[llvm-dev] How to support different versions of RVV ?

ALO via llvm-dev llvm-dev at lists.llvm.org
Wed Jul 7 00:31:48 PDT 2021


— Jojo
在 2021年7月5日 +0800 PM8:23,Alex Bradbury via llvm-dev <llvm-dev at lists.llvm.org>,写道:
> On Mon, 28 Jun 2021 at 03:04, 陆旭凡 via llvm-dev <llvm-dev at lists.llvm.org> wrote:
> >
> > Hello everyone. RVV 0.10 is now supported on the upstream LLVM RISC-V backend. However, because some RISC-V-based chip manufacturers chose RVV 0.71 version as the vector extension at the beginning, and a large number of chips that supported RVV 0.71 version are taped out. So can the community explore a way or framework to support different versions of RVV?
>
> Hi - please see here
> <https://lists.llvm.org/pipermail/llvm-dev/2020-January/138364.html>
> for a discussion on the standard policy for accepting not-yet-ratified
> extensions. In the ideal case we wouldn't need to support pre-ratified
> versions of the specification (as you've noted, the V support is
> currently behind an experimental flag and we track the latest version
> of the spec). However, I recognise there are cases where we may need
> to be more flexible. The question comes down to how many potential
> users there would be for RVV 0.71 support,
As far as I know, some computing libs like OpenCV [1], OpenBLAS [2], and ComputerLibrary
have implement the RVV 0.7.1. Also some HW core/boards with RVV 0.7.1 like D1 [3], RVB-ICE
have been release from allwinnertech, T-HEAD :)

[1] https://github.com/opencv/opencv/wiki/OpenCV-RISC-V
[2] https://github.com/xianyi/OpenBLAS
[3] https://d1.docs.allwinnertech.com
> who would be committing
> resources to maintaining, and a consideration of any other costs of
> trying to support multiple incompatible RVV versions simultaneously.
I think the vendor who implements the RVV 0.7.1 should to support/fix
multiple incompatible RVV versions simultaneously.
It’s benefit also for all vendors if community encourages to cover
usefull specs :)
> Do you have any more information on the chips targeting RVV 0.71?
>
> Best,
>
> Alex
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210707/87dfdebd/attachment.html>


More information about the llvm-dev mailing list