Profiler overhead distorting execution time for short functions

When using the MATLAB Profiler to identify performance bottlenecks, the tool records execution time and call counts for every function. However, for highly granular codebases consisting of very short, frequently called functions, the instrumentation overhead introduced by profile on can significantly inflate the reported time per call.

This creates a conflict when attempting to distinguish between a genuine algorithmic bottleneck and the measurement overhead of the Profiler itself. While timeit can be used for isolated micro-benchmarking, it does not provide the full call-stack context available in the Profiler viewer.

• How can a developer differentiate between actual function execution time and Profiler-induced overhead in a complex call tree?
• Is there a documented method to normalize Profiler results for functions with execution times near the instrumentation threshold?