The same check is just a few lines above.

The function also returns true for zero even though that is not a power
of two.

If the cached nodes like NoReg_GP, ... had no keep edge and no users
then irgwalks could miss them leading for the iredges code to miss
initializing these nodes.

This fixes bug 161.

This reverts commit 19cccd2b.

The situation can actually occur:
The spiller might reload an operand of a CopyKeep right before the CopyKeep.
This defeats the entire purpose of the CopyKeep, so the no-spill Copy right before the CopyKept node is necessary.

Except for two users, all users add 1 immediately anyway.
Of the two exceptions one of them (in kill_unused_stacknodes()) is wrong, so this is now corrected.
The other one is exchange(), which now just uses ->in, which is fine, since it handles node interna anyway.

The kills value constraint indicated that the value in a register is
modified after it is read. A copy is inserted if the value is still
required after a kills_value use.

I can't explain why the copy was created there, the comment explaining
it makes no sense.

This case cannot happen anymore nowadays, instead we have some code that
turns CopyKeeps into Keeps after the SSA reconstruction there.

It is like new_r_Proj(), but the mode is determined automatically from the out requirement of the predecessor.
This probably resolves some mode mismatches in untested code paths -- AMD64 backend and SPARC hard float, I'm looking at you!

Now register requirements always have have a register class.
This is more consistent and removes some null pointer checks.

- Rename other_different field to must_be_different
- Remove arch_register_req_type_must_be_different, it's enough to see
  if the must_be_different field is not zero.

- Rename other_same field to should_be_same.
- Remove arch_register_req_type_other_same, it's enough to see if the
  should_be_same field is not zero.

When there is a must_be_different constraint, it must be handled.
The mode of the value should not influence this.
This was some very old code.

This should hold now that no more empty spill infos are generated for Phi arguments.

This removes the last attributes from struct arch_env_t.

be: Move the info about the register classes and registers struct the arch_env_t into struct arch_isa_if_t.

Just use the selected isa interface directly.