Pi-Bench

A small benchmarking framework for the Raspberry Pi to collect accurate function timings.

• Uses macros for low overhead timing.
• Can automatically pin processes to CPUs and set priorities for single threaded applications.
• Timing data can be collected in cycles or microseconds.
• Collects L1 cache miss rates
• Allows for tracking the CPU temperature to avoid thermal throttling.
• Allows to set baselines to calculate and print relative performance.
• Supports automatic result validation
• Can save timing data & cache performance for each run to CSV

Installation

1. Clone the repository

Run git clone https://github.com/TomSchammo/pi-bench

2. Install the library

Run cd pi-bench && make install

Usage

1. Define your benchmarking parameters and declare optional ground-truth buffer

#define WARMUP_RUNS 0
#define TIMED_RUNS 1

static void *gt = NULL;

2. Define your benchmarks

The library uses the X-Macro pattern to automatically discover and collect benchmarks. So the BENCHMARKS macro needs to be defined before including the header

#define _GNU_SOURCE

#define BENCHMARKS                                                             \
  BENCHMARK_TIME_PINNED(                                                       \
      "Benchmark Name", true, false, output_buffer, BYTES_OUTPUT_BUFFER, 1,    \
      function(input, output))                                                 \

3. Include the utils header

#include <pi-bench/utils.h>

4. Run the benchmarks

RUN_BENCHMARKS();

5. Get the results

PRINT_RESULTS_INDIVIDUAL();

PRINT_RESULTS_GROUP();

SAVE("./results/");

6. Clean the environment

CLEANUP();

TODO

• Add example main.c
• Add a way to kill in-progress benchmarks