histogram/doc/fill_performance.py
Hans Dembinski 92a873c746
Faster linearize (#230)
* introduce offset for faster linearization of non-growing axes
* added traits::is_inclusive and constexpr bool inclusive() methods for builtin axes
* bug fixes for fill method when weight array is used with non-inclusive axis and when growing axes are used
* bug fix of axis::options::test(...)
* coverage tested with gcc-8, updated CONTRIBUTING.md with coverage info
2019-10-06 23:03:45 +02:00

119 lines
3.4 KiB
Python
Executable File

#!/usr/bin/env python3
# Copyright Hans Dembinski 2018 - 2019.
# Distributed under the Boost Software License, Version 1.0.
# (See accompanying file LICENSE_1_0.txt or copy at
# https://www.boost.org/LICENSE_1_0.txt)
import os
import numpy as np
import glob
import re
import json
import sys
from collections import defaultdict, OrderedDict
from matplotlib.patches import Rectangle
from matplotlib.lines import Line2D
from matplotlib.text import Text
from matplotlib.font_manager import FontProperties
import matplotlib.pyplot as plt
import matplotlib as mpl
mpl.rcParams.update(mpl.rcParamsDefault)
cpu_frequency = 0
data = defaultdict(lambda: [])
for fn in sys.argv[1:]:
d = json.load(open(fn))
cpu_frequency = d["context"]["mhz_per_cpu"]
for bench in d["benchmarks"]:
name = bench["name"]
time = min(bench["cpu_time"], bench["real_time"])
m = re.match("fill_(n_)?([0-9])d<([^>]+)>", name)
if m.group(1):
time /= 1 << 15
tags = m.group(3).split(", ")
dim = int(m.group(2))
label = re.search(
"fill_([a-z]+)", os.path.splitext(os.path.split(fn)[1])[0]
).group(1)
dist = tags[0]
if len(tags) > 1 and tags[1] in ("dynamic_tag", "static_tag"):
if len(tags) == 3 and "DStore" in tags[2]:
continue
label += "-" + {"dynamic_tag": "dyn", "static_tag": "sta"}[tags[1]]
label += "-fill" if m.group(1) else "-call"
data[dim].append((label, dist, time / dim))
time_per_cycle_in_ns = 1.0 / (cpu_frequency * 1e6) / 1e-9
plt.figure(figsize=(7, 6))
i = 0
for dim in sorted(data):
v = data[dim]
labels = OrderedDict()
for label, dist, time in v:
if label in labels:
labels[label][dist] = time / time_per_cycle_in_ns
else:
labels[label] = {dist: time / time_per_cycle_in_ns}
j = 0
for label, d in labels.items():
t1 = d["uniform"]
t2 = d["normal"]
i -= 1
z = float(j) / len(labels)
col = (1.0 - z) * np.array((1.0, 0.0, 0.0)) + z * np.array((1.0, 1.0, 0.0))
if label == "root":
col = "k"
label = "ROOT 6"
if "numpy" in label:
col = "0.6"
if "gsl" in label:
col = "0.3"
label = "GSL"
tmin = min(t1, t2)
tmax = max(t1, t2)
r1 = Rectangle((0, i), tmax, 1, facecolor=col)
r2 = Rectangle(
(tmin, i), tmax - tmin, 1, facecolor="none", edgecolor="w", hatch="//////"
)
plt.gca().add_artist(r1)
plt.gca().add_artist(r2)
font = FontProperties(size=9)
tx = Text(
-0.5,
i + 0.5,
"%s" % label,
fontproperties=font,
va="center",
ha="right",
clip_on=False,
)
plt.gca().add_artist(tx)
j += 1
i -= 1
font = FontProperties()
font.set_weight("bold")
tx = Text(
-0.5,
i + 0.6,
"%iD" % dim,
fontproperties=font,
va="center",
ha="right",
clip_on=False,
)
plt.gca().add_artist(tx)
plt.ylim(0, i)
plt.xlim(0, 80)
plt.tick_params("y", left=False, labelleft=False)
plt.xlabel("average CPU cycles per random input value (smaller is better)")
plt.tight_layout()
plt.savefig("fill_performance.svg")
plt.show()