[PATCH] D141672: [RISCV] Support vector crypto extension ISA string and assembly
Craig Topper via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Thu Feb 2 14:29:51 PST 2023
craig.topper added a comment.
In D141672#4100940 <https://reviews.llvm.org/D141672#4100940>, @reames wrote:
> This was brought up in today's RISCV sync call; I want to summarize the major points of discussion.
>
> This patch is now behind the current most recent revision on the vector-crypto spec. (https://github.com/riscv/riscv-crypto/releases/download/v20230125/riscv-crypto-spec-vector.pdf) The assumption is that the patch needs to be updated to match most recent. If there's a reason why this assumption is wrong, please explicitly describe the argument.
>
> There was confusion around the versioning scheme used in the vector-crypto spec. It looks like newer version have moved to an internally consistent (and less than 1.0) version scheme, so I think this item is resolved.
>
> There was a mention of another change being made to the spec in the near future and a desire to wait until that had happened. This isn't required by our experimental extension policy. However, in generally trying to minimize churn by delaying a couple days is not unreasonable. I'd welcome comment from someone following spec changes as to whether we should delay here or not, and why.
Here are some key points from today's crypto meeting that someone in attendance shared.
- vghmac.vv
- a new variant vghacm.vv will be added (xor then mul)
- vandn operands will be reversed to match other insns in RISC-V
- new constraint for vl and element group operations: if VLMAX < EGS (equivalently VLEN * LMUL < EGW): raise an illegal exception (on top of vl % EGS != 0)
- Will define two meta extensions Zvkn and Zvks for the RVA23 profile
- Zvkn = Zvkns + Zvknhb (AES, SHA-256, SHA-512)
- Zvks = Zvksed + Zvksh (SM3, SM4)
- Will ask to add Zvkg + Zvkb as optional for RVA23 profile
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D141672/new/
https://reviews.llvm.org/D141672
More information about the cfe-commits
mailing list