https://github.com/MasonProtter/Gaius.jl I the above and is amazed. I always thought that BLAS, MKL etc are so complex and so well optimised that they are pretty much irreplaceable. Even if Gauis is slower for cases, how close are we to doing away with BLAS altogether even as an experiment. E.g c…

It’s still missing a bunch of kernels, but it’s quite usable today for the kernels it has. https://github.com/YingboMa/MaBLAS.jl and https://github.com/chriselrod/PaddedMatrices.jl are two other examples as well. It’ll take quite awhile to make these into a full BLAS mostly because no one is working…

Regarding Gaius.jl, I wrote a bit about this here: [ANN]: PaddedMatrices.jl, Julia BLAS and partially sized arrays - #28 by Mason TLDR: I’m not working on Gaius right now and I have no real plans to. It was never a serious project, more just an exploration of what was possible to do at a high level…

After the new update to @mcabbott ’s wonderful Tullio.jl , I really feel foolish for not at least giving it a shout-out above, especially in light of Slightly better multi-threading by mcabbott · Pull Request #16 · mcabbott/Tullio.jl · GitHub . I’m sure there are many corner-cases to work out, but wit…

[image] Mason: My real hope is that someone can devise a index notation based multidimensional tensor contraction library where BLAS just comes out as a special case, rather than building up the tensor contractions out of BLAS kernels. As far as I know, such a library is pretty speculative and r…

[image] Mason: If you want to try making a julia binary that uses Tullio.jl rather than OpenBLAS so Tullio.jl doesn’t use any BLAS calls underneath the hood at all?

[image] xiaodai: so Tullio.jl doesn’t use any BLAS calls underneath the hood at all? Yes

Why does Tullio allocate that much for largish arrays? I guess addressing allocations would improve performance?

My understanding is that those allocations are inherent to multi-threading, and that’s why they only appear for large arrays (i.e. multi-threading is only turned on for sufficiently large arrays). If I understand correctly, OpenBLAS will be making the same sort of heap allocations for it’s multi-thr…

[image] Mason: Looks like I was perhaps overly pessimistic. On my machine, a multi-threaded Tullio is beating multi-threaded OpenBLAS over a gigantic range of matrix sizes, and even when it starts to lose at very large sizes, it’s doing so quite gracefully I see a larger drop: julia> foreach…

Realistically, how close is Gaius.jl to becoming a full replacement for BLAS in Julia?

Internals & Design

pablosanjose July 9, 2020, 1:29pm 3

You might want to check out [ANN]: PaddedMatrices.jl, Julia BLAS and partially sized arrays for some truly impressive benchmark results + discussion of PaddedMatrices.jl

Topic		Replies	Views
LoopVec, Tullio losing to Matrix multiplication Performance	9	834	July 25, 2024
Julia matrix-multiplication performance Performance linearalgebra	20	9133	October 30, 2022
Tullio seems two times slower than basic LoopVectorization Performance question , tullio , loopvectorization	3	1113	April 10, 2022
Speed comparison matrix multiplication in Julia Performance question , linearalgebra , optimization , tullio	45	3594	August 19, 2021
OpenBLAS is faster than Intel MKL on AMD Hardware (Ryzen) Performance blas , lapack	40	37205	June 19, 2020

Realistically, how close is Gaius.jl to becoming a full replacement for BLAS in Julia?

Related topics