[r26956]

[r26955]

- if a node was built with edges disabled gets dead, but is revived by CSE when edges are enabled, it had NO edges
- tries to keep track of such dead nodes ...

[r26954]

[r26953]

[r26952]

[r26951]

panic instead of wrong results for Set and CMov with float compare. Rename Set to Setcc and CMov to CMovcc

[r26950]

Small optimization. Note that the else case crashes libFirm and must be fixed, or disable inlining with exceptions yet.

[r26943]

[r26940]

the find_pointer_type interface still needs a mode because we might have pointers with different modes ...

[r26938]

the find_pointer_type interface still needs a mode because we might have pointers with different modes ...

small C99 fix

[r26937]

- fixed memory leak
- invalidate the analysis info only if the graph contains muxes
- fixed linking of Proj nodes
- fixed linking of mbh lists: no need to manipulate them, as new newly created block is a mbh itself

[r26936]

- update docu: collect_phiprojs() puts all partBlocks into a list of its mbh, part_block uses this info
- the lower block (old_block) of a splitted block is always a mbh
- part_block sets and restores current_ir_graph and did not depends on its preliminary setting

[r26935]

[r26932]

[r26930]

[r26926]

[r26924]

[r26922]

[r26919]

[r26918]

[r26917]

[r26916]

[r26915]

add comments and descriptions to mode_b lowering; reformat lower_mode_b.c to latest coding conventions

[r26914]

Liebe bugs part1: Eor(1, Proj(Cmp)) was broken - simply remove the optimisation since Eor(1, X) gets transformed to Not anyway and Not(Cmp) gets optimized

[r26913]

[r26910]

- There's a new type_dbg_info* that allows you to attach debug names to types
- get_type_name and get_type_ident are now more. The new ir_print_type
  and the usual ir_printf("%+F", type) help in most usage cases.
  But you should be aware that names are not guaranteed to be unique anymore
  (or positively said: You don't have no trouble anymore building unique names
   in code that creates types)
- No need to specify mode for new pointer types anymore (you can still do it
  with set_type_mode)

[r26909]

[r26904]

 - the upper block must dominate the block, else we cannot fuse
   upper and lower block

[r26899]

- place new constructed code into the right block, which might
  be neither the block of any Cmp
- only the <null, null> test can have different operands, all others
  must have one common op in both compares

- remove the meaningless <null, null> message

[r26898]

[r26897]

[r26896]

[r26895]

[r26890]

[r26885]

[r26883]

Allow the modulo-shift setting for reference modes. This is a work-around that fixes find_unsigned_mode().

[r26882]

[r26880]

(there is retrieve_dbginfo stuff already there! And don't tell me about possible
 future extensions where sets of locations are merged and someone generates
 usefull debug info from that. Noone uses this currently!)

[r26878]

- fixed a lot of comments

[r26877]