Custom C scripts to parse through large FSDB waveform files quickly. Build on top of Verdi FsdbReader library. Setup only works for BWRC.
Add necessary libraries to path and compile all binaries.
export LD_LIBRARY_PATH=/tools/synopsys/verdi/P-2019.06-SP2-12/share/FsdbReader/linux64:${LD_LIBRARY_PATH}
export LD_LIBRARY_PATH=/opt/rh/devtoolset-8/root/usr/lib64:${LD_LIBRARY_PATH}
make
This generates executables in the root directory.
First, we need to print out the FSDB waveform "header" information, which lists all signal information in the design.
./print_header <waveform.fsdb> > header.txt
Next, we parse these headers to extract signal idcodes.
./scripts/get_idcodes.py <args>
From here, we can do a number of things.
// dump one signal's (time, value changes) to a file
./dump_signal -h
// get simulation end time (in ns)
./get_sim_endtime waveform.fsdb end_time.txt
// dump each signal's toggle activity per window of time (creates a matrix)
./dump_toggles -h
// dump each signal's ratio spent with value=1 per window of time (creates a matrix)
./dump_ones -h
See the flows/flow.ipynb Jupyter Notebook for an example flow that runs a workload on a Chipyard design in RTL simulation,
then parses the waveform and extracts the time-indexed instruction trace.
The flow assumes you run the commands printed by cells marked with a # RUN THIS comment before advancing to the next cell.