GitLab

[r15830]

[r15829]

[r15828]

[r15827]

[r15826]

[r15825]

[r15824]

[r15823]

[r15819]

[r15818]

  and tarval_eor not getting correctly hashed)
- tarval_null and tarval_one are defined for irms_reference now
- introduce get_tarvall_all_one and get_mode_all_one which returns the value
  where all bits are 1 (defined for int, boolean and reference modes)
- fix TARVAL_CLASSIFY_ALL_ONE with this

[r15817]

Adapted execfreqs
Now an order of magnitude faster... at least in c-lex.c

[r15816]

[r15815]

[r15813]

[r15810]

[r15809]

[r15808]

[r15807]

make call arg/result verify a bit looser in backend (because we have half-fixed mode_b irgs sometimes

[r15806]

[r15805]

[r15804]

[r15803]

[r15802]

[r15801]

[r15800]

[r15799]

- removed some set_*_left(), these are bad for CSE
- CSE BugFix: lookup did not normalize the nodes right!

[r15798]

[r15796]

[r15795]

[r15794]

[r15793]

[r15792]

[r15791]

[r15790]

[r15789]

[r15788]

- Do not generate dest am for 8bit operations, because the stricter register constraints are not handled yet.

[r15787]

[r15786]

[r15785]

[r15784]