This update optimizes q2_0 decoding and small 32‑bit floating‑point matrix multiplications on Adreno 8xx GPUs via OpenCL enhancements.