[r19906]

[r19752]

[r19713]

[r19305]

rename set_using_visited to set_using_irn_visited, some cosmetics, remove obsolete irflag functions from public headers

[r17400]

[r17143]

These should be MacroBlock edges and can be safely ignored.

[r17102]

[r15596]

[r15351]

[r15170]

[r15168]

Replaced set_irn_n(*, -1, *) and get_irn_n(*, -1) by new get_nodes_block()/set_nodes_block() implementation
caveats:
 - the old outs do not see the anchor node yet
 - the new edges wrongly use get_irn_n(*, -1)
 - the new edges do not realize that the move of a node moves all its Projs
 - the anchor node still must have the -1 input set (due to edges bug...)

[r15135]

[r14991]

[r14952]

[r14944]

[r14456]

[r14240]

Cond nodes are a kind of cf node, too. Therefore also ignore them when copying block contents during condeval. Otherwise this leads to fake users during repeated condeval and somehow to a violation of the def-use-dominance.

[r14012]

Altering (and especially removing) control flow predecessors of a block which is not the current block of a block walk is a very bad idea.
Use the proven set-predecessors-to-Bad method instead.
This fixes compiling bzip2 with cond-eval.

[r13844]

some doxygen comments added
restructured

[r13543]

[r13510]

[r12715]

[r8729]

Only reconstruct SSA for the users of the first value in the passed value array. The others are just copies and their users are not meaningful/unrelated

[r8726]

[r8725]

[r8667]

[r8265]

[r8252]

[r8236]

[r8227]

[r8215]

[r8212]

[r8211]

add missing #endif
cleaned up pnc handling

[r8202]

Runs only the analysis right now, does not modify the program graph

[r8199]