Faster optimize_mode_switching

[Jan Hubicka]

> On Mon, Jan 14, 2002 at 10:32:06PM +0100, Jan Hubicka wrote:
> > > What we shouldn't do is just leave the live_at_start/end unset.
> > 
> > Do we really, if we are asked for global update?  I guess each pass
> > that clobbers cfg needs to ask for global one.
> 
> A global update with start blocks will not change anything 
> but the start blocks if the liveness at the beginning of
> the start block doesn't change.
That what it should.
If we keep with the current CFG API, only way to create new empty BB and not
use create_basic_block directly is split_block.

It does know how to update the liveness on block boundary, so it should not
make the update needed.  If something changes, it must come from code outside
the block so global update will touch it, as it will relalize that liveness
elsewhere propagated into the block.

I guess I will get more experience from using the flags than I am having now.
If I will hit the problem till 3.2 unfreeze, I will definitly change the behaviour.

It is probably not that important whtehr the block is updated or not, as the
local properties are trivial to compute.

Honza
> 
> 
> r~