Covector-matrix multiplication is broken

[DilumAluthge]

[DilumAluthge]

This is pretty much the last thing we have to fix in Gaius. After we fix this, we can put Gaius into "stable"/"maintenance" mode.

[DilumAluthge]

It looks like Gaius.jl might get new life! In that case, we should definitely figure out how to fix covector-matrix multiplication.

cc: @MasonProtter @chriselrod @tkf

[tkf]

we can put Gaius into "stable"/"maintenance" mode

Oh, I didn't know about the plan. My interest is rather selfish in that Gaius is nice for research playground. Gaius uses a pretty straightforward "textbook" approach to task parallelism so it's really nice to play with for me (e.g., patch it for Tapir, testbed for DAG analyzers). But, if the plan was to transition this to a maintenance mode, I don't want to disturb it.

That said, I think it'd be nice to keep Gaius alive as proof of how much you can do in straightforward task parallelism in Julia. I would have implemented DAC GEMM myself anyway if Gaius didn't exist. So I think I can help the maintenance at least as long as I'm playing with Tapir.

[DilumAluthge]

Oh, I didn't know about the plan.

Well that was the old plan, because I didn't think that we would be further developing Gaius.jl given that Octavian.jl exists.

But with Tapir, I think Gaius.jl could find new life, as @MasonProtter wrote on Zulip:

If TAPIR progresses well, then Gaius.jl might live on

giving Gaius a LRU cache, Tapir, and some more tuning would probably put it in the same ballpark as Octavian

So I think that could be a good new plan.