GitLab

This unifies the behavior in debug and optimize build, since
be_set_transformed_node also marks the node as visited.
Furthermore, nobody should call be_is_transformed for the current node,
since this would mean we ran into a cycle (which is covered by an assert in the
debug build).

Heights are the longest chain of users, not operands, in the block.
Therefore the after nodes all have the same height of 0, because they
have no users, yet.

This fixes several tests on SPARC.

This resolves problems with hidden dependencies during code selection, which cause dependency cycles and therefore fixes backend/scheduled.c.
Also it grants the code selection slightly more freedom by not arbitrarily restricting the order by stack dependencies.

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.

The new anchor gets the transformed operands earlier already.

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

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

Other backends can use this, too.
This also corrects the bug, that for Address+Const immediates the check, whether it is a tls entity, was missing.

All Projs should be transformed into to the correct block.

new_ir_node() handles dynamic ops correctly for quite some time now.

This also renders get_irg_n_anchors() obsolete, so remove it.

This is useful for backend development.

* new_ir_node() correctly handles oparity_dynamic, so remove the special case here.
* use new_similar_node() instead of new_ir_node().

All backends have their own transformers.

This is simpler and does not require queueing them anymore.

The operands of the anchor cannot be null.