We’ve discussed SIMD and vectorization extensively on this blog, and it was only a matter of time before SIMD (or vector) functions came up. In this post, we explore what SIMD functions are, when they are useful, and how to declare and use them effectively. A SIMD function is a function that processes more than… Read
Introduction Readyset is designed to serve queries from cached views with sub-millisecond latency. This post focuses on the cold path—cases where a cache miss forces execution against the base tables. In these scenari...
I built a simple web crawler using good old platform threads. It was just a multithreaded crawler, nothing fancy. But then, curiosity…
How I investigated and worked around a kernel bug that caused async-profiler to freeze my machine whenever I tried to use a profiler.
Consider this tale of I/O and performance. We’ll start with blocking I/O, explore io_uring and kqueue, and take home an event loop very similar to some software you may find familiar.
For many years, Mozilla has continued to provide Firefox for 32-bit Linux systems long after most other browsers and operating systems ended support. We made this choice because we care ...
Json parsing in C using function pointers attached to a struct
Both GCC and Clang support the -fsanitize=undefined flag which instruments the generated code to detect places where the program wanders into parts of the C language specification which are either undefined or implementation defined. Many of these are also common programming errors. It would be great if there were…
My little corner of the web for sharing wherever my interests take me.
Writing Super-Fast Matrix Multiplication with HIP, RGP, and ISA
Every field of human activity has its unique characteristics, and programming is no exception.
I recently tried using the Minisforum DEG1 GPU Dock with a Raspberry Pi 500+, using an M.2 to OCuLink adapter, and this chenyang SFF-8611 Cable. After figuring out there's a power button on the DEG1 (which needs to be turned on), and after fiddling around with the switches on the PCB (hidden under the large metal…
The “Exo 2” language allows programmers to write “schedules” that explicitly control how the compiler generates code. This allows performance engineers to transform simple programs into complex programs that do the same thing as the specification, but faster.
meshoptimizer implements several geometry compression algorithms that are designed to take advantage of redundancies common in mesh data and decompress quickly - targeting many gigabytes per second in decoding throughput. One of them, index decoder, has seen a significant and unexpected variance in performance across…
You're probably reading this page because you've attempted to access some part of my blog (Wandering Thoughts) or CSpace , the wiki thing it's part of. Unfortunately you're using a browser version that my anti-crawler precautions consider suspicious, most often because it's too old (most often this applies to versions…
I’d like to talk more about a container technology that I really like that I touched upon in a previous article : systemd-nspawn .
Owen's personal website and portfolio
In this blog post, we walk through how to use Matrix Cores in HIP kernels, with a focus on low-precision data types such as FP16, FP8, and FP4, as well as the new family of Matrix Core instructions with exponent block scaling introduced in the AMD CDNA™4 architecture. Through code examples and illustrations, we…
While not an absolute requirement for modern graphics card support on Linux, Resizeable BAR support makes GPUs go faster, by allowing GPUs to throw data back and forth on the PCIe bus in chunks larger than 256 MB. In January, I opened an issue in the Raspberry Pi Linux repo, Resizable BAR support on Pi 5. Unlike most…
A story about a bug with logical replication slot creation on Postgres read replicas
Hello, While a lot of time of 2.0 development has been spent on new features and quality of life, we still take care of the smaller details and technical improvements.
At Terrateam, we are big fans of Fly.io. The service is hosted there and it’s served us well. Just deploy your TOML file, get your infrastructure, do something else with the rest of your day. One of the interesting sides of Fly is that they invest heavily in server-side SQLite. They’ve written a number of blog posts…
You may know that, because your computer has different caches (L1, L2, L3...), and memory operations operate on cache lines of about 64 bytes each, you should write programs that exhibit locality to get maximum performance.
A very common adage in ODE solvers is that if you run into trouble with an explicit method, usually some explicit Runge-Kutta method like RK4, then you should try an implicit method. Implicit methods, because they are doing more work, solving an implicit system via a Newton method having “better” stability, should be…
If you’ve been in this trade for a while, you have probably seen dozens of debates on the merits and problems of SQL as a relational database query language. As an ORM maintainer, I have a few gripes with SQL, but overall it is workable, and anyway, it has so much inertia that there’s no point fantasizing about a…
After my previous post on ARM exploitation, where we crafted an exploit for a known vulnerability, I decided to continue the research on a more modern IoT target. In this follow-up post, I will take you through building a considerably more complex binary exploit. We will explore the path from firmware extraction and…
where my words occasionally escape /dev/null
In my previous post Serving 200 million requests per day with a cgi-bin, I did some quick performance testing of CGI using a program written in Go. Go works excellently for CGI programs, for many of the same reasons it works so well for CLI programs and system daemons. But, out of curiosity, I decided to do a bit more…
