Software Efficiency Reviews

The Golden Standard

CodeSprite can, and often does optimize software supplied as an isolated function in a source code file. We'll build the code and analyze its behaviour, then look for performance bottlenecks or opportunities to reduce code size.

The most valuable asset when optimizing code, is a working implementation with a test harness that produces reference output for a representative range of inputs. It doesn't matter how slow or large this golden standard code is, it will save a great deal of time proving that the optimized code is correct.

CodeSprite's record for bug-free optimization using testing by logical induction (i.e. it looks like it should work!) is excellent, but verification is far easier if we have reference outputs to check our implementation against.

Ideally we would have a way to build the function to be optimized in a test system with minimal module dependencies. We also prefer to have a file of data representing a reasonable subset of the possible input values including testing corner cases and transition points, and a corresponding file containing the expected outputs from the function.