Execution characteristics of SPEC CPU2000 benchmarks: Intel C++ vs. Microsoft VC++

ABSTRACT

Modern processors include features such as deep pipelining, multi-level cache hierarchy, branch predictors, out of order execution engine, and advanced floating point and multimedia units. To successfully exploit these features, architecture-aware compilers that can produce target-specific optimal codes for the applications are needed. Using the knowledge about the architectural features, the compilers can contribute to maximizing the application performance through effective pipeline scheduling, memory penalty minimization and path length reduction. A study of the execution characteristics for the binaries generated by the various compilers can provide insights about the effectiveness of the optimization options used in the compilers. The in-built performance monitoring hardware found in present day processors can be used to collect the performance metrics for the study of execution characteristics. In this paper, we compare the Intel C++ and Microsoft VC++ compilers by studying the execution characteristics of SPEC CPU 2000 benchmarks run on a Pentium IV processor. The benchmarks were compiled with identical optimization switches in both compilers and the performance metrics were collected using Intel's VTune Performance Analyzer. The analyses of results showed that the Intel C++ compiler performed better than VC++ for all considered applications and significantly better for computer visualization and graphical applications.

REFERENCES

• 1: Talla, D., and John, L. K. Execution Characteristics of Multimedia Applications on a Pentium II Processor. In Proceedings of the IEEE International Conference on Performance, Computing and Communications, (IPCC '00), (February 20--22, 2000), 516--524.
• 2: Mehis, A., Ali, R., and Radhakrishnan, R. Using Processor-Specific Optimizations to Maximize Performance on Dell Servers, Technical Article on PowerSolutions, August 2002. www.dell.com/powersolutions
• 3: Milena Milenkovic , Aleksandar Milenkovic , Jeffrey Kulick, Microbenchmarks for determining branch predictor organization, Software—Practice & Experience, v.34 n.5, p.465-487, April 2004 [doi>10.1002/spe.v34:5]
• 4: Intel VTune Performance Analyzer, www.intel.com/software/products/vtune/
• 5: Katherine E. Stewart , Steven W. White, The Effects of Compiler Options on Application Performance, Proceedings of the1994 IEEE International Conference on Computer Design: VLSI in Computer & Processors, p.340-343, October 10-12, 1994
• 6: Dileep Bhandarkar , Jason Ding, Performance Characterization of the Pentium® Pro Processor, Proceedings of the 3rd IEEE Symposium on High-Performance Computer Architecture, p.288, February 01-05, 1997
• 7: John L. Hennessy , David A. Patterson, Computer Architecture: A Quantitative Approach, Morgan Kaufmann Publishers Inc., San Francisco, CA, 2003
• 8: Brinkley Sprunt, Pentium 4 Performance-Monitoring Features, IEEE Micro, v.22 n.4, p.72-82, July 2002 [doi>10.1109/MM.2002.1028478]
• 9: MSDN Help for Microsoft VC++ Compiler
• 10: Intel C++ Compiler User Guide, www.intel.com/software/products/compilers/techtopics/ccug. htm
• 11: SPEC CPU2000 Benchmark suite, http://www.spec.org/cpu2000/

INDEX TERMS

General Terms:
Experimentation, Measurement, Performance

Keywords:
SPEC CPU2000 benchmarks, compiler optimizations, event-based sampling, performance evaluation