[llvm-dev] [cfe-dev] [LLVMdev] Clang devirtualization proposal

Fri Aug 7 18:27:40 PDT 2015

Summing up, it seems there are three choices here that are all reasonably
defensible:

1) We can change how CSE works such that equality is not sufficient for
substitutability.
2) We can change the requirements of 'readonly' (and related) parameter
attribute to be that the *memory location* is not modified by the function.
3) We can change argument elimination to not eliminate dead arguments which
might change the set of memory locations which can be read or written by
the function.

The only one of these I see as completely infeasible is #1.

For #3, this is at odds with how I understand dereferencable works, and it
seems like these should work the same way and dereferencable had
particularly good reasons to need to work the way it does.

For #2, it really does make inference significantly more tricky. Here is an
example of the new challenges: if the caller has already escaped a pointer
to the memory location (which we must assume it has) then any function
which might write to memory must be assumed to write to the memory aliased
by the passed in pointer.

Also for #2, we should probably finish converging with dereferencable and
have a size attached. This will become especially important with type-less
pointers.

I suspect #2 is the right design, mostly because I suspect most of the
interesting and important inference cases are going to be cases where we
can easily infer the stronger guarantee, and once inferred we will have
much more freedom to optimize based on this stronger guarantee... But I
can't say I'm completely confident in this model yet.

-Chandler

On Fri, Aug 7, 2015 at 6:03 PM Hal Finkel <hfinkel at anl.gov> wrote:

> *From: *"Chandler Carruth" <chandlerc at google.com>
> *To: *"Hal Finkel" <hfinkel at anl.gov>
>
> *Cc: *"Sanjoy Das" <sanjoy at playingwithpointers.com>, "llvm-dev" <
> llvm-dev at lists.llvm.org>, cfe-dev at lists.llvm.org
> *Sent: *Friday, August 7, 2015 7:35:57 PM
>
>
> *Subject: *Re: [cfe-dev] [LLVMdev] Clang devirtualization proposal
>
> On Fri, Aug 7, 2015 at 5:21 PM Hal Finkel <hfinkel at anl.gov> wrote:
>
>
>> > From: "Chandler Carruth" <chandlerc at google.com>
>> > To: "Hal Finkel" <hfinkel at anl.gov>, "Sanjoy Das" <
>> sanjoy at playingwithpointers.com>
>> > Cc: "cfe-dev at cs.uiuc.edu Developers" <cfe-dev at cs.uiuc.edu>, "LLVM
>> Developers Mailing List" <llvmdev at cs.uiuc.edu>
>> > Sent: Friday, August 7, 2015 5:52:04 PM
>> > Subject: Re: [cfe-dev] [LLVMdev] Clang devirtualization proposal
>> >
>> > On Sat, Aug 1, 2015 at 6:39 AM Hal Finkel < hfinkel at anl.gov > wrote:
>> >
>> > ----- Original Message -----
>> > > From: "Sanjoy Das" < sanjoy at playingwithpointers.com >
>> > > To: "Reid Kleckner" < rnk at google.com >
>> > > Cc: "Piotr Padlewski" < prazek at google.com >, " cfe-dev at cs.uiuc.edu
>> > > Developers" < cfe-dev at cs.uiuc.edu >, "LLVM Developers
>> > > Mailing List" < llvmdev at cs.uiuc.edu >
>> > > Sent: Saturday, August 1, 2015 1:22:50 AM
>> > > Subject: Re: [LLVMdev] [cfe-dev] Clang devirtualization proposal
>> > >
>> > > On Fri, Jul 31, 2015 at 6:18 PM, Reid Kleckner < rnk at google.com >
>> > > wrote:
>> > > > Consider this pseudo-IR and some possible transforms that I would
>> > > > expect to
>> > > > be semantics preserving:
>> > > >
>> > > > void f(i32* readonly %a, i32* %b) {
>> > > > llvm.assume(%a == %b)
>> > > > store i32 42, i32* %b
>> > > > }
>> > > > ...
>> > > > %p = alloca i32
>> > > > store i32 13, i32* %p
>> > > > call f(i32* readonly %p, i32* %p)
>> > > > %r = load i32, i32* %p
>> > > >
>> > > > ; Propagate llvm.assume info
>> > > > void f(i32* readonly %a, i32* %b) {
>> > > > store i32 42, i32* %a
>> > > > }
>> > > > ...
>> > > > %p = alloca i32
>> > > > store i32 13, i32* %p
>> > > > call f(i32* readonly %p, i32* %p)
>> > > > %r = load i32, i32* %p
>> > >
>> > > I'd say this first transformation is incorrect. `readonly` is
>> > > effectively part of `%a`'s "type" as it constrains and affects the
>> > > operations you can do on `%a`. Even if `%b` is bitwise equivalent
>> > > to
>> > > `%a` at runtime, it is "type incompatible" to replace `%a` with
>> > > `%b`.
>> > >
>> > > This is similar to how you cannot replace `store i32 42, i32
>> > > addrspace(1)* %a` with `store i32 42, i32 addrspace(2)* %b`, even
>> > > if
>> > > you can prove `ptrtoint %a` == `ptrtoint %b` -- the nature of
>> > > `store`
>> > > is dependent on the type of the pointer you store through.
>> > >
>> > > The glitch in LLVM IR right now is that the `readonly`ness of `%a`
>> > > is
>> > > not modeled in the type system, when I think it should be. An `i32
>> > > readonly*` should be a different type from `i32*`. In practice this
>> > > may be non-trivial to get right (for instance `phi`s and `selects`
>> > > will either have to do a type merge, or we'd have to have explicit
>> > > type operators at the IR level).
>> >
>> > We could do this, but then we'd need to promote these things to
>> > first-class parts of the type system (and I'd need to put further
>> > thought about how this interacts with dynamically-true properties at
>> > callsites and inlining).
>> >
>> > The alternative way of looking at it, which is true today, is that
>> > @llvm.assume is not removed even when its information is 'used'. It
>> > appears, given this example, that this is actually required for
>> > correctness, and that dead-argument elimination needs to
>> > specifically not ignore effectively-ephemeral values/arguments.
>> >
>> > What follows is an off-the-cuff response. I'm still thinking through
>> > it, but wanted to let others do so as well.
>> >
>> >
>> > There is yet another interpretation that we might use: the final
>> > transformation Reid proposed is actually correct and allowed
>> > according to the IR.
>> >
>> >
>> > Specifically, we could make 'readonly' a property of the memory much
>> > like aliasing is. That would mean that the function promises not to
>> > modify the memory pointed to by %a in this example. The optimizer
>> > gets to ignore any such modifications while remaining correct.
>>
>> We could certainly do this, but it will obviously make inference harder.
>> That might not be a good thing.
>>
>
> The other approach that seems reasonable is to push this to the caller to
> make inference in the callee easier. In that scenario, you would say that
> the readonly tells the caller that the memory location represented by the
> argument is not written *through that argument* but might be written
> through some other argument. Since the caller passes two pointers which
> alias, it cannot forward the store.
>
>
> Isn't that what happens today?
>
>
>
> The problem I see is that if the transformation in the body of the callee
> does CSE of any form, it allows dead argument elimination to remove this
> non-readonly potentially-aliasing argument.
>
>
> Right, that seems to be the problem.
>
>
>
> So if we want to go this route, I think we need to at least block dead
> argument elimination from removing a potentially writable aliasing argument
> even if it is unused in the function body, because it might be modeling a
> writable way for a particular memory location to enter the function.
>
> All in all, I would prefer the stronger guarantee of the readonly
> attribute (that the memory location is unchanged, regardless of through
> which pointer it is accessed).
>
>
> Perhaps this is indeed the only strategy that is completely
> self-consistent. I don't object to pursuring this.
>
>
>
>
>> As I said earlier, the original problem highlighted by Reid's example
>> cannot currently occur: that could only happen if you remove the
>> @llvm.assume call (thus making the other argument unused). We already don't
>> do this (because the assumes could be useful if later inlined), and now we
>> have a second reason. Regardless, because we don't actively remove
>> @llvm.assume, I'm not convinced the current state of things is currently
>> broken.
>>
>
> As others have said, *anything* which triggers CSE seems problematic here.
>
>
> Fair enough. To record the example you provided to me on IRC here:
>
> chandlerc: we start with 'if (a == b) { *b = 42; } else { x(); }'
> chandlerc: then CSE to 'if (a == b) { *a = 42; } else { x(); }'
> chandlerc: then inline to 'if (a == b) { *a = 42; } else { unreachable; }'
> chandlerc: then fold to 'if (true) { *a = 42; }'
> chandlerc: b is now dead
> chandlerc: x was marked as 'readnone'
> chandlerc: and contained unreachable
>
>
>  -Hal
>
>
>
>>
>>  -Hal
>>
>> >
>> > This would, in turn, mean that the store in Reid's "@f" function is
>> > an unobservable in the case that %a == %b through some dynamic
>> > property, whatever that may be. And as a consequence, the
>> > store-forwarding is a correct transformation.
>> >
>> >
>> > -Chandler
>> >
>> >
>> >
>> >
>> >
>> > -Hal
>> >
>> > >
>> > > -- Sanjoy
>> > > _______________________________________________
>> > > LLVM Developers mailing list
>> > > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
>> > > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>> > >
>> >
>> > --
>> > Hal Finkel
>> > Assistant Computational Scientist
>> > Leadership Computing Facility
>> > Argonne National Laboratory
>> > _______________________________________________
>> > cfe-dev mailing list
>> > cfe-dev at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>> >
>>
>> --
>> Hal Finkel
>> Assistant Computational Scientist
>> Leadership Computing Facility
>> Argonne National Laboratory
>>
>
>
>
> --
> Hal Finkel
> Assistant Computational Scientist
> Leadership Computing Facility
> Argonne National Laboratory
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150808/0230499f/attachment.html>