Real-World Benchmarking
Ever since different processor families have emerged, engineers have been looking for a way to compare processor performance. In 1997, the Embedded Microprocessor Benchmark Consortium (EEMBC) was established as a nonprofit organization with a goal to develop benchmarks that reflect the reality of today's processors and applications. These benchmarks have been developed by the member companies, among them all major processor manufacturers. The suite includes 31 benchmarks distributed among five application-specific areas, including the telecom benchmark, which reflects the kind of work DSPs typically perform.
Device scores are derived in three ways: First, as out-of-the-box C code, where you can apply any standard compiler switches without modifications to the EEMBC source code; second, as a C-optimized test where you can allow C-level optimizations through the inclusion of any legal ANSI C compiler constructs such as pragmas or intrinsics; and third as a "full fury" test, where all stops are pulled out and hand-coding is the essence.
The telecom benchmarks (http://www.eembc.org/benchmark/score/ scorereport.asp) involve a number of tests with ECL-certified scores. While you might have some difficulty evaluating all the data, EEMBC established the Telemark, an aggregated score that encapsulates the results of the individual benchmarks. These are the figures that appear in Figure 3.
J.B. and M.L.