The original AArch64 implementation of the ML-DSA intrinsic is not clearly structured and does not have a clearly visible mapping of input, intermediate and output data to registers. A reworking to address both these issues through the use of vector register sequence type should simplify life for maintainers.