Hacker News Reader

Comment by komuher | Hacker News Reader

For numeric code sure but its because Julia are using BLAS (or any other instruction for CPU/GPU u give to LLVM). Julia (no BLAS) -> In matmul its on pair with Golang and Swift and a bit slower then RUST and Nim. If u need BLAS then just use lib with that :)

View on HN

I'm having a hard time trying to follow your argument. You seem to have had some issues with Julia, and I would like to take those issues at face value. It would be nice if you could point to specific issues since we always like to fix things.

However, in response to discussion about Julia having an integrated set of abstractions that provide high performance - you are discussing a collection of features in Rust, Go, Swift, Nim, TensorFlow, Jax, PyTorch, etc. For any single feature, there can always be some other thing that does it better. But it is unclear how that makes one system better than another.

Numeric code doesn't just mean matrix multiplication. If I have a random nested for loop, it will probably run at least as fast as those languages(apart from maybe Nim, never used it) if annotated with @simd and @inbounds. If I'm operating on a small array/matrix, then Julia will blow Go/Rust out of the water via stack-allocated static arrays (https://github.com/JuliaArrays/StaticArrays.jl). These can't be implemented in Go because it doesn't support type parameters, let alone integer type parameters, and is only recently supported in Rust via const generics, which as far as I'm aware haven't stabilised yet.

Ok let me get it here cause there are a lot of misunderstanding my post on top.

Swift have built in simd same about Rust https://github.com/apple/swift-evolution/blob/master/proposa...

Nim have arraymancer https://github.com/mratsim/Arraymancer

@inbounds it is just compiler option u can use it in any language with LLVM backend i would be suprised if Julia will be faster then Rust/Swift or Nim in this regard. But True about GO in that particular case.

>@inbounds it is just compiler option u can use it in any language with LLVM backend i would be suprised if Julia will be faster then Rust/Swift or Nim in this regard.

I agree it wouldn't necessarily be faster, but it also wouldn't be slower. Plus in Rust at least disabling bounds checks requires marking code as unsafe, which really gets the community's hackles up.

It only requires unsafe if the compiler can’t figure out that removing them is okay. Often bounds checks are not emitted, even in safe code.

Actaully, julia for loops can be made to perform at BLAS levels with this package: https://github.com/chriselrod/LoopVectorization.jl.

Here are my early experiments at making a pure-julia multi-threaded BLAS using LoopVectorization.jl https://github.com/MasonProtter/Gaius.jl. It absolutely blows a naive triple for loop out of the water and is quite competitive against OpenBLAS until you get to very big sizes.

For small, statically sized arrays, LoopVectorization + tripple loops is also much faster than MArrays. LoopVectorization doesn't support SArrays yet, because you can't get pointers to them.

MArrays will be stack allocated if they don't escape.

One of my in development packages also uses it's own "stack" (MMap a chunk of memory), so that it can have pointers to fast "stack-alocated" arrays.

I played around with LLVM's alloca a bit, but it seems like I could only ever use a single alloca at a time; if I ever used more than one, LLVM would just return the same pointer each time instead of incrementing it. If I have to manage incrementing the pointers myself anyway, I may as well use my own stack, too.

For (the problems I have tested and tuned it on), LoopVectorization produces faster code than C/Fortran, e.g.: https://chriselrod.github.io/LoopVectorization.jl/latest/exa... But it may be more fair to compare it with plutocc. In my early tests (which involved much larger problem sizes), plutocc does a lot better, because (unlike LoopVectorization) it seems to consider memory/caches rather than just registers allocation and instruction costs.

Providing a well functioning BLAS experience in a language is a non-trivial task. Saying "just use lib" seems to severely understate this.

How many go or swift or Nim issues discuss matmul vs the Julia repo?

Nim have Arraymancer https://github.com/mratsim/Arraymancer sure it isn't easy but still most of top dogs have some libs supporting BLAS/CUDA etc.

Reply on news.ycombinator.com