No vector TG meeting tomorrow - preparing to start public review
There are some minor suggestions and edits (thank you!), and I will incorporate these into the v1.0 version which we’ll freeze for public review, hopefully starting in a day or at most a few days.
Remember, you can still comment during public review.
Krste
I am disappointed that the meeting was cancelled.
I am concerned that jnk0le's contributions/concerns, in particular, have been dismissed and to the extent that they were objections to the draft have been categorized as not substantial.
The meeting was a final verification/certification opportunity. A ratification of sorts.
I have found our meetings productive and collectively raised concerns and insights that have improved the V extension and its documentation.
I intended to raise some concerns that are difficult to express as an "issue" and hoped to get the groups input more dynamically.
Given the meeting is cancelled, I will type these in as best I can as issues.
I apologize that I did not anticipate the cancellation of the
meeting, and thus that I let my life interfere with providing
these concerns as issues.
[Our son suddenly was at the top of the waiting list for long
term care placement, which means in Ontario that you drop
everything to make it work. And further, with covid restrictions
we needed to ensure, on admission, he has completely set up as he
is now quarantined for 14 day. Also that they messed up his
medication and I spent the better part of a day, and the night it
happened, getting that corrected. He has Dopamine Responsive
Dystonia and takes levo/carbadopaCR and pramipexole. Thus, they
thought he has Parkinsons and concentrated his medications to
daytime hours, leaving him without his medication for 12 hours
over night. It took a full 36 hours before he was stabilized and
free of adverse effects. And this is happening on the background
of us selling our house, downsizing, moving, new rental
accommodation set up and shut off old services and legal matters,
including document preps and signing papers. Once again, I
apologize, I know of others in worse constraints that are on the
lists etc. 24/7. So it is just me, given others can contribute
when their life is hectic.]
Contributing during public review does not prepare the document for public consumption. I believe cancelling the meeting is a lost opportunity to do that.
There have been no substantial objections raised on the v1.0-rc1 draft, so I will cancel the meeting tomorrow.
There are some minor suggestions and edits (thank you!), and I will incorporate these into the v1.0 version which we’ll freeze for public review, hopefully starting in a day or at most a few days.
Remember, you can still comment during public review.
Krste
I did not expect that this meeting, of all meetings, would be cancelled.
So I’ll send my comments by email now rather than discussing an important one at the meeting and filing issues for minor ones. I think #1 is important. The rest are lesser comments:
- I’m (still) concerned about the lack of a whole register load with “mask” type hint. I think leaving it out damages the clarity of support for micro-architectural redistribution, which is critical for wide SIMD.
- If we really will not add the instruction, I’m thinking there’s a relatively simple sequence that does the job, like save vl, set it to max, do an ordinary mask load, restore vl. Could you add that sequence somewhere. Maybe in the note about not having the mask version. That would help make the completeness of support clear for micro-architectural redistribution.
- In Section 3.5, I’m not sure why vlenb must be a design-time constant. An implementation could vary it as a mode. So, I think the wording should allow that.
- Section 13.19 has a typo. It says “widening” when I think it means “narrowing.”
- In Section 14, what is the thinking on why there’s no update of the destination of reductions when vl=0? I can’t see any reason not to update – it seems easier to update – and it could, in some circumstances matter. This rule means that code has to ensure that it does not execute a reduction in a rare case that vl might be zero unless the source scalar and destination scalar are the same register. Why require that?
- In Section 7.8, I thought we wanted the set of registers a segment load/store could access to be limited to an aligned group of eight to allow optimizations on the renaming of registers. So a three register group could start at register #3, for example. But here it says that a sequence can start anywhere.
Bill
Sent: Friday, June 25, 2021 3:41 AM
To: tech-vector-ext@...
Subject: [RISC-V] [tech-vector-ext] No vector TG meeting tomorrow - preparing to start public review
EXTERNAL MAIL
There have been no substantial objections raised on the v1.0-rc1 draft, so I will cancel the meeting tomorrow.
There are some minor suggestions and edits (thank you!), and I will incorporate these into the v1.0 version which we’ll freeze for public review, hopefully starting in a day or at most a few days.
Remember, you can still comment during public review.
Krste