GitLab

[r25767]

[r25766]

[r25765]

- new style initializers are not yet dumped ...

[r25764]

[r25763]

[r25761]

- Also generate new_<nodename> functions
- Fixed new_d_Block

ir_cons.c
- Switched to gen_ir.py

[r25759]

[r25757]

- remove double analysis info drop (dominance might be still ok!)

[r25756]

[r25755]

- add aa_opt_no_alias_args and aa_opt_no_alias_args_global flags

[r25754]

[r25753]

[r25751]

[r25750]

[r25749]

[r25748]

- BugFix: cache the newly created memop, not the old one
- BugFix: must search for the Phi-translated load in the predecessors avail set
- BugFix: size of bitset is now equal to the address size (was it before too :-)

[r25720]

- phi-translate the antic sets

[r25719]

  interface
- get_Block_cfgpred_pos() added and used
- more doxygen docu

[r25718]

- update addesses of nodes due to Lod replacements
- remove unused (and wrong) code
- BugFix: used right condition to stop search of partly redundand Loads
- use Phi-translated addresses for cleanup check
- BugFix: correctly link in newly created memops into the memop chain of the previous blocks
- BugFix: always update the value map

[r25714]

[r25713]

- remove block flags for now, use block marks for marking blocks
- BugFix: fix analysis for IGNORE Stores
- BugFix: fixed creation of partly reduntant Loads: must be done BEFORE gen/kill
- removed now unused code
- kill unreachable blocks before analysis helps greatly and is simple, as we have
  to collect these data anyway

[r25703]

[r25702]

[r25701]

[r25700]

[r25699]

  exist

[r25698]

- BugFix: correctly handle different modes when creating a Phi (CopyB lowerer always uses unsigned modes :-()
- Handle loads from constant entities

[r25675]

  use local out edges (like old implementation)
- BugFix: always build Phi nodes on merge points: the lazy techniques did not
  work that simple ...
- in rare cases Loads and Stores might have NO memory

[r25663]

[r25662]

- only calculate avail_out for the blocks in first step (like GVN-PRE)
- correctly update avail_out during load_insert
- do NOT WAW same stores

[r25660]

[r25655]

[r25653]

[r25647]

[r25646]

  and the now (yet incomplete) LD/ST optimization

[r25645]

[r25643]

[r25641]

[r25640]

- Renamed type_walk_plus_frames to type_walk_prog.
- irio: Added support for the other segments

[r25598]