Solving the System Optimum Static Traffic Assignment Problem with Single Origin Destination Pair in Fuzzy Environment

An End-to-End Programming Model for AI Engine Architectures

“…This endeavor emphasized the integration of language and compiler design to furnish developers with the necessary tools for achieving performant implementations of dataflow programs such as GEMM. This work also yielded ILP and CP-SAT formulations of congestion-aware traffic assignment (Temelcan et al, 2020) for stream routing, a novel stream broadcasting primitive for combining array broadcasting and stream switch configuration and a novel approach to metaprogramming MLIR in Python.…”

“…-Key technical contribution of this work: A fully end-to-end programming model for AMD AI Engines, including a language frontend, optimal stream routing (using ILP and CP-SAT formulations of congestion-aware traffic assignment (Temelcan et al, 2020)), runtime memory management, and packaging, distribution, deployment to device; a novel stream broadcasting primitive for reducing the semantic gap between array broadcasting and stream switch configuration; a novel approach to metaprogramming MLIR in Python that enables using the same language for both metaprogramming and programming; finally, performant implementations of GEMM for the same architecture.…”