Agenda for sync meeting 1/29/21

[tlively]

The next meeting will be Friday, January 29 at 9:00AM - 10:00AM PST/ 6:00PM - 7:00PM CET. Please respond with agenda items you would like to discuss.

If this meeting doesn't already appear on your calendar, or you are a new attendee, please fill out this form to attend.

Agenda so far:

• Unsigned i64x2 comparisons (i64x2.gt_u, i64x2.lt_u, i64x2.ge_u, and i64x2.le_u instructions #414)
• i64x2.abs (i64x2.abs instruction #413)
• i64x2 min and max (i64x2.min_s and i64x2.max_s instructions #417, i64x2.min_u and i64x2.max_u instructions #418)
• Sign select (Sign Select instructions #124)
• Population Count (Population count instruction #379)
• Extended Pairwise Addition (Extended pairwise addition instructions #380)
• Prefetch (Prefetch instructions #352)
• f64x2 conversions (Double-precision conversion instructions #383)

Please comment with anything I've missed. @dtig will be conducting a provisional consensus vote on each of these PRs via GitHub comments. Please vote on each pending instruction before the meeting so that we can move things along as quickly as possible.

[Maratyszcza]

I expect that we'll have experimental data on double-precision conversions by then.

[tlively]

Thanks, I added those to the list above.

[dtig]

Following up, I've added provisional voting comments for the operations on the agenda for the meeting this week. Please vote ahead of the meeting so we can ratify/discuss as needed. @Maratyszcza do you think it makes sense to fold the discussion about packed floating point pairwise add (#375) into the extended pairwise addition discussion?

[Maratyszcza]

I think it is best to keep them separate. IMO, Extended Pairwise Addition is a high-priority instruction, but Floating-Point Pairwise Add is a nice-to-have.

[zeux]

Are signed 64-bit comparisons (#412) part of the agenda? I was under the impression that we've voted on this in the last meeting and the vote results were divisive, but that issue also has a voting comment now.

[tlively]

No, they are intentionally not on the agenda right now because we voted and failed to achieve consensus on them. That being said, the vote in the comments on that issue might still be useful in letting us know in case consensus emerges before Friday. If it looks like we have consensus after all, we can bring the issue back up for a vote at the meeting.

[lars-t-hansen]

No, they are intentionally not on the agenda right now because we voted and failed to achieve consensus on them. That being said, the vote in the comments on that issue might still be useful in letting us know in case consensus emerges before Friday. If it looks like we have consensus after all, we can bring the issue back up for a vote at the meeting.

The votes are now exactly as they were last week: five in favor, two against, and the rest presumably neutral.

[tlively]

Thank you everyone! Here are the notes.

We had successful consensus votes or unanimous consent on the following actions:

• To not include prefetch (Prefetch instructions #352) in this proposal
• To not include i64x2 min or max (i64x2.min_s and i64x2.max_s instructions #417, i64x2.min_u and i64x2.max_u instructions #418) instructions in this proposal
• To merge extended pairwise addition (Extended pairwise addition instructions #380) to the proposal
• To merge f64x2 conversions (Double-precision conversion instructions #383) to the proposal (although there was interest in splitting this into multiple votes given extra time)
• To merge popcount (Population count instruction #379) (this was more contentious; see the issue for commentary)

We also decided to push back our vote for phase 4 in the CG so that we could meet again next week to address more outstanding issues first.