SIMD Vectorization in Julia (opens in new tab)

(software.intel.com)

126 pointsmlubin11y ago11 comments

11 comments

mlu11y ago

This looks great! I used Julia in my Master's thesis and it was very fast and easy to use. One thing however was annoying: The lack of shared memory parallel computing. Unfortunately, in Julia, each parallel process has its own memory such that you have to keep multiple instances of your data and/or move data around all the time. This can be a deal breaker if working with very large data sets and actually is preventing me from using Julia even more. But I'm sure, it's on the right track.

simonster11y ago

In Julia 0.3 there are SharedArrays on Linux/OS X: http://julia.readthedocs.org/en/latest/stdlib/base/#shared-a...

Hopefully there will also be multithreading someday, although right now it's still a work in progress: https://github.com/JuliaLang/julia/tree/threads

ihnorton11y ago

Shared-memory multiprocessing is supported in the latest release on linux and mac. Multi-threading is in development and has seen some significant progress over the past few months.

> But I'm sure, it's on the right track.

Yes, and it is worth pointing out that both this article and the @simd feature were written by Arch Robison - who is also the architect of Intel's Threading Building Blocks.

illumen11y ago

Doesn't it have mmap?

ninjin11y ago

It does, in the base library in fact:

https://github.com/JuliaLang/julia/blob/master/base/mmap.jl

Here is the relevant documentation:

http://julia.readthedocs.org/en/latest/stdlib/base/#memory-m...

pjmlp11y ago

Julia is really shaping up quite nicely.

etrain11y ago

Looks like a very nice feature - but would be great if they provided some actual benchmarks showing when vectorization is faster than not. For example, if you're bound by memory bandwidth, no amount of extra compute is going to help you.

hitlin3711y ago

as always, intel comes up with nice tutorial from time to time.

3JPLW11y ago

The author of this piece, Arch Robison, has been an active contributor to Julia. Not only did he write this tutorial, but he actually contributed the `@simd` implementation himself.[1]

He also presented this information as a talk at Juliacon back in June.[2]

1. https://github.com/JuliaLang/julia/pull/5355

2. https://www.youtube.com/watch?v=zFxaK0ufL04

tokai11y ago

Really nice to see Intel giving Julia some attention.

z3t411y ago

This needs to be ported to JavaScript, or at least update the Math library to include Vectors.

j / k navigate · click thread line to collapse