Can I now rely on the cfg and basic block execution counts in
FINAL_PRESCAN_INSN ?
(Rely as in: we won't encounter dangling pointers, and we get
reasonable heuristics if at least -fguess-branch-probability
is enabled, and good heuristics if we use relevant profile based
feedback)
If yes, when did we aquire this feature?
If no, is this on the roadmap, or could it at least be fitted into the
roadmap and implemented without becoming an expert in the cfg / dataflow
internals first?
I have some detail code generation decisions to make depending on
delay slot scheduling, branch shortening, insn alternatives and exact
intruction lengths / alignment, and branch probabilities.
