dzaima wrote: Would there be some way to explicitly request to not clobber vl/vtype? An asm block is quite useful as an "optimization fence" on x86 & ARM NEON, but it'd be sad if that necessarily meant extra vsetvl spam on RVV (currently clang doesn't do much optimization on RVV intrinsics so there's not much chance for it to misoptimize, but presumably that may change; more significant for `__attribute__ vector_size` vectors, though using those with asm is non-trivial and generally not ideal on RVV; [some quite contrived examples](https://c.godbolt.org/z/jTe1WjTc9))
https://github.com/llvm/llvm-project/pull/128636 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
