Script Documentation
March 21, 2026 ยท View on GitHub
This directory contains some scripts that were used to generate slurm jobs to evaluate all competing prefetchers, as well as rolling up statistics from output files.
Creating Slurm Jobs
create_jobs.py creates slurm jobs to run a list of experiments on a list of traces. The script supports the following commandline arguments:
| Argument | Description | Default Value |
|---|---|---|
tlist | A list of traces as a CSV file. These traces will be used to run the experiments. dpc.1c.csv is an example for single-core traces, dpc.4c.csv is an example for four-core traces. | "" |
exp | A list of experiments as a CSV file. Each experiment runs on all traces. exp.csv is an example of list of experiments. Line starting with # gets ignored. | "" |
out | Name of the output jobfile. | jobfile.sh |
Rolling-Up Statistics
rollup_csv.py parses a list of output file, extracts a list of metrics, and dumps into a CSV format. The script supports the following commandline arguments.
| Argument | Description | Default Value |
|---|---|---|
tlist | A list of traces as a CSV file. These traces will be used to rollup stats. dpc.1c.csv is an example for single-core traces, dpc.4c.csv is an example for four-core traces. | "" |
exp | A list of experiments as a CSV file. exp.csv is an example of list of experiments. Line starting with # gets ignored. | "" |
mfile | A list of metrics to rollup. perf.mfile and coverage.mfile are example mfiles | "" |
ext | The extension of the stat file. Typically JSON. | JSON |
results-dir | The directory where all the output files are expected. | "" |
threads | The number of threads to launch. Useful when rolling up stats for a long list of experiments and traces. | 1 |