https://hgpu.org/?p=2797
Scaling LAPACK panel operations using parallel cache assignment