GitLab

Additionally it is subsumed by 'SETCC_TR_LEAxx' anyway.

Do not make a separate type for the x87 80bit floating point mode, but adjust the mode's type accordingly.

This removes unnecessary special case handling for 80bit floating point.

This slightly adapts the relevant code from the IA32 backend.  As soon
as a cleaner solution is implemented there, the AMD64 code should be
updated as well.

Before, printing an Address node that did have an entity attribute set
yet crashed.  Unfortunately, when using the debug extensions to break on
the creation of an Address node, the attribute was not yet set and the
debug helper still wanted to print the node, which crashed.

Must emit .quad directive for DW_FORM_ref_addrs on 64-bit targets; .long
on 32-bit targets.

It is unused.

This fixes backend/f2ull.c.

* 'get_atomic_ent_value()' is a wrapper, which tries to unify the different kind of initalizers by wastefully creating nodes on demand.
  It had only a single user, which tried to use it for method entities, which does not work anymore (see 3rd bullet point).
* 'set_atomic_ent_value()' is a thin and unnecessary wrapper to set a const node initializer.
* 'is_atomic_entity()' classifies entities having (among primitive and pointer types) method type as having an atomic initializer.
  They do not for a long time (was an address of itself).
  Except for the two functions above, this function is unused.
* 'Atomic' is a misnomer and should rather be 'scalar'.

The pointer modes must be identical now.

If a Sub has pointer mode, then the left operand has pointer mode, too.

This fixes opt/proc_cloning007.c.