Trace Scripts

Trace Profiling using Scripts

Using CCS with an XDS to capture trace data and then use post processing scripts (for profiling)

• CCS is used to setup trace configuration through trace control
• CCS can be used to setup trace jobs or AETLib can be used from the application.
• CCSv4 will automatically save the latest data collected in a .bin file (at ccsInstalDir\csv4\emulation\analysis\bin\logs).
• Use td.exe to decode the .bin file (at ccsInstalDir\ccsv4\emulation\analysis\bin) to either a text or csv file.
• td.exe has extensive capability to:

- Set configuration details; processor type, processor revision, what was used to collect the data (ETB or Trace Pod), timestamp format (abs, delta), ...

- Select trace data columns for output

- Where to start decoding as a % of the trace data from the start

- How many samples to decode

- Output can be to a file or stdout

- Format of of the output data selected

- Use "td -help" to get the complete list of command line arguments

• CCS supports Trace Job setup for the following TI scripts:

- Statistical Profiling

- Dynamic Thread Aware Call Graph

- Pipeline Stall Analysis

- Event Analysis

• Trace Jobs for scripts are selected from a list provided by trace properties actions (User Script) for Standard Trace. Expand the Script Type parameter to see a list of script specific parameters.
• The Trace Scripts can be downloaded from here.
• In cases where the CCS built-in script support, or normal trace jobs are not sufficient to provide the trace data needed, you may be able to use AETLib to setup trace jobs that can not be setup by UBM from CCS. In these cases AETLib will need to be integrated into your application. CCS, through Trace System Control, must still be used to configure and calibrate trace capture.

Instructions Per Cycle Calculation

A new trace script that was not included in the 3.x release of the Trace CSV Scripts has been created. This script, called trace_func_unit_efficiency.pl allows calculation of the scheduling efficiency of an application from output captured via trace. The script calculates the average number of instructions executed every cycle. This number is useful in power consumption calculations.

This script has not yet been integrated into a full release, so it, along with the necessary .PM module are provided here. If you already have an existing trace script package, you can extract these into the current directory. The .PM module provided is a new version that is backward compatible with existing scripts, but must be used for the trace_func_unit_efficiency script. If you don't want to merge these two sets, you can put the two included files into a separate directory. The Perl Script and PM module must reside in the same directory.

You can get the new script and module here