[PATCH] D16749: [OpenMP] Map clause codegeneration.

[Alexey Bataev via cfe-commits]

Samuel, this is true for device part of codegen. But what about host? If 
this code must be executed on host, will it be handled correctly?

Best regards,
Alexey Bataev
=============
Software Engineer
Intel Compiler Team

18.03.2016 23:03, Samuel Antao пишет:
> sfantao added a comment.
>
> Hi Alexey,
>
> I am sorry but I don't think I am following. So in your example:
>
>    #pragma omp target map(to: this->A)
>    []()->void {++A}();
>
> the map clause is going to make sure the runtime library allocates the section of `this` that contains A and that is what is used in the device. There are no new instances of A created inside the kernel. If you do:
>
>    #pragma omp target
>    []()->void {++A}();
>
> what the lambda is going to get would be the same (except that runtime had eventually to copy more data to the device), so the map clause has no effect in how captures are dealt with inside. All the map clause does is to inform the runtime about sizes and addresses and change the captures to byCopy when necessary, it does not create new instances of data inside the kernel. Maybe what you are proposing is meant for (first)private?
>
> Thanks,
> Samuel
>
>
> http://reviews.llvm.org/D16749
>
>
>