Elliot Saba

Okay, I'll just add a short-circuit to OpenBLAS's default thread calculation then. X-ref: https://github.com/JuliaPackaging/Yggdrasil/pull/5555

I am attempting to upstream this patch here: https://github.com/xianyi/OpenBLAS/pull/3773

Sadly, I can confirm that while this does help a little, it's not very significant. Here are the "commit charge" graphs for the start of the testset for both `master`...

> I'm confused, that's half of (Commit) memory, and that's useful? Yes, that is useful, but we were hoping that this would reduce the wildly high commit memory during tests,...

The thought is that OpenBLAS has [its own function that tries to auto-guess](https://github.com/xianyi/OpenBLAS/blob/7918ba11c29fc667b822162ebab112e97231bf1b/driver/others/memory.c#L350) how many threads to initialize with, which [doesn't agree with ours](https://github.com/JuliaLang/LinearAlgebra.jl/blob/8108bc4ff2dbd5b2143d6279efad640952f97f8e/src/LinearAlgebra.jl#L621), where we generally divide by two...

Yes, I think it's likely different. In my case, I think we may have tracked it down to old `liteuart` module versions being used with newer gateware; because the register...

Hello there, I'm one of the Julia developers that is impacted by this, and I thought I should explain a bit why we want stable links. The way Julia's package...

Hey Shreyas, this all looks reasonable to me; feel free to start hacking this out, and let us know when you run into something that is difficult; let's do our...

While I do expect some difference between the two implementations, we should be able to get a lot more than just 2 decimal places of accuracy. If we have to...

Agreed; I suggest tracking down the tensorflow computation and seeing if you can figure out where the difference lies; I think this is more than just floating-point noise; I think...