llama.cpp is a high-performance C/C++ inference engine for large language models (LLMs) and vision-language models (VLMs). It loads models in the GGUF format and runs them on a wide range of hardware, including CPU, CUDA, Metal, Vulkan, SYCL, OpenCL, Hexagon DSP, and more. The project provides a core library (`llama.cpp` and `ggml`), a CLI client (`llama-cli`), an OpenAI-compatible HTTP server (`llama-server`), a Svelte-based web UI, and a suite of tools for quantization, perplexity evaluation, and model conversion. It supports dozens of model architectures (LLaMA, Mistral, Qwen, Gemma, DeepSeek, etc.) and advanced features like speculative decoding, grammar-based sampling, flash attention, and LoRA adapters. The architecture is modular: the `ggml` tensor library provides a pluggable backend abstraction, while model-specific graph builders in `src/models/` construct computation graphs. The project is actively maintained with frequent releases and broad community adoption.
This is the Godot Engine, a fully-featured, cross-platform 2D and 3D game engine. The repository contains the complete source code for the engine runtime, the built-in editor, and all supporting modules (rendering, physics, audio, networking, scripting, etc.). The architecture is a monolithic C++ core with a modular plugin system, supporting multiple rendering backends (Vulkan, D3D12, Metal, OpenGL), physics engines (Jolt, GodotPhysics), and scripting languages (GDScript, C# via Mono). The editor is a full-featured IDE built on the engine itself, with scene editors, debugger, asset pipeline, and export system. The codebase is exceptionally well-structured, with clear separation of concerns between core, servers, scene, and platform layers. It demonstrates world-class engineering in real-time graphics, physics simulation, audio processing, and cross-platform abstraction. While the code is tightly coupled to the Godot framework, many individual components (math libraries, spatial data structures, serialization utilities) exhibit high craftsmanship and could be extracted with moderate effort.
ncnn is a high-performance neural network inference framework designed for mobile, embedded, and desktop deployment. It provides a self-contained C++ library with no mandatory external dependencies, supporting CPU inference with extensive SIMD optimizations (ARM NEON, x86 SSE/AVX/AVX512, RISC-V Vector, MIPS MSA, LoongArch LSX/LASX) and GPU inference via Vulkan. The framework includes a comprehensive set of neural network layers (convolution, pooling, normalization, activation, recurrent, attention, etc.) and supports multiple data types (float32, float16, bfloat16, int8). The companion pnnx tool converts models from PyTorch (TorchScript) and ONNX into ncnn format using a sophisticated graph rewriting pipeline. Python bindings and a model zoo are provided for rapid prototyping. The codebase is production-grade, used in Tencent applications like QQ and WeChat, and is actively maintained with support for modern architectures and operations (e.g., scaled dot-product attention, rotary embeddings).
clpeak is a cross-platform synthetic micro-benchmark that measures peak achievable compute performance (FLOPS/TOPS), bandwidth, latency, and AI inference throughput across CPUs, GPUs, and NPUs. It supports eight hardware backends: OpenCL, Vulkan, CUDA, ROCm/HIP, Metal, oneAPI/SYCL, native CPU (with per-ISA dispatch), and ONNX Runtime. The tool is architected as a set of backend-specific static libraries orchestrated by a common framework, with two front-ends: a CLI tool (`clpeak`) and a Flutter desktop/mobile GUI. The GUI communicates with the native library via a C-ABI FFI bridge, streaming structured events for live display. Results are serialized to a versioned JSON schema (v3) and can be compared against baselines. The codebase is notable for its meticulous anti-optimization engineering (MAD chains, compiler barriers, calibration probes) and its protobuf-free ONNX model generator. It is a mature, actively maintained project with extensive CI coverage and packaging for multiple platforms.
Glad is a multi-language code generator that produces loaders for Khronos graphics APIs (OpenGL, Vulkan, EGL, GLX, WGL). It parses official XML specification registries and generates C/C++ and Rust source files that load all required function pointers at runtime, supporting multiple contexts (MX), header-only builds, debug callbacks, on-demand loading, and alias resolution. The architecture is modular: a Python CLI dispatches to language-specific generators via a plugin system, using Jinja2 templates for output. It includes CMake integration, extensive compile/run tests, and examples for GLFW, SDL2/3, Vulkan, and embedded contexts. The codebase demonstrates strong engineering practices—clean separation of parsing, generation, and template rendering, with reusable patterns for configuration, plugin discovery, and XML processing.
NVRHI is a C++ hardware abstraction layer for real-time GPU programming. It exposes a single vendor-neutral RHI API (defined in include/nvrhi/nvrhi.h) and implements it through D3D11, D3D12, and Vulkan backends. Applications interact with interfaces such as IDevice, ICommandList, IBuffer, ITexture, and pipeline objects, while each backend translates those calls into native graphics API operations. The library is not an application or service; its primary entry points are device factory functions like nvrhi::d3d12::createDevice, nvrhi::d3d11::createDevice, and nvrhi::vulkan::createDevice, plus an optional validation layer created through nvrhi::validation::createValidationLayer.
Lemonade is a local AI inference server and desktop application that provides OpenAI/Anthropic/Ollama-compatible APIs with support for multiple backends (llama.cpp, vLLM, whisper.cpp, Kokoro, etc.). The C++ server core handles model lifecycle, routing policies, multi-modal inference, and telemetry, while the React/TypeScript frontend (Tauri desktop app and web build) offers a full GUI for chat, image/audio generation, transcription, and model management. The repository also includes a CLI client, tray application, extensive integration tests, and automated documentation generation. Architecture follows a clean backend-descriptor registry pattern with generated boilerplate, platform abstraction for process/archive/path operations, and a sophisticated routing policy engine for model selection.
Granite is a Vulkan-based real-time 3D rendering engine and application framework developed by Hans-Kristian Arntzen. It provides a comprehensive, modular library for building interactive graphics applications, including a high-level Vulkan abstraction, a declarative render graph with automatic resource management, scene loading (glTF, custom JSON), a full post-processing pipeline (TAA, SMAA, FXAA, HDR, bloom, SSAO, SSR), clustered and volumetric lighting, mesh shaders, and GPU-driven rendering. The framework also includes cross-platform audio (WASAPI, PulseAudio, Oboe) with DSP, video encoding/decoding via FFmpeg, a physics system (Bullet), a UI widget system, an entity-component system, and a flexible filesystem abstraction with network VFS support. The codebase is production-quality with advanced Vulkan feature usage, extensive tests, and tools for asset processing. While the author explicitly discourages external contributions, the repository remains actively maintained and showcases high craftsmanship in graphics engineering.
Cemu is a cross-platform Wii U emulator written in C/C++. It emulates the PowerPC-based Espresso CPU and Latte GPU via a dynamic recompiler (IML) and interpreter, and provides high-level emulation (HLE) of Wii U operating system libraries (coreinit, GX2, nn_*, etc.). The architecture is a monolithic desktop application with a wxWidgets GUI, multiple graphics backends (Vulkan, OpenGL, Metal), audio backends (Cubeb, XAudio2, DirectSound), and an abstract input layer supporting various controllers. The core emulation loop loads Wii U game images (WUD, WUX, WUHB), parses RPL modules, and translates PowerPC instructions into native code while rendering graphics via Vulkan/OpenGL/Metal. The codebase also includes a GDB stub for debugging, per-title game profiles, graphic packs for patches, and network service stubs (Miiverse, NIM). It is actively maintained with regular commits and vibrant community contributions.
LÖVE (love2d/love) is a cross-platform 2D game engine and framework. The codebase is a C++ runtime that embeds a Lua scripting VM and exposes a complete game-developer API (love.graphics, love.audio, love.physics, love.filesystem, love.math, love.window, etc.) used to build and ship games. The main executable (src/love.cpp) bootstraps a Lua state, loads src/modules/love/boot.lua, and runs the standard game loop defined in callbacks.lua (love.load/update/draw). In production this engine packages .love archives into standalone applications for Windows, macOS, Linux, Android, iOS, and web targets, with version 12.0 "Bestest Friend" currently defined in src/common/version.h. Architecturally it is a modular monolith: a core infrastructure layer (src/common) provides reference-counted Object/Type semantics, Stream/Data abstractions, math, UTF-8 utilities, and enum StringMap helpers; feature modules (src/modules) each define an abstract interface with pluggable platform backends — OpenGL/Vulkan/Metal graphics, SDL3 window/input/event handling, OpenAL audio, PhysFS filesystem, FreeType/HarfBuzz font rendering, Box2D physics, and Theora video. Lua bindings are hand-written via a consistent luax_* convention in wrap_*.cpp files. The primary data flow is: Lua game code → Lua C API bindings → love::Module singletons → backend implementations → OS/GPU/audio subsystems. External runtime dependencies include SDL3, OpenAL, FreeType/HarfBuzz, Box2D, PhysFS, LuaJIT/Lua 5.1/5.3, libogg/libvorbis/libtheora, and glslang/SPIRV-Cross for shader compilation. Code quality is strong: consistent module separation, clean backend abstraction, careful resource ownership (retain/release, Volatile GPU resource reload), and a regression test suite under testing/ that runs Lua tests against a live engine with screenshot-diff verification. Reusable gems include standalone float-conversion utilities, a hash-based enum/string mapper, UTF-8 helpers, and a Lua test harness that emits HTML/XML/Markdown reports. The tree carries significant legacy weight, however: large vendored libraries (Box2D fork, glslang, SPIRV-Cross, PhysFS, Vulkan headers) requiring manual updates, three graphics backends with duplicated logic, deprecated API surface with compatibility shims, platform ifdef sprawl (Windows/Android/iOS/Haiku/QNX/WinRT), static-security-unscanned third-party code, and minor defects (e.g., an uninitialized bool in the WinINet client). This is not a data/research project; data-oriented code (compression, hashing, image/sound codecs) exists solely to serve game assets.
shadPS4 is a PlayStation 4 emulator for Windows, Linux, and macOS, written in C++20. It emulates the PS4's AMD GCN GPU via a custom shader recompiler that translates GCN instructions to SPIR-V, and implements high-level emulation (HLE) of PS4 system libraries (kernel, libc, networking, audio, video, input, etc.). The architecture is monolithic with modular subsystems: a Vulkan renderer with texture/buffer caches, a shader recompiler with IR optimization passes, a file system abstraction supporting host files and ZArchive, and an ImGui-based debug UI. While many library functions are stubbed or partially implemented, the core GPU and CPU emulation is functional enough to run several commercial games. The codebase demonstrates strong engineering practices with reusable common utilities (bitfields, slot containers, thread-safe queues) and a well-structured shader compilation pipeline. It is a hybrid project, combining software application development with significant reverse-engineering and data analysis of PS4 firmware and GPU architecture.
MangoHud is a Linux-first, in-process performance monitoring overlay for Vulkan and OpenGL games. It injects a HUD that shows FPS, frame times, CPU/GPU load, temperatures, clocks, power, VRAM/RAM usage, I/O, network, battery, and media player metadata. The code implements this through multiple low-level integration paths: a Vulkan implicit layer, OpenGL/EGL/GLX swapchain hooks, LD_PRELOAD shims, a gamescope companion app (mangoapp), and a Windows DLL path that uses kiero to hook D3D11/D3D12 and DXGI. Hardware metrics are collected from Linux sysfs/proc interfaces, vendor libraries like NVML and XNVCtrl, and vendor-specific DRM/hwmon backends for AMD, Intel, NVIDIA, Panfrost/Panthor, and Qualcomm GPUs.
B3DFramework (namespace `b3d`, legacy `bsf` path prefixes) is a cross-platform C++17 game/application engine framework. Production code reveals a complete runtime: a scene/component model with ECS fragments, dirty-flag transform hierarchy propagation, prefab/scene serialization, async resource handles, and a plugin architecture with GPU backends (Vulkan, D3D12, Metal, Null), physics (PhysX), audio (FMOD, OpenAL), and a default renderer (RenderBeast). The engine implements its own shader language (BSL) with a compiling pipeline, a custom RTTI/reflection system with versioned binary serialization and delta compression, and a Mono-based C# scripting layer generated from C++ types via a codegen toolchain.
GPUReshape is a GPU debugging and instrumentation tool for DirectX 12 and Vulkan APIs. It provides a modular desktop application with backend layers for API interception, shader analysis and transformation capabilities, resource tracking and validation systems, and a plugin-extensible architecture built with Avalonia UI. The system includes automated testing infrastructure and message-passing communication between components.
OGRE (Object-Oriented Graphics Rendering Engine) is a mature, production-grade 3D rendering engine written in C++ with extensive cross-platform support. The codebase implements a modular plugin architecture with render system backends for Direct3D 9/11, OpenGL (legacy, 3+, ES2), Metal, Vulkan, and a software Tiny renderer. Core modules include scene management, material/shaders, resource handling, animation, particle systems, and compositor framework. Optional components provide real-time shader generation (RTShaderSystem), automatic mesh LOD generation, terrain rendering, volume rendering, world paging, and a 2D overlay system. The repository also includes Python bindings, sample applications, fuzz testing, and content pipeline tools (exporters for Maya, Milkshape, Wings3D). The architecture follows design patterns such as Singleton managers, Factory-based extensibility, and a work queue for background loading. Despite its age and size, the engine remains actively maintained with recent commits, CI/CD pipelines, and support for modern graphics APIs. Notable technical debt includes legacy render backends, deprecated APIs, and some tightly coupled subsystems.
MethaneKit is a cross-platform graphics rendering framework that provides abstractions over DirectX 12, Vulkan, and Metal APIs. It features a layered architecture with platform-independent interfaces (RHI), platform-specific implementations, an application framework, UI components, and mathematical utilities. The framework includes tutorial applications demonstrating various graphics techniques from basic triangle rendering to compute shaders and parallel rendering. Key systems include resource management, command submission, pipeline state handling, and synchronization primitives. The codebase demonstrates modern C++ practices with extensive use of templates, PIMPL idiom, and interface-based design.
This repository implements a Vulkan layer, CLI tool, and Qt UI for real-time frame generation using compute shaders. The core mechanism intercepts Vulkan swapchain creation and presentation in any Vulkan application, then injects generated frames via a separate Vulkan device. Shaders are extracted at runtime from the proprietary Lossless Scaling DLL and patched for the target format and device capabilities. The project is composed of four main components: a backend library (lsfg-vk-backend) with SPIR-V extraction and shader chain execution; a Vulkan layer (lsfg-vk-layer) that hooks instance/device/swapchain creation; a CLI tool (lsfg-vk-cli) for headless benchmarking and validation; and a Qt6 Quick UI (lsfg-vk-ui) for graphical profile management. The common library provides reusable Vulkan wrappers (buffer, image, shader, fences, descriptors) and a TOML-based configuration system with hot-reload support. The architecture is modular but tightly coupled to the proprietary DLL and the Vulkan layer loading mechanism.
vkbind is a Vulkan API loader and code generator that creates a single header file containing all Vulkan API definitions and functions for dynamically loading Vulkan functions. It parses the official Vulkan XML specifications (vk.xml and video.xml) to generate a comprehensive C header that eliminates the need for the official Vulkan SDK. The generated header includes type definitions, enumerants, function prototypes, and loader functions like vkInit(), vkInitInstanceAPI(), and vkBindAPI() for accessing Vulkan functionality across different platforms.
Yave is a C++20 Vulkan-based real-time graphics engine paired with a custom ImGui-driven editor and a self-contained core utility library named `y`. The repository is structured as three layers: `y` provides reusable containers, math, concurrency, IO, serialization, and a test framework; `yave` implements the engine itself (Vulkan device/window management, frame graph, render passes, ECS, asset pipeline, physics, animation, and scripting); and `editor` is a full 3D scene editor application with workspaces, widgets, gizmos, importers, thumbnails, and undo/redo. The renderer supports deferred PBR, cascaded shadow maps, TAA, bloom, SSAO/RTAO, RTGI, atmosphere scattering, and tone mapping, and it runs through a declarative frame graph with automatic resource pooling and barrier insertion. The editor is not a separate toy: it provides asset browsing, glTF/image import, material editing, scene outliner, inspector, file system views, performance metrics, GPU profiling, and RenderDoc integration. The engine uses a modern ECS with component groups, async asset pointers, and a serialization framework that supports schema evolution. The `y` library is particularly notable for its clean, high-performance primitives: SSO strings, flat hash maps with SIMD probing, sparse component sets, dependency-aware job system, and a forward/backward-compatible binary serialization layer. Externally, it depends on Vulkan, VMA, Volk, Dear ImGui, JoltPhysics, LuaJIT, tinygltf, stb, Tracy, and spirv_reflect, mostly vendored. While the project is described by its author as a learning/pet engine, the code is generally well-structured and professionally organized. However, it is a single-author codebase with many `Y_TODO` markers, some stubbed platform paths, global device singletons, and an explicit compiler-dependent serialization format. Tests exist mainly for the `y` core containers and math, not for the engine or editor. The repository is active and uses current C++/Vulkan practices, but should be treated as an evolving research-quality engine, not a stable library.
Alimer is a modern cross-platform .NET 10 game engine featuring a modular architecture with Entity-Component-System (ECS) design, multi-backend graphics rendering (Vulkan, Direct3D 12, Metal), physics integration, audio subsystem, and asset pipeline. The engine provides a complete framework for game development with systems for rendering, physics, audio, input, and scene management. It follows a layered architecture where platform-specific implementations are abstracted behind clean interfaces, allowing developers to build games that run on Windows, Linux, and macOS. The codebase demonstrates strong engineering practices with well-defined boundaries between systems, extensible asset importing, and backend-agnostic graphics abstractions.