APEX Compute: A Zero-Dependency C++20 Engine for Vectorized Matrix Multiplication
APEX Compute is a C++20 engine built from scratch with no external dependencies, designed to run cache-tiled, explicitly vectorized matrix multiplication (GEMM) on both ARM and x86 processors. The project pairs its low-level SIMD kernels with an automated benchmarking tool, apex-bench, and a differential validation suite, apex-validate, to check results for correctness. It also offers an optional natural-language layer powered by Gemma 4, which converts a user's stated intent into a structured hardware run plan. The project's source code is available in its public repository.

