange corner cases can be
removed with the direct link.

Make the list scheduler a set of helper functions, instead of a complex
design with a set of callbacks.
Simplified list scheduler code.

In their current state they produce very bad code and are just a
maintenance burden.