[cfe-dev] [RFC] Delayed target-specific diagnostic when compiling for the devices.

Finkel, Hal J. via cfe-dev cfe-dev at lists.llvm.org
Mon Feb 25 16:25:11 PST 2019 

On 2/25/19 12:38 PM, Alexey Bataev wrote:

Hi Hal, I made a simple reproducer, seems to me float128 is already emitted correctly for NVPTX target. I just need to check the alignment and the record layout. And tere is another one question because of this. What if there is a difference between the host and the device in the memory layout for different types? Shall we use the host memory layout when we try to compile for the device? But it may lead to the problems with the memory access on the devices. Maybe, we should emit an error message if there is a difference in the type layout on the host and on the device?


I think that you need to use the host layout, otherwise there is no reasonable way to directly share the data structures. An error makes sense if there's a difference and we actually need to generate device code for operations on those elements.

 -Hal


-------------
Best regards,
Alexey Bataev

25.02.2019 13:33, Finkel, Hal J. пишет:

Hi, Alexey,

Thanks again for continuing to work on this. Yes, this is what we had in mind. Regarding how to do it, I *think* that you need to emit it as a i128, essentially, however, it occurs to me that it might not be that simple. If the alignment of the float128 is the same as i128, then I think that a single i128 will do. If the alignment of float128 is less than that of i128, then you need to emit a series of smaller fields to match the layout based on the alignment of the unsupported type.

 -Hal

On 2/25/19 12:10 PM, Alexey Bataev wrote:

I'm not saying that this is "terribly different", I'm just asking how to represent unsupported types. But seems to me I was wrong about codegen. The data types are emitted correctly, but we cannot use it for any expression.

-------------
Best regards,
Alexey Bataev

25.02.2019 12:53, Justin Lebar пишет:
> Hi Justin, no, the idea was different. Hal and David want to be able to compile the unsupported types, but emit errors for expressions with the unsupported types. It is required for unified memory support. I.e., if the type is used in, say, a structure or class, this class must be compiled correctly, even if the type is not supported. Hal, David, is this your intention?

To me that idea is not terribly different.

It seems to me all we're saying is, in addition to the rule of "emit an error if required to codegen something we don't support", we add the rule of "learn to codegen structs that contain unsupported types".  This should be relatively simple so long as one never *uses* the type, right?

On Mon, Feb 25, 2019 at 9:27 AM Alexey Bataev <a.bataev at outlook.com<mailto:a.bataev at outlook.com>> wrote:

Hi Justin, no, the idea was different. Hal and David want to be able to compile the unsupported types, but emit errors for expressions with the unsupported types. It is required for unified memory support. I.e., if the type is used in, say, a structure or class, this class must be compiled correctly, even if the type is not supported. Hal, David, is this your intention?

-------------
Best regards,
Alexey Bataev

25.02.2019 12:24, Justin Lebar пишет:
> I can try to disable emission of the error messages about unsupported type for NVPTX devices, but how we're going to emit it in the PTX format?

I thought the idea in this thread was: if we have to codegen something the backed doesn't support, then we emit an error.

That is, it's not "no errors for unsupported types," it's "deferred errors for unsupported types."

On Mon, Feb 25, 2019, 7:40 AM Alexey Bataev <a.bataev at outlook.com<mailto:a.bataev at outlook.com>> wrote:

Hi Hal, David, I have a question about the unsupported types. Ok, I can try to disable emission of the error messages about unsupported type for NVPTX devices, but how we're going to emit it in the PTX format? PTX supports only f16, f32 and f64 type. If we going to enable float128 type, for example, there is no way to emit it for NVPTX correctly. Any ideas how to do this? Because currently, I think, it will just lead to the incorrect codegen and will cause a crash in the NVPTX backend.

-------------
Best regards,
Alexey Bataev

17.01.2019 12:40, Finkel, Hal J. пишет:

On 1/17/19 11:11 AM, Alexey Bataev wrote:


The compiler does not know anything about the layout on the host when
it compiles for the device.



No, the compiler does know about the host layout (e.g., can't we
construct this by calling getContext().getAuxTargetInfo(), or similar?).




We cannot do anything with the types that are not supported by the
target device and we cannot use the layout from the host. And it is
user responsibility to write and use the code that is compatible with
the the target devices.

He/she does not need to use macros/void* types, there are templates.



No. This doesn't solve the problem (because you still need to share the
instantiations between the devices). Also, even if it did, does not
address the legacy-code problem that the feature is intended to address.
The user already has classes and data on the host and wishes to access
*parts* of that data on the device. We should make as much of that work
as possible.




You cannot use classes, which use types incompatible with the device.
There is a problem with the data layout on the device and we just
don't know how to represent such classes on the device.



There's no reason for this to be true. To be clear, the model of a
shared address space only makes sense, from a user perspective, if the
data layout is the same between the host and the target. Not mostly
similar, but the same. Otherwise, users will constantly be tracking down
subtle data-layout incompatibilities.

Thanks again,

Hal




-------------
Best regards,
Alexey Bataev
17.01.2019 11:47, Finkel, Hal J. пишет:


On 1/17/19 9:52 AM, Alexey Bataev wrote:


Because the type is not compatible with the target device.


But it's not that simple. The situation is that the programming
environment supports the type, but *operations* on that type are not
supported in certain contexts (e.g., when compiled for a certain
device). As you point out, we already need to move in this explicit
direction by, for example, allowing typedefs for types that are not
supported in all contexts, function declarations, and so on. In the end,
we should allow our users to design their classes and abstractions using
good software-engineering practice without worrying about access-context
partitioning.

Also, the other problem here is that the function I used as an example
is a very common C++ idiom. There are a lot of classes with function
that return a reference to themselves. Classes can have lots of data
members, and those members might not be accessed on the device (even if
the class itself might be accessed on the device). We're moving to a
world in which unified memory is common - the promise of this technology
is that configuration data and complex data structures, which might be
occasionally accessed (but for which explicitly managing data movement
is not performance relevant) are handled transparently. If use of these
data structures is transitively poisoned by use of any type not
supported on the device (including by pointers to types that use those
types), then we'll force unhelpful and technically-unnecessary
refactoring, thus reducing the value of the feature.

In the current implementation we pre-process the source twice, and so we
can:

 1. Use ifdefs to change the data memebers when compiling for different
targets. This is hard to get right because, in order to keep the data
layout otherwise the same, the user needs to understand the layout rules
in order to put something in the structure that is supported on the
target and keeps the layout the same (this is very error prone). Also,
if we move to a single-preprocessing-stage model, this no longer works.

 2. Replace all pointers to relevant types with void*, or similar, and
use a lot of casts. This is also bad.

We shouldn't be forcing users to play these games. The compiler knows
the layout on the host and it can use it on the target. The fact that
some operations on some types might not be supported on the target is
not relevant to handling pointers/references to containing types.

Thanks again,

Hal




-------------
Best regards,
Alexey Bataev

17.01.2019 10:50, Finkel, Hal J. пишет:


On 1/17/19 9:27 AM, Alexey Bataev wrote:


It should be compilable for the device only iff function foo is not used
on the device.


Says whom? I disagree. This function should work on the device. Why
should it not?

 -Hal




-------------
Best regards,
Alexey Bataev

17.01.2019 10:24, Finkel, Hal J. пишет:


On 1/17/19 4:05 AM, Alexey Bataev wrote:


Best regards,
Alexey Bataev



17 янв. 2019 г., в 0:46, Finkel, Hal J. <hfinkel at anl.gov><mailto:hfinkel at anl.gov> написал(а):




On 1/16/19 8:45 AM, Alexey Bataev wrote:

Yes, I thought about this. But we need to delay the diagnostic until
the Codegen phase. What I need is the way to associate the diagnostic
with the function so that this diagnostic is available in CodeGen.

Also, we need to postpone the diagnotics not only for functions,
but,for example, for some types. For example, __float128 type is not
supported by CUDA. We can get error messages when we ran into
something like `typedef __float128 SomeOtherType` (say, in some system
header files) and get the error diagnostic when we compile for the
device. Though, actually, this type is not used in the device code,
the diagnostic is still emitted and we need to delay too and emit it
only iff the type is used in the device code.



This should be fixed for CUDA too, right?

Also, we still get to have pointers to aggregates containing those types
on the device, right?



No, why? This is not allowed and should be diagnosed too. If somebody tries somehow to use not allowed type for the device variables/functions - it should be diagnosed.


Because this should be allowed. If I have:

struct X {
  int a;
  __float128 b;
};

and we have some function which does this:

X *foo(X *x) {
  return x;
}

We'll certainly want this function to compile for all targets, even if
there's no __float128 support on some accelerator. The whole model only
really makes sense if the accelerator shares the aggregate-layout rules
of the host, and this is a needless hassle for users if this causes an
error (especially in a unified-memory environment where configuration
data structures, etc. are shared between devices).

Thanks again,

Hal




Thanks again,

Hal




-------------
Best regards,
Alexey Bataev
15.01.2019 17:33, John McCall пишет:


On 15 Jan 2019, at 17:20, Alexey Bataev wrote:
This is not only for asm, we need to delay all target-specific
diagnostics.
I'm not saying that we need to move the host diagnostic, only the
diagnostic for the device compilation.
As for Cuda, it is a little but different. In Cuda the programmer
must explicitly mark the device functions,  while in OpenMP it must
be done implicitly. Thus, we cannot reuse the solution used for Cuda.


All it means is that you can't just use the solution used for CUDA
"off the shelf".  The basic idea of associating diagnostics with the
current function and then emitting those diagnostics later when you
realize that you have to emit that function is still completely
applicable.

John.


--
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory



--
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory

--
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20190226/9bbaf19f/attachment.html>

More information about the cfe-dev mailing list