================
@@ -7532,6 +7532,25 @@ A managed variable can be accessed in both device and
host code.
}];
}
+def CUDAClusterDimsAttrDoc : Documentation {
+ let Category = DocCatDecl;
+ let Content = [{
+In CUDA/HIP programming, the ``__cluster_dims__`` attribute can be applied to
a kernel function
----------------
erichkeane wrote:
Once they are spelled like that, they are no longer an attribute though, they
are a function macro. So referring to them as attribute in this with that
spelling isn't quite right.
I've a number of complaints with the documentation here, but I guess I can live
with settling back on 'consistently wrong'.
IMO, what we REALLY need to do, and probably SHOULD before we do any further
CUDA work, is to have a full 'Clang CUDA' documentation page written. We are
such a different enough dialect from the NVIDIA CUDA that I'm not comfortable
with us relying on THAT documentation for much of anything.
As far as the above: A more-correct way would be to document it with the
`__attribute__` spelling, then reference the macro as a "by the way, the
library exposes it as a function-like macro __cluster_dims__ <and so forth>".
https://github.com/llvm/llvm-project/pull/156686
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
