This directory groups provider- and platform-specific execution design.

CUDA contains the current low-level buffer/runtime guides, active scheduler and provider designs, and GPU ecosystem surveys.
MPI contains the distributed persistent-object and dispatch design.

These directories mix current low-level foundations with planned heterogeneous execution. Each document identifies its status. The current working backend path is summarized by KernelDispatch" and @ref "/home/runner/work/uni20/uni20/docs/linalg/README.md" "Linear Algebra". @section autotoc_md821 Source Navigation - @ref "/home/runner/work/uni20/uni20/src/uni20/backend/README.md" "Provider wrapper source map" - @ref "/home/runner/work/uni20/uni20/src/uni20/linalg/backends/README.md" "Operation-tag backend source map"

There is no MPI source subtree yet; the MPI documentation is forward design.