Statistical Profiling

From Texas Instruments Embedded Processors Wiki

Statistical Profiling is a method of profiling an application by taking samples of the Program Address at regular intervals and associating them with either a a)specific function, b) specific memory range. A simple statistical analysis is performed on the captured data in an effort to determine the areas where an application spends most of its time.

The advantage over traditional profiling is the amount of data necessary to be captured. Traditional profiling captures every executed PC address. Statistical profiling only captures a fraction of the entire execution path. Statistical Profiling does not provide cycle accurate results. It's merely a means to give a developer an overview of the CPU consumption by his application, and a handful of functions that should be targeted for optimization.

Beginning with CCS SR10, XDS560 Trace and the UBM plugin provide a means for capturing of statistical profiling data. In order to post process this data, you'll need the Trace CSV Scripts that can be obtained from the location below.

https://www-a.ti.com/downloads/sds_support/applications_packages/trace_csv_scripts/index.htm

Leave a Comment