propagate_freq enormous performance hit in 3.1

[Jan Hubicka]

> 
> Michael, you improved the speed of some of the other flow routines
> before, do you know whether similar algorithms can apply here?
Since I can't get that far - can you take a look how does the natural loops
look like?
How much are they nested?

Limiting to fixed nest is IMO not problem, as the outermost loops will get
frequency of 0 anyway.

Honza
> 
> Brad