[RFC] propagating loop dependences from trees to RTL (for SMS)

[Paolo Bonzini]

> 1. Introduce a new BB bit flag and set it for the header BB of a loop that
> has no data dependencies. This approach already works, but only if the old
> loop optimizer (pass_loop_optimize) is disabled (otherwise the bit doesn't
> survive).

Which will happen in 4.2.

> One potential problem is that the loop header BB may change
> between the tree-level and SMS as result of some optimization pass (can
> that really happen?)

I don't know, but I'd not answer with a definite no.

I'm not an expert, but ISTR that SMSable loops are all do-while loops, 
and composed of a single basic block  If this is true, why not mark the 
loop *body* instead of the header?

> 3. Use insn_notes - introduce a new note "NOTE_INSN_NO_DEPS_IN_LOOP" to be
> inserted after the "NOTE_INSN_LOOP_BEG" for relevant loops.

This is a no-no.  pass_loop_optimize will disappear in 4.2, and loop 
notes will disappear together with it (and with much joy).

Paolo