Integer Linear Programming versus Dynamic Programming for Optimal Integrated VLIW Code Generation
Andrzej Bednarski and Christoph Kessler
To our knowledge there is only one Integer Linear Programming (ILP)
formulation in the literature that fully integrates all steps of
code generation, i.e. instruction selection, register
allocation and instruction scheduling, on the basic block level.
We give in this paper an improved version of this ILP formulation
that also covers VLIW processors.
Moreover, our ILP formulation does no longer require preprocessing
the basic block's data flow graph to support instruction selection.
Please contact our webadmin with any comments or changes.