cleaner docs, sim perf

README.md

@@ -3,6 +3,8 @@
 PPSFP (Parallel-Pattern Single-Fault Propagation) style fault simulators and other utility functions for use in various projects.
 These fault simulators are meant to be correct (ensured by unit-tests) and reasonably fast.
 
+Fault simulation has $O(n^3)$ complexity. Every *fault* is simulated with every *pattern* involving every combinational *gate* in the circuit. We measure performance in terms of `gates * faults * patterns / second`. How fast can we make it?
+
 ## Quick Start
 
 This project has submodules. To ensure everything is up-to-date, run
@@ -15,9 +17,15 @@ This project manages reproducible programming environments with:
 - [uv](https://docs.astral.sh/uv/) for managing python environments.
 - [nix](https://nixos.org) for managing non-python tools and benchmark designs. Follow [this guide](https://librelane.readthedocs.io/en/stable/installation/nix_installation/index.html) or [this guide](https://github.com/fossi-foundation/nix-eda/blob/main/docs/installation.md) to setup [nix-eda](https://github.com/fossi-foundation/nix-eda/tree/main) binary cache to avoid re-building EDA-related tools.
 
-Smoke-test: `uv run pytest`
+Run all unit-tests: `uv run pytest`
+
+Run a naive, baseline fault simulation with 1024 random patterns: `uv run main.py tests/c6288.bench`
 
-For running Jupyter Notebooks in the reproducible programming environment, install a kernelspec with this command:
+Same on an 18k-gate circuit: `uv run main.py polito-itc99-b15-sky130`
+
+`nix develop` makes [quaigh](https://github.com/coloquinte/quaigh) available, another simulator/atpg written in rust.
+
+To run Jupyter Notebooks in the reproducible programming environment, install a kernelspec:
 ```
 uv run ipython kernel install --user --env VIRTUAL_ENV $(pwd)/.venv --env PATH ${PATH} --name=uv-env
 ```
@@ -27,8 +35,7 @@ uv run jupyter lab
 ```
 Choose `uv-env` as kernel in Jupyter Lab.
 
-`uv run main.py <circuit>` runs a naive, baseline fault simulation.
 
-`nix develop` makes [quaigh](https://github.com/coloquinte/quaigh) available, another simulator/atpg written in rust.
+
 
 

main.py

@@ -2,6 +2,7 @@
 import argparse
 import subprocess
 from pathlib import Path
+import time
 
 import numpy as np
 
@@ -13,11 +14,11 @@ from fsim.static import LineRoles, FaultSet
 
 def main():
 
-    parser = argparse.ArgumentParser(description='Fault Stats')
+    parser = argparse.ArgumentParser(description='A basic stuck-at fault simulator.')
-    parser.add_argument('-t', '--tlib', default='SKY130', help=f'techlib of circuit. default: SKY130, available: {sorted(techlib_by_name.keys())}.')
+    parser.add_argument('-t', '--tlib', default='SKY130', help=f'Techlib for verilog circuit. Default: SKY130, available: {sorted(techlib_by_name.keys())}.')
-    parser.add_argument('-p', '--patterns', default=1024, help='Number of random patterns to simulate.')
+    parser.add_argument('-p', '--patterns', default=1024, help='Number of random patterns to simulate. Default: 1024.')
-    parser.add_argument('--seed', type=int, default=42, help='Random seed for reproducibility (default: 42).')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed for reproducibility. Default: 42.')
-    parser.add_argument('circuit', help='gate-level verilog file or nix package to import. See "nix flake show github:s-holst/benchmark-circuits" for available packages.')
+    parser.add_argument('circuit', help='Gate-level verilog, bench, or nix package to import. See available packages: "nix flake show github:s-holst/benchmark-circuits".')
     args = parser.parse_args()
     args.patterns = int(args.patterns)
     args.tlib = techlib_by_name[args.tlib]
@@ -75,6 +76,8 @@ def main():
     undetected = set()
     detected = set()
 
+    start_time = time.perf_counter()
+
     with log.progress() as p:
         for fidx, fault in enumerate(injection_faults):
             fault_site = fault//2
@@ -91,7 +94,11 @@ def main():
             else:
                 detected.add(fault)
 
-    log.info(f'detected by simulation: {len(detected)}/{len(injection_faults)}  -  {len(detected)/len(injection_faults)*100:.1f}%')
+    sim_time = time.perf_counter() - start_time
+    log.info(f'fsim time: {sim_time:.2f}s')
+    sim_performance = stats['comb'] * len(injection_faults) * patterns.shape[1] / sim_time
+    log.info(f'fsim performance: {sim_performance:.2e} gfp/s')
+    log.info(f'detected by simulation: {len(detected)}/{len(injection_faults)}  -  {len(detected)/len(injection_faults)*100:.2f}%')
 
 if __name__ == "__main__":
     main()