software-renderer

This project is a high-performance 3D rendering engine that implements the entire graphics pipeline on the CPU. It avoids all hardware acceleration APIs (OpenGL/Vulkan) to explore close-to-metal optimizations and data-oriented architectural patterns.

Engineering Highlights

• Multi-Threaded Rasterization: Utilizes dynamic thread pools with lock-free atomic counters for load-balanced spatial binning and rasterization.
• Data-Oriented Design: Implemented Structure of Arrays (SoA) for vertex and primitive data to maximize CPU cache efficiency and SIMD utilization.
• Aggressive Culling: Features object-level frustum culling and per-object light culling, reducing memory bandwidth from dozens of MBs down to a few KBs per frame.
• Custom Fragment Shading: Supports Blinn-Phong illumination with sequential-squaring power approximations to avoid expensive math library calls.

Performance Benchmarks

The engine demonstrates that software rendering is viable for complex scenes through optimized memory access and threading, achieving 120K+ triangles at 120 FPS while being illuminated with blinn-phong shading and multiple dynamic lights.