[PATCH] D134445: [PR57881][OpenCL] Fix incorrect diagnostics with templated types in kernel arguments

[Anastasia Stulova via Phabricator via cfe-commits]

Anastasia added a comment.

In D134445#3920448 <https://reviews.llvm.org/D134445#3920448>, @rjmccall wrote:

> In D134445#3920257 <https://reviews.llvm.org/D134445#3920257>, @Anastasia wrote:
>
>> In D134445#3920188 <https://reviews.llvm.org/D134445#3920188>, @rjmccall wrote:
>>
>>> You really can't ask whether a class template pattern is standard layout; it's not meaningful.
>>
>> Well the templates have to be instantiated with concrete types as kernels can't be called from the host code. But if the concrete type is a reference or pointer there is no full instantiation apparently as AST dumps in the review description show.
>
> My objection here is to the code that does `CXXRec = CXXRec->getTemplateInstantiationPattern()`.  That is the unsubstituted template pattern.  You then ask if it has standard layout.  I do not think this is the right thing to do.

Yes this only works if the type doesn't depend on the template parameter like in the reported test case. But it is not a universal solution.

>>> How are pointers and references passed to kernels?  Does the pointee get copied or something?  If so, you may have a requirement that pointee types be complete, in which case the only problem is probably that you're doing this check too soon, or doing them on declarations rather than on definition/use.
>>
>> Ok, the kernel function actually already contains the instantiated templates since they can't be templated themselves.
>
> Yes, I understand that.
>
>> My understanding is that references and pointers to templated types are not required to be fully instantiated as they are not ODR-used?
>
> It's not the ODR, but yes, in standard C++, declaring a parameter of pointer type does not require the pointee type to be complete, and if the pointee type is a template type, the compiler is not permitted to immediately try to instantiate it.  Of course, OpenCL `kernel` declarations are not standard C++ and can use different rules, and I think the fact that OpenCL has this semantic restriction on pointer arguments to kernels is pretty good license to do so.  There are some places in C++ (e.g. initialization) that do instantiate templates if a definition is available, and you can request this by calling `Sema::isCompleteType` instead of `hasDefinition()`.

Ok, thanks! I will look into this.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134445/new/

https://reviews.llvm.org/D134445