Re: [PATCH][PR tree-optimization/67816] Fix jump threading when DOM removes conditionals in jump threading path

[Jeff Law <law@redhat.com>]

On 10/08/2015 03:56 AM, Richard Biener wrote:
> On Thu, Oct 8, 2015 at 12:00 AM, Jeff Law <law@redhat.com> wrote:
>> On 10/07/2015 02:26 AM, Richard Biener wrote:
>>
>>>
>>> Hmm, other passes avoid all this by not removing edges or blocks
>>> themselves but leaving that to cfgcleanup.  They simply replace the
>>> condition in GIMPLE_CONDs or the switch value in GIMPLE_SWITCHes and
>>> let cleanup_control_expr_graph do the hard part of the job.
>>>
>>> Now - I don't know how that would interact with jump threads
>>> covering those paths, that is, what kind of "mess" jump threading
>>> would produce with the conditions/switches mangled that way.
>>>
>>> The other nice thing is that CFG cleanup properly deals with loop
>>> structure changes.
>>
>> Threading will handle it correctly as that's essentially how we handled it
>> until last week ;-)
>>
>> The "problem" is the threader thread through those blocks.  And while that
>> works correctly, it is quite wasteful in terms of compile time and memory
>> consumption due to the block duplication required to preserve side effects.
>> It also churns the blocks & ssa-names used in the isolated regions which
>> makes comparing debugging dumps before/after threading much harder than it
>> ought to be.
>
> Yes, but as you remove jump threading paths you could leave the CFG change to
> cfg-cleanup anyway?  To get better behavior wrt loop fixup at least?
So go ahead and detect, remove the threading paths, but leave final 
fixup to cfg-cleanup.  I can certainly try that.

It'd actually be a good thing to experiement with regardless -- I've 
speculated that removing the edges in DOM allows DOM to do a better job, 
but never did the instrumentation to find out for sure.  Deferring the 
final cleanup like you've suggested ought to give me most of what I'd 
want to see if there's really any good secondary effects of cleaning up 
the edges in DOM.

jeff