GitLab

- Make API private as it should only be called by backends
- Let ia32 backend create special ia32_Set nodes instead of relying
  on muxes which must not be touched anymore
- Does not create ConvB nodes anymore but instead produces the Cmp
  directly. (All backends did this anyway during code-selection so I was
  able to leave this case out code-selection)
- First collect nodes to lower, then lower them. This avoids robustness
  problems when transforming the graph while at the same time walking
  it.

This fixes opt/fehler202.c.

Implement a generic way to lower compound-call-arguments.
Parameters are prepared on the stackframe and passed as a pointer
to the function.

This should also fix compound returns in combination with
-msoft-float on sparc

They really are const before lowering. So only mark the function
as non-const after lowering

This reverts commit 72c631f3.

This commit also adds a generic Alloc/Free lowering helper
Unfortunately we often produce invalid code, because currently our
stackpointer wiring logic is too weak and produces invalid results for
example when using alloca in a loop.

This fixes opt/fehler200.c.

Old version used parameter.

Backends can configure CopyB lowering, so that it's possible to keep
CopyB nodes in a certain size range for special backend-specific
optimizations.  Furthermore, large CopyBs are turned into memcpy calls.

The lowering phase had a lower_bitfield parameter.
Instead of passing "true", the (e.g. EDG) frontends now must
set_irg_state(irg, IR_GRAPH_STATE_IMPLICIT_BITFIELD_MASKING).

This is necessary, because not only the lowering phase is affected.
Additionally, the load store optimisation should handle Sels differently.

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).

This was done by removing unnecessary options/options that only
allowed marginal tweakings to improve performance while at the same time
making the interface cumbersome.

Fixes issue 64, because casting long to int does work on 64bit.

There might be some ProjL nodes reachable via out edges
and the following "local" phase does enqueue_users for
optimized nodes, which might visit the 'undead' ProjL.