Optimal Integrated VLIW Code Generation with Integer Linear Programming

Abstract

We give an Integer Linear Programming (ILP) solution that fully integrates all steps of code generation, i.e. instruction selection, register allocation and instruction scheduling, on the basic block level for VLIW processors.In earlier work, we contributed a dynamic programming (DP) based method for optimal integrated code generation, implemented in our retargetable code generator OPTIMIST. In this paper we give first results to evaluate and compare our ILP formulation with our DP method on a VLIW processor. We also demonstrate how to precondition the ILP model by a heuristic relaxation of the DP method to improve ILP optimization time.

Modeling Universal Instruction Selection

“…These have subsequently been extended to DAG-based program graphs [17,18,25], but at the loss of optimality. Together with instruction scheduling or register allocation, the problem has also been approached using integer programming (IP) [7,21,36] and constraint programming (CP) [6,19,30]. Far fewer methods exist for global instruction selection, which so far only has been approached as a partitioned Boolean quadratic problem [9,15,16].…”

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].…”

Instruction Selection

“…In 2006, Bednarski and Kessler [43] developed an integrated code generation design where both pattern matching and pattern selection are solved using integer programming. The scheme-which later was applied by Eriksson et al [115], and is also described in an article by Eriksson and Kessler [116]-is an extension of their earlier work where instruction selection had previously more or less been ignored (see [199,200]).…”

“…These problems notwithstanding, there do exist several techniques that I believe have shown considerable promise-namely those based on methods for combinatorial optimization (see for example Wilson et al [340], Bashford and Leupers [40], Bednarski and Kessler [43], Floch et al [131], and Arslan and Kuchcinski [30]). …”