Compilation and Scheduling Techniques for Embedded Systems

Embedded applications are constantly increasing in size, which has resulted in increasing demand on designers of digital signal processors (DSPs) to meet the tight memory, size and cost constraints. With this trend, memory requirement reduction through code compaction and variable coalescing techniques are gaining more ground. Also, as the current trend in complex embedded systems of using multiprocessor system-on-chip (MPSoC) grows, problems like mapping, memory management and scheduling are gaining more attention. The first part of the dissertation deals with problems related to digital signal processors. Most modern DSPs provide multiple address registers and a dedicated address generation unit (AGU) which performs address generation in parallel to instruction execution. A careful placement of variables in memory is important in decreasing the number of address arithmetic instructions leading to compact and efficient code. Chapters 2 and 3 present effective heuristics for the simple and the general offset assignment problems with variable coalescing. A solution based on simulated annealing is also presented. Chapter 4 presents an optimal integer linear programming (ILP) solution to the offset assignment problem with variable coalescing and operand permutation. A new approach to the general offset assignment problem is introduced. Chapter 5 presents an optimal ILP formulation and a genetic algorithm solution to the address register allocation problem (ARA) with code transformation techniques. The ARA problem is used to generate compact codes for array-intensive embedded applications. In the second part of the dissertation, we study problems related to MPSoCs. MPSoCs provide the flexibility to meet the performance requirements of multimedia applications while respecting the tight embedded system constraints. MPSoC-based embedded systems often employ software-managed memories called scratch-pad memories (SPM). Scheduling the tasks of an application on the processors and partitioning the available SPM budget among those processors are two critical issues in reducing the overall computation time. Traditionally, the step of task scheduling is applied separately from the memory partitioning step. Such a decoupled approach may miss better quality schedules. Chapters 6 and 7 present effective heuristics that integrate task allocation and SPM partitioning to further reduce the execution time of embedded applications for single and multi-application scenarios