[r14479]

[r14438]

[r14330]

finish support for custom backend node attributes, separate x87 attributes from normal ia32 attributes to test this

[r14328]

- "Fix" TEMPLATE backend
- Load TEMPLATE backend in bemodules
- fix ppc32 not scheduling alot of nodes

[r14319]

[r14287]

[r14277]

[r14275]

[r14270]

[r14261]

[r14205]

emit execfreqs for all blocks. ignore the am_flavour setting and look at the actual values of the AM variables

[r13915]

workaround memperm problems with different entity sizes (not nice, but still better than the assert until we have a new spillslot coalescer)

[r13841]

last change with mode suffix was wrong for x87 nodes, don't exchange left/right for dest addr mode for non-commutative nodes

[r13751]

[r13749]

[r13652]

[r13587]

unified .h include guards
fixed include structure (every files includes needed headers)

[r13560]

[r13508]

[r13446]

[r13441]

[r13401]

- Started working on flags

[r12716]

- Need to mark used entities globally, not per-irg

[r12703]

[r12643]

  switching in x87
- Only emit externally allocated symbols that are actually used

[r12625]

- more work on state registers for x87 fpu rounding modes

- Replace limit callback in register requirements with a simple bitset
- Fix backends, generators and the rest of the code for the change register
  requirements structure
- Eliminated the custom register_requirement structures in the backend, they
  are pointless now, use the normal arch_register_req_t
- Fix TEMPLATE backend
- Extract emitter framework from ia32 and provide generic header/code for it
- Port TEMPLATE backend to new emitter
- Extract gas specific emit stuff (declarations, sections) from ia32 backend
  and provide it as generic header/code. Make TEMPLATE backend use that
- Remove several FIRM_DBG_REGISTERs from functions which are called very often.
  Before we had millions of string hashtable lookups because of that. The
  prefered way is to have 1 static debug module per file and initialize it in
  the module constructor. No passing around in environment structs and in no
  case should you dynamically register it in single functions.

	 when the operand is a constant
- workaround gas bugs which swaps fsubp,subrp, fdivp, fdivrp

- correctly capture spills and reloads in coalescer code

- use mode_Iu for all nodes producing integer values (because registers are mode_Iu), this gives more possibilities for cse
- use mode_D for all nodes producing float values, this gives more possibilities for cse
- Introduce NoReg and Unknown nodes so we don't have strange projs from the start barrier
- Rewrite transform phase:
  - while transforming we now build a completely new graph on a new irg (while looking at the old graph)
  - no need to transform constants in advance anymore
  - several smaller fixes+cleanups