Re: ptrlist-iterator performance on one wine source file

[Luc Van Oostenryck <luc.vanoostenryck@gmail.com>]

On Sat, Jul 29, 2017 at 6:25 PM, Christopher Li <sparse@chrisli.org> wrote:
> On Sat, Jul 29, 2017 at 12:04 PM, Luc Van Oostenryck
> <luc.vanoostenryck@gmail.com> wrote:
>>
>> You can't do that once cycles are involved. You need something
>> like the marking algorithm used by kill_unreachable_bbs() for that.
>> And it was such a  cycle that created the problem with the false
>> "crazy programmer" warning.
>
> No I don't think so. The find dominator already taking the cycles into
> account. By definition if X dominate Y, means every execution flow
> from entry point to Y will need to go through X. If X was not reachable,
> nor does Y. It does not change where the block get deleted. It just don't
> not need to do the marking algorithm. That is the point of dominator tree.

OK, I misread and misunderstood that you was talking about  the
dominator *tree*.

The real problem with such a tree is that you need to maintain it as
it potentially changes each time there is a change in the CFG.
And of course, building this tree is not linear (in the number of BBs)
while finding the dead BBs is linear.

-- Luc