Thank you for clarifying. It still seems wrong, somehow. Admittedly I don't know enough detail about Thumb and the others, but if the variable-length encoding i

Thank you for clarifying. It still seems wrong, somehow. Admittedly I don't know enough detail about Thumb and the others, but if the variable-length encoding is a simple design like RISC-V's, it's hard to imagine that extra layer to be so expensive as to eliminate the benefits of an effectively increased I$. It would be nice to be able to dig deeper into the details of this.