Greg Lueck
Greg Lueck
> ping https://github.com/orgs/intel/teams/dpcpp-specification-reviewers We just had a meeting about this yesterday, so I presume the next step is that @tiwaria1 will update this PR.
I'd like to take another look at this since it's been a while since my last review. I promise to do this by Thursday or Friday of this week.
@kbobrovs and @mdtoguchi: I have removed "Draft" status. Please approve or give comments. @intel/dpcpp-specification-reviewers: I will need an approval from one of you too.
@AerialMantis mentioning you here so you can see this proposal. Comments welcome!
@intel/llvm-gatekeepers can this be merged?
> This might be a good idea if USM pointers can be nested inside classes in user code. I don't know if this is possible with USM pointers though. Yes,...
> * Try to pass structs with global pointer inside, but convert them to structs with generic pointers inside to avoid further problems with IR emission (that will be non-trivial...
> At least for now I don't see other ways to preserve performance advantage and not decompose structs with pointers. All you're doing here is building in an assumption to...
> Actually, it is possible to get a pointer with local address space in kernel arguments using local accessor. So, I'm not sure BE can make this assumption. This is...
> It's much easier to do in CFE where SPIR kernel function wrapper is created. I think maintaining address space adjustment in LLVM IR -> LLVM IR transformation or in...