Open-source AMDGCN kernels for optimizing LLM inference

Open-source AMDGCN kernels designed to optimize large language model inference performance. Provides low-level GPU optimization for AMD hardware running LLM workloads.