We still have blow for integer part vx combine. It is totally OK from my side to defer this to next stage 1 for risk consideration, as well as there is a clear line. Policy is policy I think.
- vmsltu.vx - vmslt.vx - vmsleu.vx - vmsle.vx - vmsgtu.vx - vmsgt.vx Pan -----Original Message----- From: Jeff Law <[email protected]> Sent: Monday, November 24, 2025 9:19 PM To: Robin Dapp <[email protected]>; Li, Pan2 <[email protected]>; [email protected] Cc: [email protected]; [email protected]; Chen, Ken <[email protected]>; Liu, Hongtao <[email protected]> Subject: Re: [PATCH v1 0/2] RISC-V: Combine vec_duplicate + vmsltu.vv to vmsltu.vx on GR2VR cost On 11/24/25 3:55 AM, Robin Dapp wrote: > I don't have issues with going forward here but not that we're past stage 1 > and > this is not really a regression. The changes in all of these are very > mechanical so the risk is low (and I would be OK with it) but I'd still like > some consensus. > > Kito, Jeff, any opinion? We can also discuss tomorrow in the sync call. > > Pan, do you know how many of those are left or in your plan still? I'll only be around for ~5 minutes tomorrow as I have a personal item to attend to. So I probably can't really lead or participate in that discussion. I agree the risk is low, but it's also the case that we need to draw the line somewhere and Richi is communicating the deadlines well in advance so that folks aren't surprised. I'd be OK with letting it through given its risk profile, but we do need to wrap up the stage1 efforts. Jeff
