https://github.com/llvm-beanz commented:
We talked about this yesterday in a call with @Keenuts and @s-perron. I think one of the problems we have with HLSL is that we haven't had explicit address space annotations except for `groupshared` in the past. We do need to do that in the future, but I'm not sure this was the right slicing. In HLSL `static` variables are stored in thread-local addresses, but have thread lifetime, whereas function local variables are stored in thread-local addresses but have function lifetime. In both cases the storage is thread-local, and they can be used interchangeably as addresses so we should have them in the same address space, but with different storage classes and lifetimes. There are cases (specifically around the implicit `this` object) where we'll need to follow the OpenCL approach and insert addrspacecasts that we can clean up later in the compiler. I don't think we have a good way around that until we can evolve the language to express address spaces more comprehensively. Those issues will need to be addressed in the language, and we have a few different related proposals: * https://github.com/microsoft/hlsl-specs/blob/main/proposals/0006-reference-types.md * https://github.com/microsoft/hlsl-specs/pull/325 I've also filed an issue to track specifying this further: https://github.com/microsoft/hlsl-specs/issues/364 https://github.com/llvm/llvm-project/pull/122103 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
