[clang] [llvm] [RISCV] Make zvknha a subset of zvknhb (PR #178680)

[Nicolas Brunie via llvm-commits]

nibrunieAtSi5 wrote:

> We don't usually do this unless there is a sentence in the spec indicating one extension implies or depends on another. I don't think such a sentence exists.
> 

I believe there is not such sentence in the spec but this is an oversight. Zvknhb is clearly a superset containing Zvknha, and the fact that the vector crypto spec defines them independently is not ideal IMO.

>  In hardware, would you have two different circuits implementing SHA256 if you support zvkhna and zvknhb?

Definitely if you support both, it would be sub-optimal to duplicate that hardware.


https://github.com/llvm/llvm-project/pull/178680