Update: This follows my post on AMD shipping Loom v0.1 for local AI on its CPUs, GPUs, and NPUs. Loom 0.2 is out, Phoronix reports. AMD describes 0.2.0 as expanding complete-program C++ authoring, runtime-provided managed objects, stateful finite services, and packaged AMD GPU, NPU, and Vulkan execution. The shared target compiler now stretches across x86, AIE2P, AMDGPU, SPIR-V, and WebAssembly, with work to cut fixed and superlinear costs in JIT compilation, graph construction, GPU launch and replay, pipeline planning, and long-context serving.
Tools, examples, serving, and hardware qualification now use one libamdf-backed AMD HAL through the existing --device=amdgpu spelling. The native path supports RDNA and CDNA execution and shared CPU, GPU, and NPU memory contracts. The former ROCr-backed driver and its legacy selector are removed, so this is not a drop-in assumption if you were on that older path. I have not installed 0.2.
Review before you install: read the release notes yourself, back up your config, and try new versions somewhere you can afford to break first.
My take
v0.1 was the first stake in the ground. v0.2 looks like the one where C++ authoring and a single AMD HAL start to matter, if you are actually targeting AMD hardware locally. Read the notes, especially if you were on the legacy driver path, and try it somewhere you can afford to break first.
God Bless, and May the Source Be with You!