c

On the topic of byte order, have you had a look at the proposed riscv bit manipulation instructions? The grev and gzip instructions in particular are very simple and elegant, and may replace a number of more specialized instructions. The first is a generalized reverse, which is capable of both bit reversal and byte swapping. For every bit in the control word, the corresponding pairs of 2^n bits are swapped and the effects are combined. The second is a generalized zip which is more opaque without a diagram, but can zip/unzip any 2^n bit sizes. Both would be a welcome addition to any instruction set.

For detail, see xbitmanip-draft.pdf at https://github.com/cliffordwolf/xbitmanip

It looks like the Mill has a bitwise reverse instruction, and is using shuffle for byte swapping, which limits the functionality to bits and bytes. The shuffle itself appears more limited then some would like, and Grant also mentioned a shuffle that can use two element sources; something which is actually implemented on PPC with vec_perm, and which is extremely useful on that platform.

With that said, I appreciate that the Mill is different enough that it will have its own considerations. I’m still curious if any thought has been given to vec_perm or something similar, and how it might fit into the Mill.