Stream: git-wasmtime

Topic: wasmtime / PR #3304 Implement `vany_true` and `vall_true`...


view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 09:44):

afonso360 opened PR #3304 from interp-vanyall to main:

This PR also changes the interfaces of extractlane and vectorizelanes. They now operate on Value instead of i128

This is a more ergonomic interface since pretty much every operation we do between these two functions is implemented on Value. With this interface we can also now start vectorizing bool values with the same interface.

I suspect that this change also makes insert_lane and extract_lane work on bool vecs, but I haven't tested it and we don't have existing tests for it.

This has an unfortunate downside that we had to add some clone's in some places. i128 is Copy but Value is not, so we have to explicitly annotate these moves.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 10:20):

afonso360 updated PR #3304 from interp-vanyall to main.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 11:27):

afonso360 updated PR #3304 from interp-vanyall to main.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 13:13):

bjorn3 submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 13:13):

bjorn3 created PR review comment:

Can you instead add a TODO?

view this post on Zulip Wasmtime GitHub notifications bot (Sep 06 2021 at 16:31):

afonso360 updated PR #3304 from interp-vanyall to main.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 07 2021 at 16:50):

abrown submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 07 2021 at 16:50):

abrown submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 07 2021 at 16:50):

abrown created PR review comment:

Side comment, no changes needed: yeah, wish this was easier. The decision to store the vector immediates in a side table was made to keep the size of InstructionData under some number of bytes. I wonder if things have changed enough since then for us to stuff 16 bytes into InstructionData.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 07 2021 at 16:50):

abrown merged PR #3304.


Last updated: Dec 23 2024 at 13:07 UTC