llvm dev - Jul 2019 - [RFC] Changing X86 data layout for address spaces

Looks like I made my point in an accidentally really confusing way.  Let 
me try again w/Matt's correction in mind.

I want to make sure that the middle end optimizer code is *driven by the 
data layout*.  I am not trying to express an opinion on how that data 
layout is populated (frontend, backedge, black magic, what have you).  I 
just want to make sure that we don't end with the middle end having to 
know that address space "56" has special meaning beyond what is
encoded
in the data layout.  To say it differently, I want to make sure that a 
different frontend targeting a different backend is not effected by the 
proposed changes.

Philip

On 7/31/19 9:26 AM, Arsenault, Matthew wrote:
> The datalayout itself is currently not considered a point of 
> configurability. It’s a static, backend/codegen defined property. The 
> target machine machine returns the datalayout for the given triple and 
> is the source of truth. The front end is responsible for creating a 
> module with an exactly matching datalayout and is not free to 
> customize the datalayout this way. I’m not sure what there is to gain 
> by avoiding putting this information in the backend.
>
> -Matt
>
> ------------------------------------------------------------------------
> *From:* llvm-dev <llvm-dev-bounces at lists.llvm.org> on behalf of
Philip
> Reames via llvm-dev <llvm-dev at lists.llvm.org>
> *Sent:* Wednesday, July 31, 2019 12:15
> *To:* Amy Huang; John Reagan
> *Cc:* llvm-dev
> *Subject:* Re: [llvm-dev] [RFC] Changing X86 data layout for address 
> spaces
>
> Please review the properties of an address space which are 
> configurable via the data layout.  For example, bitwidth is one of 
> those parameters.  If that parameter space covers your needs, then you 
> do not need LLVM side support.
>
> Philip
>
> On 7/30/19 2:59 PM, Amy Huang wrote:
>> Thanks for the info--
>> It seems like the way to do this is for clang to use address spaces 
>> to represent different sized pointers in the IR. If that's the
case,
>> then LLVM still has to know about those specific address spaces. I 
>> don't see a clear way to avoid adding knowledge of the address
spaces
>> to LLVM.
>> I think that's the approach we'll go with, unless there are
more
>> objections.
>>
>> On Sun, Jul 28, 2019 at 8:23 AM John Reagan 
>> <john.reagan at vmssoftware.com <mailto:john.reagan at
vmssoftware.com>> wrote:
>>
>>     That is basically what we do today to provide mixed sized
>>     pointers with
>>     our legacy frontends.  They generate IR to our old code generator
>>     which
>>     has ADDR32 and ADDR64 datatypes.  We use a 64-bit address data
layout
>>     and then typecast the 32-bit forms to/from the underlying 64-bit
>>     addresses.  I have been warned that such rampant typecasting might
>>     interfere with certain optimizations or TBAA data.  We haven't
>>     investigated that yet.  Since clang doesn't go through our
IR-to-IR
>>     converter, we'll have to teach clang to do the same.
>>
>>     On 7/26/2019 6:35 PM, Amy Huang wrote:
>>     >       *Clang* can assign whatever meaning to an AS it wishes,
>>     and their
>>     >     properties are configurable via data layout configuration.
>>     >
>>     >
>>     > I think Clang checks that its data layout matches the LLVM
>>     target data
>>     > layout, so I'm not sure how to go about implementing the
>>     address spaces
>>     > as a change in Clang.
>>     > Thoughts?
>>     >
>>     > The other issue I was running into is how to do the address
>>     space casts
>>     > in clang - my current thought was to lower the address space
>>     casts in
>>     > LLVM to the sign extension / zero extension / truncation.
>>     >
>>     > Thanks,
>>     > Amy
>>     >
>>     > On Mon, Jul 22, 2019 at 5:06 PM Philip Reames via llvm-dev
>>     > <llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>
>>     <mailto:llvm-dev at lists.llvm.org
>>     <mailto:llvm-dev at lists.llvm.org>>> wrote:
>>     >
>>     >     Please don't bake in knowledge *in LLVM* of a address
space
>>     unless
>>     >     there's a strong need.  *Clang* can assign whatever
meaning
>>     to an AS
>>     >     it wishes, and their properties are configurable via data
>>     layout
>>     >     configuration.
>>     >
>>     >     The standards of review are much lower for a Clang only
>>     change as it
>>     >     can be revised arbitrarily without breaking other uses of
>>     LLVM.
>>     >     Building in first class knowledge to LLVM itself is a
breaking
>>     >     change (potentially) for other frontends.
>>     >
>>     >     Philip
>>     >
>>     >     On 7/22/19 1:19 PM, Reid Kleckner via llvm-dev wrote:
>>     >>     Yes, the address spaces are intended to be generally
>>     useful for
>>     >>     other applications. Glad to hear that you would find
it
>>     useful. :)
>>     >>
>>     >>     For now we were only going to add them for x86, but it
>>     should be
>>     >>     easy to add such address spaces to other targets.
>>     >>
>>     >>     On Mon, Jul 22, 2019 at 1:13 PM John Reagan via
llvm-dev
>>     >>     <llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>
>>     <mailto:llvm-dev at lists.llvm.org
>>     <mailto:llvm-dev at lists.llvm.org>>> wrote:
>>     >>
>>     >>         Amy, when you say "implement MSVC's
extensions", is that
>>     >>         MSVC/LLVM or
>>     >>         are you going to add these to clang? OpenVMS has
>>     dual-sized
>>     >>         pointers as
>>     >>         well and want to see them in clang. One of my
engineers
>>     >>         noticed that
>>     >>         address space 4 was a 32-bit pointer address space
(or at
>>     >>         least it used
>>     >>         to be).  I haven't seen any formal discussion
of that
>>     over in
>>     >>         the cfe
>>     >>         dev list.  I'm happy to help out where we can.
>>     >>
>>     >>         John
>>     >>
>>     >>  _______________________________________________
>>     >>         LLVM Developers mailing list
>>     >> llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>
>>     <mailto:llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>>
>>     >> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>     >>
>>     >>
>>     >>  _______________________________________________
>>     >>     LLVM Developers mailing list
>>     >> llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>
>>     <mailto:llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>>
>>     >> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>     >     _______________________________________________
>>     >     LLVM Developers mailing list
>>     > llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>
>>     <mailto:llvm-dev at lists.llvm.org <mailto:llvm-dev at
lists.llvm.org>>
>>     > https://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/20190731/3b5a5b9a/attachment.html>

On Wed, Jul 31, 2019 at 10:29 AM Philip Reames via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Looks like I made my point in an accidentally really confusing way.  Let
> me try again w/Matt's correction in mind.
>
> I want to make sure that the middle end optimizer code is *driven by the
> data layout*.  I am not trying to express an opinion on how that data
> layout is populated (frontend, backedge, black magic, what have you).  I
> just want to make sure that we don't end with the middle end having to
know
> that address space "56" has special meaning beyond what is
encoded in the
> data layout.  To say it differently, I want to make sure that a different
> frontend targeting a different backend is not effected by the proposed
> changes.
>
Got it. :) Yes, we have no intention of teaching the middle end about these
address spaces. The usual rules around addrspacecast should apply.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190731/255eaa4c/attachment.html>

I really hate to dip my toe in here, because it will only reveal my total
ignorance, but….

Do the "usual rules around addrspacecast" say when different address
spaces can alias?  I remember somebody using address spaces to represent
something like special off-to-the-side device memory, which obviously could
never alias main memory, whereas other uses like the 32/64-bit thing will
certainly have the 32-bit space aliasing (likely disjoint parts of) the 64-bit
space, and the exact mapping of 32-to-64 might vary across OSes.
--paulr

From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Reid
Kleckner via llvm-dev
Sent: Wednesday, July 31, 2019 1:44 PM
To: Philip Reames
Cc: llvm-dev; John Reagan
Subject: Re: [llvm-dev] [RFC] Changing X86 data layout for address spaces

On Wed, Jul 31, 2019 at 10:29 AM Philip Reames via llvm-dev <llvm-dev at
lists.llvm.org<mailto:llvm-dev at lists.llvm.org>> wrote:

Looks like I made my point in an accidentally really confusing way.  Let me try
again w/Matt's correction in mind.

I want to make sure that the middle end optimizer code is *driven by the data
layout*.  I am not trying to express an opinion on how that data layout is
populated (frontend, backedge, black magic, what have you).  I just want to make
sure that we don't end with the middle end having to know that address space
"56" has special meaning beyond what is encoded in the data layout. 
To say it differently, I want to make sure that a different frontend targeting a
different backend is not effected by the proposed changes.
Got it. :) Yes, we have no intention of teaching the middle end about these
address spaces. The usual rules around addrspacecast should apply.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190731/8fcc564f/attachment.html>

Sorry, this thread disappeared into my email filters.

So to clarify/summarize, the proposed change here is to change the x86
backend data layout to encode the pointer sizes for these three address
spaces. It seems like there's general agreement that this is fine, as long
as there aren't other changes to the 'middle end'--

On Wed, Jul 31, 2019 at 10:44 AM Reid Kleckner <rnk at google.com> wrote:
> On Wed, Jul 31, 2019 at 10:29 AM Philip Reames via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> Looks like I made my point in an accidentally really confusing way. 
Let
>> me try again w/Matt's correction in mind.
>>
>> I want to make sure that the middle end optimizer code is *driven by
the
>> data layout*.  I am not trying to express an opinion on how that data
>> layout is populated (frontend, backedge, black magic, what have you). 
I
>> just want to make sure that we don't end with the middle end having
to know
>> that address space "56" has special meaning beyond what is
encoded in the
>> data layout.  To say it differently, I want to make sure that a
different
>> frontend targeting a different backend is not effected by the proposed
>> changes.
>>
> Got it. :) Yes, we have no intention of teaching the middle end about
> these address spaces. The usual rules around addrspacecast should apply.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190808/cac26a9a/attachment.html>