GitLab

belive: Replace the union for head and nodes of the liveness sets by a struct with a flexible array member.

This fixes opt/fehler284.c.

belower: Just hand the information about the moved Perm to lower_perm_node() instead of reducing its with at the end of push_through_perm().

This was promised in the comments years ago, but was never implemented.

Now the number of outputs is adjusted, too.
This did not cause any problems, because the Perm is lowered right afterwards.

belower: In push_through_perm() merge identifying the move frontier into the actual Perm moving loop.

* This seems to remove the non-dying check.
  But the actual moving loop already had a more strict check for any operand in the Perm's register class.
* Remove the explicit is_Phi() test.
  Phis are schedule_first, so the other part of the condition covers them, too.

Registers do not have type "4" anymore, so the condition is always false.

Otherwise, we may replace an ia32_Sub by an ia32_Cmp
if the result proj has a different mode than mode_gp.

Also provide a cleaner, better documented and slightly faster
implementation.

- block_strictly_dominates() is trivially to compute yourself anyway.
- free_dom() and free_postdom() did not free any memory anyway and the
  irg_properties handling is performed by confirm_irg_properties()
  anyway.