Acceleration

Hardware acceleration recipes that optimize inference throughput through autotuning, kernel fusion, memory mapping, and quantized arithmetic. These examples demonstrate low-level performance techniques beyond SIMD and GPU categories.