Please tell me you published your findings even as an issue on the llama.cpp GitHub
If you read the code you would realize that the issue is....
in amdkfd and hsakmt
Yes, that means AMD is sitting on both sides. They wrote software that doesn't work with their own software.
he is still closing his jaw
Spoiler alert: the problem didn't actually get fixed despite the jaw on the floor.
Here they are: https://gist.github.com/birep/6f2c8d490c7a29820997d57bd654c3...