Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Instruction-level performance modeling and characterization of multimedia applications

Conference ·
OSTI ID:350949
 [1];  [2]
  1. Los Alamos National Lab., NM (United States). Scientific Computing Group
  2. Louisiana State Univ., Baton Rouge, LA (United States). Dept. of Computer Science

One of the challenges for characterizing and modeling realistic multimedia applications is the lack of access to source codes. On-chip performance counters effectively resolve this problem by monitoring run-time behaviors at the instruction-level. This paper presents a novel technique of characterizing and modeling workloads at the instruction level for realistic multimedia applications using hardware performance counters. A variety of instruction counts are collected from some multimedia applications, such as RealPlayer, GSM Vocoder, MPEG encoder/decoder, and speech synthesizer. These instruction counts can be used to form a set of abstract characteristic parameters directly related to a processor`s architectural features. Based on microprocessor architectural constraints and these calculated abstract parameters, the architectural performance bottleneck for a specific application can be estimated. Meanwhile, the bottleneck estimation can provide suggestions about viable architectural/functional improvement for certain workloads. The biggest advantage of this new characterization technique is a better understanding of processor utilization efficiency and architectural bottleneck for each application. This technique also provides predictive insight of future architectural enhancements and their affect on current codes. In this paper the authors also attempt to model architectural effect on processor utilization without memory influence. They derive formulas for calculating CPI{sub 0}, CPI without memory effect, and they quantify utilization of architectural parameters. These equations are architecturally diagnostic and predictive in nature. Results provide promise in code characterization, and empirical/analytical modeling.

Research Organization:
Los Alamos National Lab., Scientific Computing Group, NM (United States)
Sponsoring Organization:
USDOE Assistant Secretary for Management and Administration, Washington, DC (United States)
DOE Contract Number:
W-7405-ENG-36
OSTI ID:
350949
Report Number(s):
LA-UR--99-303; CONF-990606--; ON: DE99002730
Country of Publication:
United States
Language:
English