Code Instruction Selection Based on SSA-Graphs

Abstract

Abstract. Instruction selection for embedded processors is a challenging problem. Embedded system architectures feature highly irregular instruction sets and complex data paths. Traditional code generation techniques have difficulties to fully utilize the features of such architectures and typically result in inefficient code. In this paper we describe an instruction selection technique that uses static single assignment graphs (SSA-graphs) as underlying data structure for selection. Patterns defined as graph grammar guide the instruction selection to find (nearly) optimal results. We present an approach which maps the pattern matching problem to a partitioned boolean quadratic optimization problem (PBQP). A linear PBQP solver computes optimal solutions for almost all nodes of a SSA-graph. We have implemented our approach in a production DSP compiler. Our experiments show that our approach achieves significant better results compared to classical tree matching.

Complete and Practical Universal Instruction Selection

“…Together with instruction scheduling and register allocation, instruction selection has also been approached using integer programming [8,25,54] and constraint programming [7,23,44]. Far fewer methods exist for global instruction selection, which so far only has been approached as a partitioned Boolean quadratic problem [11,18,19] or using constraint programming [29]. Common among these techniques is that they are restricted to tree-based and DAG-based patterns, whereas our approach can operate on full-fledged pattern graphs.…”

SSA-Based Register Allocation with PBQP

“…By backpropagating the reductions, the selection of the smaller instance can be extended to a selection of the original PBQP instance. Originally, there are four reductions [5,9,16]: RE: Independent edges have a cost matrix that can be decomposed into two vectors u and v, i.e. each matrix entry C ij has costs u i + v j .…”

Graph Rewriting for Hardware Dependent Program Optimizations

“…This works fine for all our test programs and covers most, but not all, dependencies between rich instructions. Therefore, our current research is to use a heuristic PBQP 4 -solver for rule selection [12,13].…”