We are pleased that our new paper describing AI-based register allocation in the compiler has already been presented at MOCAST 2024:
Fabian Stuckmann and Guillermo Payá Vayá: "A Graph Neural Network Approach to Improve List Scheduling Heuristics under Register Pressure"
The paper investigates the application of Graph Neural Networks (GNNs) to improve classical list scheduling heuristics in compiler optimization. List scheduling heuristics are techniques used to efficiently assign data to registers. However, these heuristics run into problems when it comes to register pressure - the situation where there is more data than available registers. Fabian presents a new approach that aims to improve decision making within these heuristics by using GNNs.