reading list

Zed uses Conflict-Free Replicated Data Types (CRDTs) to enable smooth collaboration in text editing. This technology allows multiple users to edit documents independently while ensuring that their changes merge correctly, avoiding conflicts. By structuring data with logical locations and unique identifiers, Zed maintains a consistent experience for all users, even...

The author argues that current concurrency methods, like the "go statement" in Golang, are similar to the outdated "goto" statement, making code harder to manage and understand. Just as removing goto from programming languages led to better control flow and new features, eliminating go statements could improve concurrency APIs. The...

To improve skills, focus on deliberate practice, which involves challenging yourself just beyond your current abilities. It's essential to track measurable progress and avoid getting stuck in a cycle of overwhelming difficulty or mindless repetition. Consistent effort and strategic learning will lead to long-term success and mastery.

Frustum culling helps reduce rendering time by only processing objects within the viewer's view. By using techniques like oriented bounding boxes and SIMD, the author improved frame time from 8.2 ms to 3.3 ms. This optimization makes both CPU and GPU work more efficiently.

A user named JJB9922 is proposing to merge five code changes into the main branch of a GitHub repository. The changes include fixing an incorrect import, converting to data-oriented design, and optimizing memory usage. The code has been reviewed and approved by another user named seatedro.

Multiplayer games using the Source Engine rely on a Client-Server networking model, where the server sends updates to clients about the game world state and receives input commands. Challenges like network bandwidth limitations and latency affect gameplay, causing delays and potential packet loss issues. To address these challenges, the Source...

Yoke is a project that allows you to manage infrastructure as code by writing definitions in Go or Rust, which are then compiled to WebAssembly. It simplifies the creation of Kubernetes manifests and reduces boilerplate code, making deployments easier. Yoke's security model uses WebAssembly to sandbox code and limit access...

This post discusses how a load balancer distributes HTTP requests to servers using different methods. Round robin is a simple approach but can lead to dropped requests when server performance varies. More advanced methods like dynamic weighted round robin and least connections optimize for both server power and request latency,...

fastDOOM is a version of the classic game DOOM that runs 30% faster without losing features. It achieves this speed through various optimizations and code improvements. The performance gains were especially notable on demanding maps, demonstrating the effectiveness of the enhancements made.

Phil Eaton shares his journey from being a web developer to a database developer over ten years. He learned about databases through personal projects and co-founding a startup, eventually landing a job at EnterpriseDB, a major contributor to Postgres. After a year there, he reflects on the rewarding experience of...

Zig 0.14.0 introduces several updates, including changes to the standard library, compiler, and build system. Notable features include multithreaded backend support, incremental compilation, and new builtins like @FieldType. The release also simplifies the way libraries are linked and improves error reporting for undefined behavior.

This article explains how to use noise functions for generating natural-looking landscapes in code. It covers different types of noise, such as red, pink, and white noise, and how to modify them for various effects. The author also shares techniques for combining noise functions to create unique textures and patterns.

Noise functions like Simplex and Perlin noise are used to create 2D maps by assigning values for elevation and moisture. By combining different frequencies of noise, we can generate varied terrains, such as hills and valleys, and define biomes like forests and oceans. This technique is popular for terrain generation...

The article discusses how to calculate 17 billion similarities in a project called Cookit. It covers techniques for performance optimization, such as using bit masks and managing memory usage. Additionally, the author shares tips on debugging high memory consumption in applications.

In C++, using `__builtin_unreachable()` can help optimize functions by indicating that certain code paths are never executed, leading to smaller assembly code. By ensuring input constraints, such as the length being a multiple of 32, the compiler can eliminate unnecessary branches and improve performance. However, applying this optimization with `std::vector`...

The FFmpeg School of Assembly Language teaches how to write assembly code for high-performance image processing. It focuses on using Intel syntax and SIMD (Single Instruction Multiple Data) to handle multiple data elements simultaneously. The lessons aim to help learners understand assembly language and its application in FFmpeg, making it...

Loops can slow down programs, so optimizing them is crucial for better performance. Techniques like restructuring loops, eliminating unnecessary statements, and using faster operators can significantly speed up code execution. It’s important to balance optimization with code readability to maintain clarity for future developers.

Johnny’s Software Lab specializes in software performance and provides solutions for optimization issues. The article discusses a complex code example that, while difficult to read, offers significant performance improvements over a simpler version. Although the code is considered "horrible," it effectively enhances speed, especially in scenarios where performance is crucial.

Programmers should only optimize for performance after profiling shows a real need. Introducing too much indirection can slow down a program significantly, making it hard to fix later. Sometimes, major rewrites can yield huge speed improvements, but small tweaks often won't achieve the same results.

When designing a performance-sensitive computer system, understanding the relative costs of different operations is crucial. It's essential to grasp the trade-offs between network I/O, disk I/O, load from DRAM, and more. Back-of-the-envelope analysis and microbenchmarking are valuable skills in comparing performance without building two alternative designs. Jeff Dean's table of...

The document discusses building a high-performance fuzzer with GPUs for fuzzing embedded software in the cloud. Fuzzing involves supplying programs with randomized inputs to uncover unexpected behavior, with coverage-guided fuzzers focusing on finding inputs that explore new code paths. The document explains the challenges of fuzzing embedded software due to...

This article introduces vectorization, a technique that allows a CPU to process multiple values in one instruction, improving performance. It compares scalar and vector versions of a code example, showing how vectorization can handle more data in fewer instructions. Future articles will explore various topics related to vectorization, including best...

The text discusses designing a SIMD algorithm that efficiently handles u64 values without requiring full additions. It explores mapping data in SIMD lanes and highlights the need for a flexible implementation that can adjust the number of lanes for better performance. The goal is to optimize the algorithm for processing...

Cache memory in advanced processors can speed up performance but also introduces unpredictability that can surprise designers. This paper discusses how cache behavior affects real-time embedded systems, particularly using the Intel 80486 as an example. It suggests careful measurement and cache management to mitigate performance issues in critical applications.

Most programming jobs focus on creating software that solves business problems, often in boring ways. Instead of calling yourself a programmer, emphasize how you help businesses increase revenue or reduce costs. Strong communication and negotiation skills are crucial for showcasing your value in the workplace.

The author created a simple search engine that indexes HackerNews, using an inverted index for faster document retrieval. They implemented text normalization and a ranking algorithm based on word frequency and rarity to improve search results. This project demonstrates the basics of building a search engine, and the author encourages...

The text explains how a CPU performs simple operations using registers and an arithmetic/logic unit (ALU). It highlights the role of registers in storing data, addresses, and managing program state, as well as how pointers and local variables are handled in memory. Additionally, it discusses the importance of separating memory...

Will Crichton presents a new approach to teaching programming languages by using systems programming languages like WebAssembly and Rust to make theoretical concepts more relatable. This curriculum helps students connect their experiences with low-level programming to programming language theory, making learning more grounded and enjoyable. Feedback from students indicates that...

Reading Time: ~15 minutes Writing systems in C++ is a balancing act between flexibility, performance, and maintainability. Over time, I’ve developed a personal framework that helps me approach these systems in a modular way, ensuring they’re efficient, scalable, and easy to understand (especially for someone new reading my code). This...

As no-nonsense as possible 2024-09-07 18:42:57 Saturday Resources I have looked at or utilized, no particular order. This is a growing list that I'll update periodically. Structure is: {{ Language | Media Format | Field }} Resource Enjoy! Resources {{ C++ | Text | Programming Languages }} LearnCPP {{ C++...

Why I don't hate it 2024-09-09 18:57:14 Monday Zig this, Rust that. Odin here, GoLang there. I get it! C++ has flaws. If the standard library was in a book, it'd be thicker than a bowl of oatmeal. If you hop on a codebase that's survived 4 major C++ releases,...

std.debug.print("pretty good..."); 2024-10-13 21:14:37 Sunday Zig is a pretty new systems programming language. It is yet another 'C killer', and I think (for once) it could be just that—with a caveat. Of course, C is sitting on billions of devices and isn't going anywhere in my lifetime. However, if I...

Data Oriented Design from the Lens of an OOP Guy 2025-01-04 18:48:19 Saturday My first programming language was Java. Back when I was 13 years old, trying to get into the Minecraft modding scene. At this time, Martin's Clean Code had firm roots in the language's community. Meaningful names, small...

Or how I got sidetracked from the SICP book 2025-01-10 22:35:25 Friday My first programming language was Java, and like a lot of young programmers, I was introduced to the world of programming through the lens of Object-Oriented Programming (OOP). Moving to C broke this mold a little, with C...

Reinforcement learning uses rewards to help algorithms learn the best actions to take in various situations. It balances exploration of new moves with exploiting known good moves, similar to playing chess without analyzing every option. The Actor-Critic model improves learning efficiency by combining two networks: one that decides actions and...

C is considered a safe programming language because it has a long history of successful, trusted software development, despite some criticisms regarding its security. Many developers appreciate C for its explicitness and control, which can lead to more secure code when used correctly. The language's widespread use and robust tooling...

Richard W. Hamming reflects on his experiences with great scientists and their differences from others. He emphasizes that success in science requires emotional commitment and the ability to seize opportunities. Hamming believes that studying one's successes is crucial for understanding how to achieve greatness in research.

In "You and Your Research," Hamming emphasizes the importance of working on significant problems to achieve success. He encourages hard work, learning from successes, and collaborating with important people in your field. Hamming believes that focusing on excellence and adapting to change is crucial for making meaningful contributions.

Dr. Richard W. Hamming discussed why some scientists make significant contributions while others are forgotten during a seminar at Bell Labs. He emphasized the importance of working on important problems and having the courage to pursue great ideas. Hamming also noted that successful scientists often turn challenges into opportunities for...

Data Oriented Design (DoD) focuses on organizing data to improve computer performance, especially in high-demand applications like games and simulations. Unlike Object-Oriented Programming (OOP), which can complicate data access, DoD uses structures of arrays to enhance cache efficiency and parallel processing. While DoD offers performance benefits, it may not always...

The text discusses using 'index-handles' instead of raw or smart pointers for memory management in C and C++. It suggests centralizing memory management into systems, grouping items into arrays, and converting handles to pointers only when necessary. By following specific rules, such as not storing pointers and using handle-to-pointer conversion,...

Mitchell Hashimoto shares a structured approach to understanding and contributing to complex software projects. He emphasizes starting as a user, tracing features, and gradually learning the codebase while engaging with the project's community. By breaking the process into manageable steps, he encourages others to tackle complexity with confidence.

This article introduces a series about debuggers, highlighting their essential role in programming and software development. The author emphasizes that debuggers are not just tools for fixing errors, but also valuable for analyzing and understanding code behavior. Upcoming posts will explore various aspects of debuggers, including their architecture and features.

Programs run in their own virtual address spaces, allowing them to use memory independently without interfering with each other. The operating system manages these spaces and translates virtual addresses to physical addresses, enabling efficient memory usage. This system allows many programs to run simultaneously, each perceiving access to a large...

"The Book of Shaders" is a beginner-friendly guide to understanding Fragment Shaders. It covers topics like image processing, 3D graphics, and simulations in a step-by-step manner. The authors, Patricio Gonzalez Vivo and Jen Lowe, combine art and technology to help readers explore this complex subject.

This page provides resources for amateur compiler writers, focusing on practical and useful literature. It includes recommendations for books, papers, and algorithms related to compiler construction, optimization, and code generation. Suggestions for additional materials are welcome, and the document organization is arbitrary.

The arena allocator is a simpler alternative to traditional manual memory management in C, which often leads to complex issues with malloc and free. Instead of managing individual allocations, the arena allocator groups memory usage together, making it easier to handle lifetimes and release memory. This approach improves performance and...

The author reflects on their year of writing C code, finding it enjoyable and productive. They emphasize the importance of choosing the right language for each problem and share insights on the benefits of using C over C++ in certain scenarios. Additionally, they discuss the advantages of C99 improvements and...

In this video series, Charles Cabergs explains advanced concepts in the C programming language, starting with function pointers. He demonstrates how function pointers allow functions to be passed as arguments to other functions, making code more flexible and generic. The video also covers practical examples, such as using function pointers...

The perfect voxel engine allows developers to easily create and manipulate voxel data for games. It should efficiently handle various features like lighting, physics, and procedural generation. By focusing on flexible data structures and conversion processes, this engine can support dynamic gameplay and user modifications.

The paper discusses high-performance AI compilers, focusing on efficient code generation and hardware-specific optimizations using techniques like tiling and kernel fusion. It highlights the Linalg dialect, which simplifies tensor operations and allows for effective optimizations across different frameworks. The use of libraries like libsxmm enhances performance by fusing operations and...

MLIR is a versatile compiler infrastructure designed to address software fragmentation and improve compilation for different hardware. It aims to reduce the cost of building domain-specific compilers and facilitate the connection of existing compilers. MLIR offers a standardized approach to code generation and optimization across various application domains and hardware...

ggml is a lightweight, open-source machine learning library focused on Transformer inference, developed in C and C++. It is minimalistic and easy to compile, making it a popular choice for on-device machine learning tasks. The library supports various hardware backends and is still evolving, which may require users to have...

Ming-Lun "Allen" Chou shares tutorials on game physics simulation and engine architecture on his website. The series covers topics like motion dynamics, collision detection, and constraint-based physics. These resources aim to help others learn about implementing realistic physics in games.

The text is about a website called x86re.com that offers resources on reverse engineering. It includes a series of tutorials for beginners, covering topics like executables and Portable Executable files. The goal is to help newcomers understand x86 syntax and stack frames in reverse engineering.

Rust has been redesigning its compiler to support incremental compilation, which allows for faster builds by only recompiling changed parts of the code. This new approach uses a query-based system instead of the traditional pipeline, enabling more efficient dependency tracking and caching of results. Although this method adds complexity, it...

Monolith is a large-scale recommendation system designed to improve real-time learning and model quality by using a collisionless embedding table. It allows for fast synchronization of model parameters, enabling it to quickly adapt to user feedback. This system consistently outperforms traditional methods while maintaining efficient memory use.

Dithering is a technique that uses a limited color palette, like black and white, to create the illusion of more colors by adjusting pixel brightness. Various methods, such as ordered dithering and error diffusion, can enhance the visual quality of images by distributing quantization errors. Techniques like Bayer and blue...

Zig arrays have a fixed length known at compile-time, but for situations where the length is unknown until runtime, dynamic allocations like std.ArrayList are often used. std.BoundedArray is a solution for cases where the maximum length is known, combining the length and array into a single structure for better safety...

The article introduces io_uring, a new Linux IO interface designed for efficient asynchronous input/output operations. It explains how io_uring uses shared memory and eliminates the need for locking between applications and the kernel, improving performance. The article also highlights the significant increase in IOPS (input/output operations per second) that io_uring...

This text discusses a programmer-friendly I/O abstraction using io_uring for Linux and kqueue for FreeBSD/macOS to handle file descriptor readiness efficiently. It highlights the benefits of batching I/O requests to reduce system call overhead and simplify event handling in applications. The proposed solution includes a central I/O dispatcher to manage...

The content discusses the Zig Showtime, a show created during the pandemic lockdowns with goals to entertain, share interesting projects, and provide hosting experience. The show aims to also give community members a platform to share their projects. Mitchell Hashimoto discusses his new terminal emulator, Ghosty, highlighting its features and...

The text discusses the creation of a music player application using different programming languages and frameworks. It highlights challenges faced, such as dependencies on C++ and issues with features breaking due to updates from external services like YouTube. The author emphasizes the importance of a systems-minded approach to improve code...

Attention mechanisms are crucial for how transformers, like those used in large language models, process information by calculating the "usefulness" of each word in relation to others. The encoder uses self-attention to analyze words within the same sentence, while the decoder employs masked self-attention to consider only previous words. Each...

The video discusses challenges with C programming, including complicated code and error handling. It introduces Zig, a programming language that simplifies code and improves error management. Zig also functions as a C compiler, making it easier for developers to write and debug software.

The article explains how to implement a simple neural network in C, consisting of input, hidden, and output layers. It describes the training process using forward and backward propagation to improve accuracy over multiple epochs. By the end of training, the model achieves an accuracy of 98.22% on the test...

The article discusses how dynamic lighting in the 9th generation of gaming consoles has declined in quality compared to previous generations. It emphasizes the need for better image quality and performance instead of simply increasing resolution. The author argues for improved lighting techniques that enhance visual standards while being more...

The Byte Latent Transformer (BLT) is a new model that processes raw byte data without using tokenization, achieving performance similar to traditional models while being more efficient. It dynamically groups bytes into patches based on their information density, which helps improve scaling and robustness against noisy inputs. BLT demonstrates significant...