[r15999]

[r15997]

[r15986]

[r15927]

- more sophisticated address mode code selection, brings huge speedups in
  164.gzip

[r15918]

- Allow Source/DestAM with 8/16bit modes where this is profitable

[r15906]

[r15882]

[r15766]

[r15730]

[r15716]

[r15690]

[r15667]

[r15657]

- Source and Destination address mode are now constructed during the normal
  transform phase and not as a separate phase
- Remove the old ImmOp stuff
- introduce 8, 16bit cmps, tests
- still some nodes miss SourceAM handling, but generally we should be as good/better than the old code, so I'm comitting now

[r15642]

do Conv after Set command (instead of mov 0,) before the set, should be as fast as before but opens up other optimisation possibilities

[r15591]

[r15494]

[r15491]

fixed several AM problems (and disabled most of the new optimisation possibilities again, as we have problems with heights not being recomputed)

[r15433]

[r15419]

[r15329]

[r15172]

Do not unconditionally set the unsigned pnc flag, but only for floating point comparisons. This fixes r15142.

[r15153]

[r15142]

[r15057]

[r15016]

[r14926]

[r14919]

- Cmov and Set can do AddressMode (for the comparison not for the result) now
- Rewrote should_be_same constraint handler and fixed a little bug in it

[r14879]

[r14858]

[r14832]

[r14827]

For a TestJmp it is not necessary that the And has a constant as operand. Every kind of operand is ok.

[r14826]

[r14821]

[r14815]

[r14727]

[r14716]

[r14635]

[r14604]

[r14559]

[r14513]