From David Weaver: What is the proper (canonical) way to include the U extension, when listing the ISA of an implementation?

[rpsene]

Section 27.11 of the RV Unpriv spec clearly specifies the canonical order of the ISA extensions. It even includes the “S” and “H” extensions. …However, oddly, it does not list the “U” extension (User mode).

What is the proper (canonical) way to include the U extension, when listing the ISA of an implementation? It seems that the ISA string for an RV32GC implementation with U (RV32GCU?) should be different from one that does not include U (RV32GC).

I couldn’t find that information in either the Unpriv or Priv specs, and don’t recall seeing it elsewhere.
Any pointers where to find it would be much appreciated!

[rpsene]

Reply from @aswaterman:

The naming chapter in the unpriv spec doesn't actually define "S" and "H" extensions; rather, it defines S and H as prefixes for multi-letter extension names that pertain to privileged-architecture features.

I don't believe there is a canonical way to indicate user-mode support within an ISA string. I've seen vendors indicate privileged-architecture support separate from ISA string, e.g., "an RV32GC implementation with M and U privilege modes".

The riscv-isa-manual github repo issue tracker is a good place for these questions, BTW. It spreads the load of answering them (and might result in better answers than mine).