[PATCH] D12241: [CUDA] Change initializer for CUDA device code based on CUDA documentation.

[Artem Belevich via llvm-commits]

tra added a comment.

In http://reviews.llvm.org/D12241#236830, @tra wrote:

> Following code compiles with nvcc, but errors out with clang and this patch due to InitExpr check in CodeGenModule.cpp. It looks like the check needs to be more selective.
>
>   struct c {
>     c() {}
>   };
>  
>   __shared__ c var;
>  
>


Looks like we don't even need explicit constructor to fail:

  struct c { };
  __shared__ c var;

I believe it's something that we *do* want to support.


http://reviews.llvm.org/D12241