stepanov

System:

Power Macintosh 8500/180 = PowerPC 604e 180MHz, 256K L2

Compilers:

CW11 = Metrowerks CodeWarrior 11 -opt all -inline all -w off -opt schedule601 -d MAC -dialect cplus

MrC = Apple MrC 3.0.1f1c8 -opt speed -inline all

Libraries:

mathlib = Apple MathLib 2.0.3

powermathlib = Apple PowerMathLib 3.0d1

libmoto = Motorola libmoto provided with MotoTools 3.0

Introduction to the Stepanov Benchmark

The size of the data abstraction penalty is important to C++ programmers because if it is small enough, the C++ programmer can use the data abstraction features of C++ anywhere in a program, including within computational kernels. If the data abstraction penalty is too large, the C++ programmer will be forced, for performance reasons, to remove some or all of the data abstractions from the C++ program, and hence lose the benefits of data abstraction.

The Stepanov Benchmark is a collection of 13 kernels written in C++, each performing the same computation. Kernel zero is written at a low level, using no C++ data abstraction features. The other kernels use various C++ data abstraction features, in various combinations.

The Stepanov Benchmark reports results on a kernel by kernel basis by computing the ratio of this kernel's running time with the running time of kernel zero. The Stepanov Benchmark also reports the geometric mean of the kernel results as the abstraction penalty.

Detailed results

MrC 3.0 final

Metrowerks Codewarrior 11