david-horner / isa-bend

RISC-V Instruction Set Manual

This repository contains the LaTeX source for the draft RISC-V Instruction Set Manual. At the time of this writing, none of these specifications have been formally adopted by the RISC-V Foundation.

This work is licensed under a Creative Commons Attribution 4.0 International License. See the LICENSE file for details.

The Manual is split up into the following volumes:

• Volume I: User-Level ISA
• Volume II: Privileged Architecture

Official versions of the specifications are available at https://riscv.org/specifications/

Compiled versions of the most recent drafts of the specifications are available at https://github.com/riscv/riscv-isa-manual/releases/latest

The canonical list of open-source RISC-V implementations' marchid CSR values is available at https://github.com/riscv/riscv-isa-manual/blob/master/marchid.md

isa-bend's Issues

Relax constraint on vl exceed register group effective VLEN.

With the two significant changes of
a) V0.9 register structure moving from vertical to horizontal widening and
b) eliminating SLEN
element width structural dependencies are eliminated.

As a consequence maintaining SEW/LMUL ratio is also removed.

This allows software to address a register groups operated as two or more consecutive but separate register groups..

Currently a vsetvl[i] instruction would be needed to set a lower LMUL.
However, it would be advantageous to allow a lower than LMUL aligned register group to execute when vl > EMUL/EEW.
In this case only the first EMUL/EEW elements are processed.

This would be especially useful in transforming a fully populated LMUL=8 register group two LMUL=4 groups with the lower elements remaining unchanged and the upper being volatile. (more valid combinations are possible)

The drops a run time alignment check that is a worthwhile trade-off to provide the functionality.

I believe this is a viable post V1.0 relaxation.
So, I believe it should also be consciously assessed for V1.0.

This is the corollary to #523

table of valid combinations

LMUL	r	r+1	r+2	r+3	r+4	r+5	r+6	r+7
8	8	1	2	1	4	1	2	1
4	4	1	2	1
2	2	1

mixed size opperations in EDIV