[llvm-dev] [RFC] Refactor class hierarchy of VectorType in the IR

[John McCall via llvm-dev]

On 22 May 2020, at 12:42, Sander De Smalen wrote:
> Hi John,
>
> Not sure if I've missed this, but is there a reason that Chris' 
> suggestion to do a mechanical search/replace of VectorType -> 
> FixedVectorType wouldn't work for your local changes? Given that your 
> downstream changes interpret VectorType as a fixed vector type, it 
> should be safe to do that. It will also be an effort that you'll only 
> need to do once when you next merge with upstream master, after that 
> you can just use FixedVectorType in your downstream repo.
>
> Chris has been doing some great work to split out fixed-width vectors 
> from scalable vectors which is an important step towards preparing 
> LLVM IR for scalable vectors. The bulk of mechanical changes have been 
> done and a bunch of us are now working to ensure LLVM IR 
> handles/distinguishes these vector types correctly. It would be a pity 
> and great setback if that would be a wasted effort.

I’m not asking for that to be a wasted effort at all.  I’m asking 
for fairly mechanical changes to some of your mechanical changes.

> We have a fortnightly SVE/SVE2 sync-up call with people in the LLVM 
> community interested in scalable vectors where we normally discuss 
> these things. If you want to join the call next Thursday [1], it seems 
> like a fair topic to bring up in that meeting?

If you’re willing to hold off that long, sure.

John.