I honestly don’t know what AMD is thinking here, only a few months after this being released and making decent progress even though actually matching CUDA was certain to be a multi-year endeavor.
They’re definitely not buying themselves any goodwill (it looks like a legal self-justifying hack job, the kind of thing that an org will do to avoid having hassles for their own sake) and it’s not as if this was endangering adoption of ROCm–that keeps sliding down the drain just fine on its own.