[llvm-dev] RFC: New aggressive dead code elimination pass

Fri Mar 25 16:03:35 PDT 2016

On Fri, Mar 25, 2016 at 3:43 PM, Mehdi Amini <mehdi.amini at apple.com> wrote:

>
> On Mar 25, 2016, at 3:30 PM, Daniel Berlin via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> Make most things update post-dominance info and preserve it.
>
> Our other alternative to not take up too much time seems to be invasive
> surgery on how BB successors/predecessors work so they are a constant time
> array.  I say this because GCC recomputes post-dominators roughly 15-19
> times per compilation, and can do it about 3-5x faster than we can. All
> profiling i've done basically says all our time is spent trying to get at
> successors and predecessors in dominance/post-dominance respectively, which
> takes basically no time in gcc, because the edge lists are an array.
>
> (Note that if you look at generic dominance code, like
> http://www.cs.princeton.edu/~rwerneck/dominators/, it's much faster than
> what we do on the same graphs. This is true even though we use the same
> algorithms .....)
>
> Given it seems unlikely we are going to change the internal representation
> anytime soon (or at least i've not seen a proposal), updating post-dom
> seems the easier answer.
>
>
> Are we talking about the basic-blocks edges only?
>

Yes. Only the successor and predecessors.

> I'm not sure changing the IR for the BBs would be a lot more work than
> preserving dominance analysis everywhere, or I am totally underestimating
> one / overestimating the other?
>

The way "edges" work right now is that it walks the use/users lists.  In
the case of predecessors, it walks the user list of the bb, and sees which
ones are terminator instructions, and then hands you the parent bb's of
those.

In the case of successor, the operand array stores it, but it requires some
indirect loads per successor.

The advantage to this scheme is that if you set the operand of the
terminator, you've updated the edge. It requires no other work.
Chandler was strongly against edge cuts not being super-fast constant
time[1]
If you move to edge arrays, it now requires finding and updating the
terminators, unless you keep them both as use lists somehow (which are
still about 1.5-2x as slow as straight arrays for dominators purposes), or
always have the appropriate terminator around.

In any case,  it's really hard to have a case where you don't have to
update some stuff on edge redirection, even if you can store stuff to do it
in constant time.

For example,  you would have to keep an array of (bb, index in succ/pred
array) and (terminators, operandno) or something similar to give you
constant time cutting (because you have to update both ends, so you need to
know where everything is both directions from whatever list you are looking
at)

if you can deal with linear time this is much easier.

[1]I don't think it matters as much as he does. They don't happen that
often, and even in the case of bugpoint, most blocks do not have a ton of
edges, so it won't slow much down for bugpointing real programs.  The
significantly better cache behavior may even make up for it in practice.

> --
> Mehdi
>
>
>
>
>
> On Thu, Mar 24, 2016 at 11:56 PM, Eric Christopher <echristo at gmail.com>
> wrote:
>
>> What do you have in mind here?
>>
>> On Thu, Mar 24, 2016, 7:28 PM Daniel Berlin via llvm-dev <
>> llvm-dev at lists.llvm.org> wrote:
>>
>>> Yeah, that was gonna be my question.
>>> If so, my view is we should just bite the bullet and start threading
>>> post dominance through the compiler.
>>> (assuming anyone wants to help. I'm tackling the memoryssa updating
>>> stuff with george ATM).
>>>
>>>
>>>
>>> On Thu, Mar 24, 2016 at 7:04 PM, Hal Finkel <hfinkel at anl.gov> wrote:
>>>
>>>> [+Danny]
>>>>
>>>> ----- Original Message -----
>>>> > From: "Justin Bogner via llvm-dev" <llvm-dev at lists.llvm.org>
>>>> > To: "David Callahan via llvm-dev" <llvm-dev at lists.llvm.org>
>>>> > Sent: Wednesday, March 23, 2016 12:36:50 PM
>>>> > Subject: Re: [llvm-dev] RFC: New aggressive dead code elimination pass
>>>> >
>>>> > David Callahan via llvm-dev <llvm-dev at lists.llvm.org> writes:
>>>> > > Hi,
>>>> > >
>>>> > > I have a new variant of Aggressive Dead Code Elimination that also
>>>> > > removes dead branching. It is designed to minimize the cost of
>>>> > > control-dependence analysis in the common case where almost the
>>>> > > entire
>>>> > > program is live. It also can optionally remove dead but
>>>> > > may-be-infinite loops.
>>>> > >
>>>> > > When enabled for –O3 (replacing current ADCE pass) and removing
>>>> > > loops,
>>>> > > impact on SPEC2006 is in the noise but it impacts internal
>>>> > > benchmarks
>>>> > > suites 1-2% with a comparable increase in compile time.  My
>>>> > > expectation would be to enable –O3 only until we have some
>>>> > > experience
>>>> > > with cost but I suspect it should be fine –O2.
>>>> >
>>>> > Just to clarify, you're saying that both runtime and compile time
>>>> > impact
>>>> > were in the noise on SPEC, right?
>>>> >
>>>> > > What information would the community like to see about such a
>>>> > > change
>>>> > > before I put up a diff and (including tweaks to unit tests).
>>>> >
>>>> > I'm not sure that there's much to discuss in the abstract - it's much
>>>> > easier to evaluate this kind of thing when there's a patch to refer
>>>> > to.
>>>> > Presumably people will want to try the patch out on their own
>>>> > internal
>>>> > benchmarks as well.
>>>>
>>>> +1
>>>>
>>>> Does it use post-dominance information?
>>>>
>>>>  -Hal
>>>>
>>>> >
>>>> > > Thanks
>>>> > > david
>>>> > > _______________________________________________
>>>> > > LLVM Developers mailing list
>>>> > > llvm-dev at lists.llvm.org
>>>> > > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>> > _______________________________________________
>>>> > LLVM Developers mailing list
>>>> > llvm-dev at lists.llvm.org
>>>> > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>> >
>>>>
>>>> --
>>>> Hal Finkel
>>>> Assistant Computational Scientist
>>>> Leadership Computing Facility
>>>> Argonne National Laboratory
>>>>
>>>
>>> _______________________________________________
>>> LLVM Developers mailing list
>>> llvm-dev at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>
>>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160325/0b8e36ad/attachment.html>