GitHub - luxcpp/gpu: Lux GPU: An array framework for Apple and NVIDIA silicon.

Lux GPU Core

Lightweight plugin-based GPU acceleration library for blockchain and ML workloads.

Architecture

This is the core library only. It provides:

Stable ABI (backend_plugin.h) - Plugin contract
Plugin Loader - Dynamic loading of backend plugins
CPU Fallback - Builtin CPU backend for any platform
Tests - Backend-agnostic test harness

Backend plugins are built and distributed separately:

Plugin	Repo	Platform	Dependencies
Metal	`luxcpp/metal`	macOS arm64	MLX, Metal.framework
CUDA	`luxcpp/cuda`	Linux, Windows	CUDA Toolkit, CCCL
WebGPU	`luxcpp/webgpu`	All	Dawn/wgpu, gpu.cpp

Building

# Core only (CPU backend)
cmake -B build
cmake --build build

# Run tests
ctest --test-dir build

Usage

#include <lux/gpu.h>

int main() {
    // Initialize (loads best available backend)
    lux_gpu_init();

    // Or specify backend explicitly
    // lux_gpu_set_backend(LUX_BACKEND_CUDA);

    LuxContext* ctx = lux_gpu_create_context(-1);

    // Allocate and compute...
    LuxBuffer* buf = lux_gpu_alloc(ctx, 1024 * sizeof(float));

    lux_gpu_free(ctx, buf);
    lux_gpu_destroy_context(ctx);
    lux_gpu_shutdown();
}

Backend Selection

At runtime, backends are selected in priority order:

CUDA - If NVIDIA GPU detected
Metal - If macOS arm64
WebGPU - Cross-platform fallback
CPU - Final fallback (always available)

Override via environment or API:

export LUX_BACKEND=cuda  # or metal, webgpu, cpu

Plugin Loading

Backends are loaded from:

LUX_GPU_BACKEND_PATH environment variable
System library paths (/usr/lib/lux-gpu, etc.)
Relative to executable

Plugin naming: libluxgpu_backend_<name>.{so,dylib,dll}

ABI Stability

The plugin ABI is versioned. Plugins must match the core ABI version:

// backend_plugin.h
#define LUX_GPU_ABI_VERSION 1

License

BSD-3-Clause-Eco