- We now always output dwarf3 so we can use DW_OP_call_frame_cfa and avoid
  construction location lists, but just reuse the callframe info lists.
- Backends have to emit debug info as callframe calculation changes:
  The ia32 backend has a preliminary implementation which assumes esp
  offset of frame_type_size at the beginning of a block (currently
  always true), the no-omit-fp mode assumes ebp relative addressing
  (which is correct except for the prolog/epilogue insns)

unknown_type, code_type, none_type, unknown_entity reference are hold in
the irprog now. This makes handling more consistent since now all types
and entities are equally part of irprog.

returns_twice handling was broken in the ia32-backend (the only real
beabi user) for a long time now. Just noone noticed because our
frontends lacked the special treatment to set the flag correctly.
Turned out setjmp/longjmp on ia32 save/restore the registers anyway
so not doing anything is fine here.

This tries to get the names in a consistent format. We basically have 2
views on register allocation/constraints now:

1) Register Requirements and Assigments are per-instruction. Each
   instruction has requirements on its inputs and outputs, register get
   assigned for outputs of an instruction (assignment is an attribute of
   an instruction, not of the Proj-nodes).
   The internal datastructures model this form!

   The functions here have the form:
     arch_get_irn_register_req_in(node, input_nr)
     arch_get_irn_register_in(node, input_nr)
     arch_get_irn_register_req_out(node, output_nr)
     arch_set_irn_register_out(node, output_nr, register)

2) Register Requirements and Assignments are on firm-values. This view
   allows to query/assign requirements and registers at the Proj nodes
   instead of the repsective predecessors.
   This is a convenience interface modeled on top of the other!

   The functions have the form:
     arch_get_irn_register_req(node)
     arch_get_irn_register(node)
     arch_set_irn_register(node, register)

So sparc can handle this case now.

Use a special kind of entity on the frame type instead of a value_type
struct inside a method type. This makes replacement of function types
slightly easier (it's still a complex operation though) and handling in
the backend a bit more consistent since it's more or less a normal stack
access (with special offsets).

The callee/caller saved information is not constant accross different
ABI's, so don't make it constant. Instead, all BE that still use beabi
provide a callback now.
This allows to implement support for x64_64/Win32 and is a necessary step
for the combined x86 BE.

merge start-block successor with start-block in backend (necessary after equivalent_node_Block is gone)

Fixes opt/localopts2

Because of Phis blocks cannot be equivalent. The cfopt must do all
block merging.

For example, consider the case where block A has one predecessor
block B, which contains a Jmp. Naively, we might exchange(A,B).
However, A might contain a Phi with one operand, then the graph
is broken afterwards, since we move the Phi into B as well.

Difference of idx for Calls with no order relation is stable,
but may lead to circular dependencies. Fixed by looking at
the heights first.

Fixes backend/transform_loop.c (with -O0)

[r28436]

[r28340]

[r28337]

[r28272]