copy_node_attr() just a few lines before does that already.

This is not a hard constraint like the others and does not break the register allocation, if a Perm is pushed across the node with should_be_same.
This restores the behavior before the register constraint reorganization and removes lots of xchgs in the testsuite again.

Now copy coalescing sees that CopyKeep is costly (just like Copy) and hopefully produces better code.

This fixes backend/spillslot.c.

An amd64_add has mode_T by default.

This only checked that Phis are NOPs.
be_verify_register_allocation() does the same and more.

It gets overwritten just a few lines later.

We never used the distinction of modifying the carry flag or not, so just simplify the specification a bit.

This was lost for the unsigned multiplication some years ago.

The floating point control word is not modeled, is not that kind of flags and the floating point flags are not modeled explicitly.

The port can be encoded as immediate in the range of an unsigned byte ('N'), not a signed byte ('K').

- 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.

We can easily identify the outputs producing a stack pointer by looking
if the stack pointer register is assigned. Unfortunately this also
needed a new mechanism to tell the verifier that some registers are not
in SSA form (all the time) which needs a few extra lines in each
backend.

Get the obstack from the given start node.

Use a flexible array, so we can iterate over the linked list only once instead of twice.

This fixes gcc/pr43784.c.

This fixes several test on ARM.