jitterplot

#!/usr/bin/env python3
# SPDX-License-Identifier: MIT

import os
import sys
import json
import argparse
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

__version__ = '0.3'

# silence SettingWithCopyWarning globally
# https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#evaluation-order-matters
pd.options.mode.chained_assignment = None


def plot_cdf(filename, outfilename):
    with open(filename) as file:
        rawdata = json.load(file)

    cpu_id = 0
    fig, ax = plt.subplots()
    while True:
        if str(cpu_id) not in rawdata['cpu']:
            break

        cid = str(cpu_id)
        data = rawdata['cpu'][cid]
        bins = [int(i) for i in data['histogram'].keys()]

        values = [int(i) for i in data['histogram'].values()]
        cumulative = np.cumsum(values) / np.sum(values)
        # show 99.9 and 99.99 percentiles in legend
        p3_idx = list(map(lambda i: i > 0.999, cumulative)).index(True)
        p4_idx = list(map(lambda i: i > 0.9999, cumulative)).index(True)
        p3 = bins[p3_idx]
        p4 = bins[p4_idx]
        pmax = bins[-1]
        ax.step(bins, cumulative,
                label='cpu{}: p99.9={}, p99.99={}, max={}'
                      .format(cid, p3, p4, pmax))

        cpu_id = cpu_id + 1

    L = ax.legend()
    plt.grid(color='lightgrey', linestyle='-', linewidth=1, which='both')
    plt.yticks(ticks=np.arange(0, 1.1, 0.1))
    plt.xlabel('jitter [us]')
    plt.ylabel('probability')

    plt.setp(L.texts, family='monospace')
    if outfilename is not None:
        plt.savefig(outfilename)
    plt.show()


def plot_histogram(filename, outfilename):
    with open(filename) as file:
        rawdata = json.load(file)

    cpu_id = 0
    fig, ax = plt.subplots()
    while True:
        if str(cpu_id) not in rawdata['cpu']:
            break

        cid = str(cpu_id)
        data = rawdata['cpu'][cid]
        d = {int(k): int(v) for k, v in data['histogram'].items()}
        lbl = 'cpu{} min{:>3} avg{:>7} max{:>3}'.format(
            cid,
            rawdata['cpu'][cid]['min'],
            rawdata['cpu'][cid]['avg'],
            rawdata['cpu'][cid]['max'])
        ax.bar(list(d.keys()), list(d.values()),
               log=True, alpha=0.5, label=lbl)

        cpu_id = cpu_id + 1

    L = ax.legend()
    plt.setp(L.texts, family='monospace')
    plt.xlabel('jitter [us]')
    plt.ylabel('frequency')
    if outfilename is not None:
        plt.savefig(outfilename)
    plt.show()


def load_samples(filename):
    dt = np.dtype([('CPUID', 'u4'),
                   ('Seconds', 'u8'),
                   ('Nanoseconds', 'u8'),
                   ('Value', 'u8')])
    data = np.fromfile(filename, dtype=dt)
    df = pd.DataFrame(data)
    return df


def plot_all_cpus(df, outfilename):
    ids = df["CPUID"].unique()
    max_jitter = max(df["Value"])

    fig = plt.figure()
    axes = fig.subplots(len(ids))
    for ax, data in zip(
        iter(axes),
        (df[df["CPUID"] == id] for id in ids),
    ):
        data["Time"] = data["Seconds"] + data["Nanoseconds"] * 10**-9
        ax.plot("Time", "Value", data=data)
        ax.set_xlabel("Time [s]")
        ax.set_ylabel("Latency [us]")
        ax.set_ylim(bottom=0, top=max_jitter)
    if outfilename is not None:
        plt.savefig(outfilename)
    plt.show()


def main():
    ap = argparse.ArgumentParser(
        description='Plot statistics collected with jitterdebugger')
    ap.add_argument('--version', action='version',
                    version='%(prog)s ' + __version__)
    ap.add_argument('--output', help='output file name to save figure',
                    default=None, action='store', type=str)
    sap = ap.add_subparsers(dest='cmd')

    hrs = sap.add_parser('hist', help='Print historgram')
    hrs.add_argument('HIST_FILE')

    crs = sap.add_parser('cdf', help='Plot cumulative densitiy function')
    crs.add_argument('CDF_FILE')

    srs = sap.add_parser('samples', help='Plot samples graph')
    srs.add_argument('SAMPLE_FILE')

    args = ap.parse_args(sys.argv[1:])
    if args.cmd == 'hist':
        fname = args.HIST_FILE
        if os.path.isdir(fname):
            fname = fname + '/results.json'
        plot_histogram(fname, args.output)
    elif args.cmd == 'cdf':
        fname = args.CDF_FILE
        if os.path.isdir(fname):
            fname = fname + '/results.json'
        plot_cdf(fname, args.output)
    elif args.cmd == 'samples':
        fname = args.SAMPLE_FILE
        if os.path.isdir(fname):
            fname = fname + '/samples.raw'

        df = load_samples(fname)
        plot_all_cpus(df, args.output)


if __name__ == '__main__':
    main()