On the Benefits of Speculative Trace Scheduling in VLIW Processors

VLIW processors are statically scheduled processors and their performance depends on the quality of the compiler’s scheduler. We propose a scheduling scheme where the application is first divided into decision trees and then further split into traces. We have developed a tool “SpliTree” to generate traces automatically. Using dynamic branch prediction for selecting the root of the decision tree from which the traces are scheduled using our scheme, we obtain approximately 1.4x performance improvement over that using decision trees for Spec92int benchmarks simulated on TriMedia processor.

[1]  Edward S. Davidson,et al.  Highly concurrent scalar processing , 1986, ISCA 1986.

[2]  Jan Hoogerbrugge Dynamic branch prediction for a VLIW processor , 2000, Proceedings 2000 International Conference on Parallel Architectures and Compilation Techniques (Cat. No.PR00622).

[3]  John R. Ellis,et al.  Bulldog: A Compiler for VLIW Architectures , 1986 .

[4]  Scott Mahlke,et al.  Effective compiler support for predicated execution using the hyperblock , 1992, MICRO 1992.

[5]  Mayan Moudgill,et al.  Scalable instruction-level parallelism through tree-instructions , 1997, ICS '97.

[6]  Scott A. Mahlke,et al.  Effective compiler support for predicated execution using the hyperblock , 1992, MICRO 25.

[7]  Thomas M. Conte,et al.  Treegion Scheduling for Highly Parallel Processors , 1997, Euro-Par.

[8]  Lex Augusteijn,et al.  Instruction Scheduling for TriMedia , 1999, J. Instr. Level Parallelism.