banach-space wrote: > I'd just like to point out that ElementCount does scale to multidimensional > vectors (you can simply have an array of them).
So, for example, a scalable array of scalable vectors? IIRC, that's not supported in LLVM IR, but CIR could take a different approach. There's nothing inherent in `ElementCount` that would prevent this, right? But then, should CIR follow that design (i.e. a vector of vectors)? I don't have a strong opinion on this at the moment. > I think targeting SME via CIR should still use the ACLE intrinsics Why ACLE rather than LLVM intrinsics? Does ACLE allow us to model something that would otherwise be lost at the LLVM IR level? I'm curious because experimenting with SME support in CIR is something we definitely intend to do sooner or later. https://github.com/llvm/llvm-project/pull/231906 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
