Seeking recommendation on SIMD learning materials 🙃

akrauze · March 1, 2025, 4:28pm

Hi. I recently decided to start learning SIMD, and after a couple of hours of looking at manuals without any prior experience I have managed to vectorize a simple function. It was fun and I would like to continue learning, but I think it would be more fruitful if I would do it in a more structured way.

I have tried searching online for some materials, but so far I have only found either manuals, cheat sheets, or random blogs with solitary examples. So I would like to ask you, do you know any good books/materials/tutorials/courses/exercises/web series/talks/etc. that could help me with my learning?

More specifically I am looking for something that would:

cover mostly x86_64
explain how it exactly works from the perspective of a CPU; what registers are involved, how instruction "latency" looks like compared to "normal" scalar operations, how to read generated assembly, etc.
explain history of SIMD extentions, and what each of them introduced
preferably be in a written form
would include many exercises; I would like to try to come up on my own with solutions to specific algorithms, and then to compare them to already existing implementations. Generally I learn best by doing.
I want to write it in Rust, but since Intel's intrinsics names are the same, the learning materials can by in C/C++; I don't think it should make a difference
if it is of really high quality I am willing to pay for such materials
give me general knowledge and teach techniques of approaching different problems (I know that I can only master it by continuously using it; I just want to give myself a proper base from which I can start)

If you could recommend me something like this, I would be very obliged.

casibbald · March 1, 2025, 5:19pm

I assume you have looked at resources such as the following:

github.com/rust-lang/portable-simd

beginners-guide.md

master


# Beginner's Guide To SIMD

Hello and welcome to our SIMD basics guide!

Because SIMD is a subject that many programmers haven't worked with before, we thought that it's best to outline some terms and other basics for you to get started with.

## Quick Background

**SIMD** stands for *Single Instruction, Multiple Data*. In other words, SIMD is when the CPU performs a single action on more than one logical piece of data at the same time. Instead of adding two registers that each contain one `f32` value and getting an `f32` as the result, you might add two registers that each contain `f32x4` (128 bits of data) and then you get an `f32x4` as the output.

This might seem a tiny bit weird at first, but there's a good reason for it. Back in the day, as CPUs got faster and faster, eventually they got so fast that the CPU would just melt itself. The heat management (heat sinks, fans, etc) simply couldn't keep up with how much electricity was going through the metal. Two main strategies were developed to help get around the limits of physics.
* One of them you're probably familiar with: Multi-core processors. By giving a processor more than one core, each core can do its own work, and because they're physically distant (at least on the CPU's scale) the heat can still be managed. Unfortunately, not all tasks can just be split up across cores in an efficient way.
* The second strategy is SIMD. If you can't make the register go any faster, you can still make the register *wider*. This lets you process more data at a time, which is *almost* as good as just having a faster CPU. As with multi-core programming, SIMD doesn't fit every kind of task, so you have to know when it will improve your program.

## Terms

SIMD has a few special vocabulary terms you should know:

* **Vector:** A SIMD value is called a vector. This shouldn't be confused with the `Vec<T>` type. A SIMD vector has a fixed size, known at compile time. All of the elements within the vector are of the same type. This makes vectors *similar to* arrays. One difference is that a vector is generally aligned to its *entire* size (eg: 16 bytes, 32 bytes, etc), not just the size of an individual element. Sometimes vector data is called "packed" data.

This file has been truncated. show original

casibbald · March 1, 2025, 5:20pm

and:

akrauze · March 1, 2025, 5:59pm

This one I have read. But this is not quite what I need, because I firstly want to learn just SIMD, before using abstractions.

I haven't found it, but after a glance this looks very promising. Thanks!

robertknight · March 1, 2025, 7:45pm

C++ conference talks have been a useful source of information about SIMD, where the lessons are usually platform agnostic. Some examples:

Another talk about SIMD and performance more generally:

https://www.youtube.com/watch?v=vIRjSdTCIEU

The author of this talk has an online book at Algorithms for Modern Hardware - Algorithmica which covers both CPU architecture, which is very useful to understand for getting the best performance out of SIMD code, as well as a chapter (10) that covers SIMD specifically.

blonk · March 2, 2025, 1:58am

The ffmpeg project somewhat recently released an assembler course. I haven't had time to look at it, but when they announced it they made it clear that the primary objective was to teach people SIMD (specifically because they need more people to help out with ffmpeg). So with the caveat that I haven't actually looked it myself, you might be interested in FFmpeg School of Assembly Language.

Topic		Replies	Views
Any good resources on using simd in rust? help	2	443	May 30, 2020
Are there any comprehensive primers on Rust SIMD? help	1	268	July 11, 2022
What's everyone working on this week (39/2021)? community	4	823	December 26, 2021
Advice on Learning resources help	5	546	September 14, 2024
Blog post: using SIMD to speed up dynamic programming algorithm	2	1249	January 12, 2023

Seeking recommendation on SIMD learning materials 🙃

Related topics