llama.cpp

84778e97 - CUDA/HIP: Share the same unified memory allocation logic. (#12934)

CUDA/HIP: Share the same unified memory allocation logic. (#12934) Replace compile-time `GGML_HIP_UMA` with environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY`. This unifies the usage on NVIDIA and AMD GPUs, and allows a single binary to be shared between integrated and dedicated GPUs.