From f06e10639178becfc3fb3a5ce2e8a6ef66170cbe Mon Sep 17 00:00:00 2001
From: Guillaume Damiand <guillaume.damiand@liris.cnrs.fr>
Date: Tue, 25 May 2021 15:57:09 +0200
Subject: [PATCH] Add scripts; update readme
---
README.md | 94 +++++-
python_scripts/generate_data.py | 211 ++++++++++++
python_scripts/histogram_segment.py | 129 ++++++++
python_scripts/parselog.py | 22 ++
python_scripts/plot_result.py | 492 ++++++++++++++++++++++++++++
python_scripts/run_benchmark.py | 173 ++++++++++
6 files changed, 1120 insertions(+), 1 deletion(-)
create mode 100644 python_scripts/generate_data.py
create mode 100644 python_scripts/histogram_segment.py
create mode 100644 python_scripts/parselog.py
create mode 100644 python_scripts/plot_result.py
create mode 100644 python_scripts/run_benchmark.py
diff --git a/README.md b/README.md
index b138f5f..0c32658 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,95 @@
# stripped-hds
-Stripped halfedge data structure for parallel computation of arrangements of segments
\ No newline at end of file
+Stripped halfedge data structure for parallel computation of arrangements of segments.
+
+Let us denote by SRC_DIR the root directory of the gitlab repository.
+
+1. Each data is a text file containing a list of segments. In the gitlab repository, these files are compressed.
+ To uncompress all the data files:
+
+ # We suppose the current directory is SRC_DIR
+ gunzip -r data/
+
+2. Install required dependencies
+
+ Boost, with components system and filesystem
+ Intel TBB
+ Qt5
+ CGAL, with component Qt5
+
+3. Compile the programs
+
+ # We suppose the current directory is SRC_DIR
+ mkdir builds
+ cd builds
+ mkdir release
+ cd release
+ cmake -DCMAKE_BUILD_TYPE=Release ../../src
+ make
+
+4. Run the main program
+
+ One example of run:
+
+ # We suppose the current directory is SRC_DIR/builds/release
+ ./parallel-arrangement -t1 ../../data/mixerP4.txt -nbs 8 -nbt 8 -crop -draw
+
+ The parameter after -t1 is the filename
+ -nbs gives the number of strips
+ -nbt gives the number of thread
+ -crop is an option to crop segments in the strips
+ -draw will draw the final arrangement
+
+ Run
+ ./parallel-arrangement -h
+ for an help showing all options.
+
+5. You can use different scripts to reproduce results of the paper
+
+ 5.1 To compare computation times of CGAL and our method, for GIS dataset.
+
+ # We suppose the current directory is SRC_DIR
+ ./run-gis-test.sh 5
+
+ Run our program on all the gis data.
+ 5 is the number of runs for each file (in the paper we use 5 runs);
+ 1 is used by default.
+
+ Computation times of the method are stored in file times.dat;
+ number of cells in nbcells.dat; and times of the different parts
+ of our method in detailed-times.dat.
+
+ 5.2 To draw computation time graphs for stochastics dataset
+
+ You need python3, numpy, matplotlib
+
+ # We suppose the current directory is SRC_DIR
+ python3 python_scripts/run_benchmark.py --log outputmixerP4.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/mixerP4.txt
+
+ Run our program on the given dataset, and store the log output in a file.
+
+ --log gives the log filename
+ --thread gives the number of threads to test
+ --repeat gives the number of runs for each file (in the paper we use 5 runs)
+ --crop is an option to crop segments in the strips
+ last parameter is a directory of a list of files to test. When a directory is used,
+ the program is ran for each file in the directory.
+
+ In the paper we used:
+
+ python3 python_scripts/run_benchmark.py --log outputmixerP4.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/mixerP4.txt
+ python3 python_scripts/run_benchmark.py --log FRA.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/countries/FRA.txt
+ python3 python_scripts/run_benchmark.py --log gauss-short.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/gauss-short/
+ python3 python_scripts/run_benchmark.py --log gauss-long.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/gauss-long/
+ python3 python_scripts/run_benchmark.py --log best.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/best/
+ python3 python_scripts/run_benchmark.py --log worst.log --thread 1 2 4 8 16 32 --repeat 5 --crop data/worst/
+
+
+ To visualize the results, you need to use plot_result.py, for example:
+
+ python3 python_scripts/plot_result.py --no-load --no-error outputmixerP4.log
+ python3 python_scripts/plot_result.py --no-load --no-error FRA.log
+ python3 python_scripts/plot_result.py --no-load --no-error gauss-short.log
+ python3 python_scripts/plot_result.py --no-load --no-error gauss-long.log
+ python3 python_scripts/plot_result.py --no-load --no-error best.log
+ python3 python_scripts/plot_result.py --no-load --no-error worst.log
\ No newline at end of file
diff --git a/python_scripts/generate_data.py b/python_scripts/generate_data.py
new file mode 100644
index 0000000..25f99fa
--- /dev/null
+++ b/python_scripts/generate_data.py
@@ -0,0 +1,211 @@
+import argparse
+import itertools
+import os
+import subprocess
+import sys
+import time
+
+from enum import Enum
+
+RANDOM_SEGMENTS_BIN = "../src/builds/release/tools/random-segments"
+SHP_TO_TXT_BIN = "../src/builds/release/tools/shp-to-txt"
+
+class Profile(Enum):
+
+ GAUSS = "gauss"
+ RAND = "rand"
+ BEST = "best"
+ WORST = "worst"
+
+ def __str__(self):
+ return self.value
+
+ def __repr__(self):
+ return str(self)
+
+def bin_exist(path):
+
+ return os.path.isfile(path) and os.access(path, os.X_OK)
+
+def random_segments(xmin=0.0, xmax=10000.0, ymin=0.0, ymax=10000.0,
+ n=100, size=1000.0, error=0.05,
+ output="./segments.txt",
+ seed=None, fixed=False):
+
+ args = [
+ RANDOM_SEGMENTS_BIN,
+ "--xmin", str(xmin),
+ "--xmax", str(xmax),
+ "--ymin", str(ymin),
+ "--ymax", str(ymax),
+ "-n", str(n),
+ "--size", str(size),
+ "--error", str(error),
+ "--output", output
+ ]
+ if seed:
+ args.append("--seed")
+ args.append(str(seed))
+ if fixed:
+ args.append("--fixed ")
+
+ proc = subprocess.run(args, stdout=subprocess.PIPE)
+
+ return proc
+
+def shp_to_txt(profile, n,
+ band=None, size=None,
+ output=None):
+
+ args = [
+ SHP_TO_TXT_BIN
+ ]
+
+ if profile == Profile.RAND:
+
+ args.extend([ "-t2", str(n), str(size) ])
+ if size == 0.0:
+ filename = "random-%d.txt" % n
+ else:
+ filename = "random-%d-size-%.2f.txt" % (n, size)
+
+ elif profile == Profile.BEST:
+
+ args.extend([ "-t5", str(n), str(band), str(size) ])
+ if band == 1:
+ if size == 0.0:
+ filename = "random-%d.txt" % n
+ else:
+ filename = "random-%d-size-%.2f.txt" % (n, size)
+ else:
+ if size == 0.0:
+ filename = "random-%d-best-%d.txt" % (n, band)
+ else:
+ filename = "random-%d-size-%.2f-best-%d.txt" % (n, size, band)
+
+ elif profile == Profile.WORST:
+
+ args.extend([ "-t6", str(n) ])
+ filename = "random-%d-worst.txt" % (n)
+
+ else:
+ raise NameError("'" + profile + "' is not a valid profile for shp-to-txt.")
+
+ proc = subprocess.run(args, stdout=subprocess.PIPE)
+
+ if output:
+ os.rename(filename, output)
+
+ return proc
+
+if __name__ == "__main__":
+
+ parser = argparse.ArgumentParser()
+
+ parser.add_argument("--profile", "-p", required=True, type=Profile,
+ choices=list(Profile),
+ help="set generation profile")
+
+ parser.add_argument("--nseg", "-n", required=True, type=int, nargs="+",
+ help="set number of segments")
+ parser.add_argument("--size", "-s", required=True, type=float, nargs="+",
+ help="set size of segments")
+ parser.add_argument("--batch", "-b", required=True, type=int,
+ help="set number of segments generations")
+
+ parser.add_argument('--verbose', '-v', action='count', default=0,
+ help="set verbosity level")
+ parser.add_argument("--dir", "-d", default=None,
+ help="set generated data directory, default is cwd")
+
+ gaussg = parser.add_argument_group("gauss profile configuration")
+
+ gaussg.add_argument("--error", type=float, default=0.10,
+ help="set segment length error (=2*sigma/mu), default value is 0.10")
+ gaussg.add_argument("--fixed", action="store_true",
+ help="fix segment size")
+
+ bestgr = parser.add_argument_group("best profile configuration")
+
+ bestgr.add_argument("--band", type=int, nargs="+",
+ required=(str(Profile.BEST) in sys.argv),
+ help="set numbers of bands")
+
+ binary = parser.add_argument_group("binaries configuration")
+
+ binary.add_argument("--bin-gauss", default=RANDOM_SEGMENTS_BIN,
+ help="set 'random-segments' binary path, default value is set line 10")
+ binary.add_argument("--bin-shp", default=SHP_TO_TXT_BIN,
+ help="set 'shp-to-txt' binary path, default value is set line 11")
+
+ args = parser.parse_args()
+
+ RANDOM_SEGMENTS_BIN = args.bin_gauss
+ SHP_TO_TXT_BIN = args.bin_shp
+
+ if args.dir == None:
+ args.dir = os.getcwd()
+
+ if not bin_exist(RANDOM_SEGMENTS_BIN):
+ raise NameError("'" + RANDOM_SEGMENTS_BIN + "' random-segments path does not exist. "
+ "Set variable 'RANDOM_SEGMENTS_BIN' line 10 or option --bin-gauss to a correct path.")
+ if not bin_exist(SHP_TO_TXT_BIN):
+ raise NameError("'" + SHP_TO_TXT_BIN + "' shp-to-txt path does not exist. "
+ "Set variable 'SHP_TO_TXT_BIN' line 11 or option --bin-shp to a correct path.")
+
+ if not os.path.isdir(args.dir):
+ raise NameError("'" + args.dir + "' directory does not exist.")
+
+ # Si le mode est 'worst' alors la taille des segments est inutiles.
+ if args.profile == Profile.WORST:
+ args.size = [0]
+
+ iter = 0
+ total_iter = len(args.nseg) * len(args.size) * args.batch
+ if args.profile == Profile.BEST:
+ total_iter *= len(args.band)
+
+ def advance_and_print():
+
+ global iter
+ iter += 1
+ if args.verbose > 1:
+ print("progress: %d / %d" % (iter, total_iter))
+
+ def format_path(nseg, size, i, band=None):
+
+ filename = "data-%s-%d" % (args.profile, nseg)
+ if args.profile != Profile.WORST:
+ filename += "-size-%d" % size
+ if args.profile == Profile.BEST and band != None:
+ filename += "-band-%d" % band
+ filename += "-%d.txt" % i
+
+ return os.path.join(args.dir, filename)
+
+ start = time.time()
+
+ for nseg, size, i in itertools.product(args.nseg, args.size, range(args.batch)):
+
+ if args.profile == Profile.BEST:
+
+ for band in args.band:
+ shp_to_txt(Profile.BEST, nseg, size=size, band=band,
+ output=format_path(nseg, size, i, band=band))
+ advance_and_print()
+
+ elif args.profile == Profile.GAUSS:
+
+ random_segments(n=nseg, size=size, error=args.error, fixed=args.fixed,
+ output=format_path(nseg, size, i))
+ advance_and_print()
+
+ else:
+
+ shp_to_txt(args.profile, nseg, size=size,
+ output=format_path(nseg, size, i))
+ advance_and_print()
+
+ stop = time.time()
+ if args.verbose:
+ print("ellapsed time:", stop - start, "sec")
diff --git a/python_scripts/histogram_segment.py b/python_scripts/histogram_segment.py
new file mode 100644
index 0000000..b7a126c
--- /dev/null
+++ b/python_scripts/histogram_segment.py
@@ -0,0 +1,129 @@
+import argparse
+import numpy as np
+import matplotlib.collections as mcollec
+import matplotlib.colors as mcolors
+import matplotlib.pyplot as plt
+import os
+
+def make_size_bar(ax, size, bins):
+
+ ax.hist(size, bins=bins, density=True)
+
+ ax.set_xlabel("size")
+ ax.set_ylabel("frequency")
+ ax.grid(True)
+
+ return ax
+
+def make_dir_bar(ax, dir, bins):
+
+ ax.hist(dir, bins=bins, density=True)
+
+ ax.set_xlabel("direction [deg]")
+ ax.set_ylabel("frequency")
+ ax.grid(True)
+
+ return ax
+
+def make_pos_bar(ax, x, y, bins):
+
+ hist = ax.hist2d(x, y, bins=bins, density=True)
+ cbar = plt.colorbar(hist[3], ax=ax)
+
+ ax.set_xlabel("x")
+ ax.set_ylabel("y")
+ cbar.ax.set_ylabel("frequency")
+
+ return ax
+
+if __name__ == "__main__":
+
+ parser = argparse.ArgumentParser()
+
+ parser.add_argument("input",
+ help="input result file")
+ parser.add_argument("--bins", "-b", type=int, default=16,
+ help="set histogram number of bins")
+
+ parser.add_argument("--split", "-s", action="store_true",
+ help="one figure per graph instead of subplots")
+ parser.add_argument("--draw", "-d", action="store",
+ help="draw segments on another figure and export to PDF")
+ parser.add_argument("--export", "-e", action="store_true",
+ help="export histograms")
+ parser.add_argument("--noshow", "-n", action="store_true",
+ help="dont show figures")
+ parser.add_argument("--ext", "-x", action="store", default="png",
+ help="set export extension, default is png")
+
+ args = parser.parse_args()
+
+ data = np.genfromtxt(args.input, dtype=float, skip_header=1)
+
+ x0 = data[:, 0]
+ y0 = data[:, 1]
+ x1 = data[:, 2]
+ y1 = data[:, 3]
+
+ dx = x1 - x0
+ dy = y1 - y0
+
+ size = np.sqrt(np.square(dx) + np.square(dy))
+ dir = np.arctan(dy / dx) * 180.0 / np.pi
+
+ x = 0.5 * (x0 + x1)
+ y = 0.5 * (y0 + y1)
+
+ basename = os.path.splitext(os.path.basename(args.input))[0]
+
+ if args.split:
+
+ fig0 = plt.figure()
+ fig1 = plt.figure()
+ fig2 = plt.figure()
+
+ make_size_bar(fig0.add_subplot(), size, args.bins)
+ make_dir_bar (fig1.add_subplot(), dir, args.bins)
+ make_pos_bar (fig2.add_subplot(), x, y, args.bins)
+
+ if args.export:
+ fig0.savefig("%s-hist-size.%s" % (basename, args.ext))
+ fig1.savefig("%s-hist-dir.%s" % (basename, args.ext))
+ fig2.savefig("%s-hist-pos.%s" % (basename, args.ext))
+
+ else:
+
+ fig = plt.figure()
+
+ make_size_bar(fig.add_subplot(1, 3, 1), size, args.bins)
+ make_dir_bar (fig.add_subplot(1, 3, 2), dir, args.bins)
+ make_pos_bar (fig.add_subplot(1, 3, 3), x, y, args.bins)
+
+ if args.export:
+ fig.savefig("%s-hist.%s" % (basename, args.ext))
+
+ if args.draw:
+
+ palette = list(mcolors.TABLEAU_COLORS.values())
+ n = len(palette)
+
+ lines = []
+ colors = []
+ for i in range(data.shape[0]):
+
+ row = data[i, :]
+ lines.append([(row[0], row[1]), (row[2], row[3])])
+
+ colors.append(palette[i % n])
+
+ fig = plt.figure()
+ ax = fig.add_axes([0, 0, 1, 1])
+
+ ax.add_collection(mcollec.LineCollection(lines, color=colors))
+ ax.autoscale()
+ ax.axis('off')
+
+ fig.savefig(args.draw)
+
+ if not args.noshow:
+ plt.show()
diff --git a/python_scripts/parselog.py b/python_scripts/parselog.py
new file mode 100644
index 0000000..d0516dc
--- /dev/null
+++ b/python_scripts/parselog.py
@@ -0,0 +1,22 @@
+import argparse
+import numpy as np
+import matplotlib.pyplot as plt
+import os
+import re
+
+## Le Thread.log c'est le resultat de
+# cat toto.log | grep "\[Threa" >! toto-thread.log
+
+
+with open("toto-thread.log") as file:
+
+ for line in file:
+ p = re.compile( 'time=\d+.\d+' )
+ alltimings = p.findall( line )
+ print(len(alltimings))
+ allT = []
+ for athread in alltimings:
+ p = re.compile( '\d+.\d+' )
+ t = p.findall( athread )
+ allT = allT + t
+ print(allT)
diff --git a/python_scripts/plot_result.py b/python_scripts/plot_result.py
new file mode 100644
index 0000000..6156768
--- /dev/null
+++ b/python_scripts/plot_result.py
@@ -0,0 +1,492 @@
+import argparse
+import itertools
+import numpy as np
+import matplotlib.colors as mcolors
+import matplotlib.pyplot as plt
+import os
+import warnings
+
+# PB: see "matplotlib named_colors"
+PREFERED_COLORS = {
+ "load": mcolors.TABLEAU_COLORS["tab:blue"],
+ "local": mcolors.TABLEAU_COLORS["tab:orange"],
+ "global": mcolors.TABLEAU_COLORS["tab:green"],
+ "idle": mcolors.TABLEAU_COLORS["tab:red"],
+}
+
+def last_word(line):
+ return line.rsplit(" ", 1)[1]
+
+def parse_total(lines,
+ cgal=False):
+ index = 4 if cgal else 10
+ return float(last_word(lines[index]))
+
+def parse_load(lines,
+ cgal=False):
+ index = 3 if cgal else 7
+ return float(last_word(lines[index]))
+
+def parse_local(lines):
+ return float(last_word(lines[8]))
+
+def parse_global(lines):
+ return float(last_word(lines[9]))
+
+def parse_thread(lines):
+
+ kv = {}
+ for line in lines[1].split("[")[1:]: # data per thread
+ line = line.split()
+ b = int(line[2].split("=")[1])
+ t = float(line[-1].split("=")[1])
+ kv[b] = t
+
+ return [ kv[k] for k in sorted(kv) ]
+
+def parse_memory(lines,
+ cgal=False):
+ index = 2 if cgal else 6
+ return int(lines[index].split(" ", 3)[2])
+
+def parse_traverse(lines):
+ lines = lines[11].split(" times ", 1)
+ n = int(lines[0].rsplit(" ", 1)[1])
+ t = float(lines[1].split(" ", 1)[0])
+ return t / n
+
+def parse_cdarts(lines):
+
+ total = 0
+ for line in lines[3].split("|")[0:-1]: # data per band
+ line = line.rsplit(", ", 1)
+ lcd = int(line[0].rsplit("(", 1)[1])
+ rcd = int(line[1].rsplit(")", 1)[0])
+ total += lcd + rcd
+
+ return total
+
+def read_log(filename):
+
+ data = {}
+
+ with open(filename) as file:
+
+ it = iter(file)
+ for _ in it:
+
+ line = next(it).strip()
+ method = line.rsplit(" ", 1)[1]
+
+ if method == "cgal":
+
+ if "cgal" not in data:
+ data["cgal"] = {
+ "total": { "detail": [] },
+ "load": { "detail": [] },
+ "global": { "detail": [] },
+ "memory": { "detail": [] }
+ }
+
+ entry = data["cgal"]
+ lines = [ next(it).strip() for _ in range(5) ]
+
+ total = parse_total (lines, cgal=True)
+ load = parse_load (lines, cgal=True)
+ entry["total"]["detail"] .append(total)
+ entry["load"] ["detail"] .append(load)
+ entry["global"]["detail"].append(total - load)
+ entry["memory"]["detail"].append(parse_memory(lines, cgal=True))
+
+ else:
+
+ nbthread = int(method)
+
+ if method not in data:
+ data[method] = {
+ "total": { "detail": [] },
+ "load" : { "detail": [] },
+ "local": { "detail": [] },
+ "global": { "detail": [] },
+ "thread": { "detail": [ [] for _ in range(nbthread) ] },
+ "memory": { "detail": [] },
+ "traverse": { "detail": [] },
+ "cdarts": { "detail": [] }
+ }
+
+ entry = data[method]
+ lines = [ next(it).strip() for _ in range(12) ]
+
+ entry["total"] ["detail"].append(parse_total (lines))
+ entry["load"] ["detail"].append(parse_load (lines))
+ entry["local"] ["detail"].append(parse_local (lines))
+ entry["global"] ["detail"].append(parse_global(lines))
+ entry["memory"] ["detail"].append(parse_memory(lines))
+ entry["traverse"]["detail"].append(parse_traverse(lines))
+ entry["cdarts"] ["detail"].append(parse_cdarts(lines))
+
+ try :
+ thread = parse_thread(lines)
+ for i in range(nbthread):
+ entry["thread"]["detail"][i].append(thread[i])
+ except:
+ warnings.warn("found thread without band in")
+ for i in range(nbthread):
+ entry["thread"]["detail"][i].append(np.nan)
+
+ for method in data:
+ for key in data[method]:
+
+ entry = data[method][key]
+
+ arr = np.array(entry["detail"]).transpose()
+ entry["mean"] = np.nanmean(arr, axis=0)
+ entry["dev"] = np.nanstd(arr, axis=0)
+
+ return data
+
+def remove_idle(data):
+
+ for m in data:
+ if m != "cgal":
+
+ t_local = data[m]["local"] ["mean"]
+ t_thread = data[m]["thread"]["mean"]
+ offset = np.min(t_local - t_thread)
+
+ data[m]["local"]["mean"] -= offset
+ data[m]["total"]["mean"] -= offset
+
+ return data
+
+#def fill_cgal(data):
+#
+# if "cgal" in data:
+#
+# entry = data["cgal"]
+#
+# entry["load"] = { "mean": 0.0, "dev": 0.0 }
+# entry["local"] = { "mean": 0.0, "dev": 0.0 }
+# entry["global"] = entry["total"]
+# entry["thread"] = { "mean": np.array([ 0.0 ]), "dev": np.array([ 0.0 ]) }
+# entry["memory"] = { "mean": 0.0, "dev": 0.0 }
+# entry["traverse"] = { "mean": 0.0, "dev": 0.0 }
+#
+# return data
+#
+#def extract_field(data, methods, field):
+# mean = list(map(lambda m: data[m][field]["mean"], methods))
+# dev = list(map(lambda m: data[m][field]["dev"], methods))
+# return mean, dev
+
+def extract_field(data, methods, field):
+ mean = []
+ dev = []
+ for m in methods:
+ if field in data[m]:
+ mean.append(data[m][field]["mean"])
+ dev .append(data[m][field]["dev"])
+ else:
+ if field == "thread":
+ mean.append([ 0.0 ])
+ dev .append([ 0.0 ])
+ else:
+ mean.append(0.0)
+ dev .append(0.0)
+ return mean, dev
+
+def make_simple_bar(ax, data, methods, field, without_error):
+
+ mean, dev = extract_field(data, methods, field)
+ ind = np.arange(0, len(methods))
+
+ b = ax.bar(ind, mean, yerr=dev)
+
+ if without_error:
+ for e in b.errorbar[2]:
+ e.remove()
+
+ ax.set_xticks(ind)
+ ax.set_xticklabels(methods)
+ ax.grid(True)
+
+ return ax
+
+def make_total_bar(ax, data,
+ without_error=False, without_load=False):
+
+ methods = data.keys()
+ mean, dev = extract_field(data, methods, "total")
+
+ ind = np.arange(0, len(methods))
+
+ if without_load:
+ load_mean, _ = extract_field(data, methods, "load")
+ y = np.array(mean) - np.array(load_mean)
+ else:
+ y = mean
+ b = ax.bar(ind, y, yerr=dev)
+
+ if without_error:
+ for e in b.errorbar[2]:
+ e.remove()
+
+ ax.set_xticks(ind)
+ ax.set_xticklabels(methods)
+ ax.set_xlabel("method - threads")
+ ax.set_ylabel("duration [sec]")
+ ax.grid(True)
+
+ return ax
+
+def make_step_bar(ax, data,
+ without_error=False, without_load=False):
+
+ methods = data.keys()
+ t0_mean, t0_dev = extract_field(data, methods, "load")
+ t1_mean, t1_dev = extract_field(data, methods, "local")
+ t2_mean, t2_dev = extract_field(data, methods, "global")
+
+ ind = np.arange(0, len(methods))
+
+ if without_load:
+
+ p1 = ax.bar(ind, t1_mean, yerr=t1_dev, color=PREFERED_COLORS["local"])
+ p2 = ax.bar(ind, t2_mean, yerr=t2_dev, bottom=t1_mean, color=PREFERED_COLORS["global"])
+
+ bars = (p1, p2)
+ steps = ("local", "global")
+
+ else:
+
+ p0 = ax.bar(ind, t0_mean, yerr=t0_dev, color=PREFERED_COLORS["load"])
+ p1 = ax.bar(ind, t1_mean, yerr=t1_dev, bottom=t0_mean, color=PREFERED_COLORS["local"])
+ p2 = ax.bar(ind, t2_mean, yerr=t2_dev, bottom=(np.array(t0_mean) + np.array(t1_mean)), color=PREFERED_COLORS["global"])
+
+ bars = (p0, p1, p2)
+ steps = ("load", "local", "global")
+
+ if without_error:
+ for e in itertools.chain(*[ b.errorbar[2] for b in bars ]):
+ e.remove()
+
+ ax.set_xticks(ind)
+ ax.set_xticklabels(methods)
+ ax.set_xlabel("method - threads")
+ ax.set_ylabel("duration [sec]")
+ ax.grid(True)
+
+ ax.legend([ b[0] for b in bars ], steps)
+
+ return ax
+
+def make_detail_bar(ax, data,
+ without_error=False, without_load=False):
+
+ methods = list(data.keys())
+ t0_mean, t0_dev = extract_field(data, methods, "load")
+ t1_mean, _ = extract_field(data, methods, "local")
+ t2_mean, t2_dev = extract_field(data, methods, "global")
+ t3_mean, t3_dev = extract_field(data, methods, "thread")
+
+ ind = np.arange(0, len(methods))
+
+ w = 0.8
+ x = []
+ height = []
+ width = []
+ yerr = []
+ bottom = []
+
+ height_idle = []
+ bottom_idle = []
+
+ for i in range(len(methods)):
+
+ n = 1 if methods[i] == "cgal" else int(methods[i])
+ wi = w / float(n)
+ a = (w - wi) / (float(n) - 1.0) if n != 1 else 0.0
+ b = float(ind[i]) + 0.5 * (wi - w)
+
+ for j in range(n):
+
+ t0_mean_i = 0.0 if without_load else t0_mean[i]
+
+ x .append(a * j + b)
+ height.append(t3_mean[i][j])
+ width .append(wi)
+ yerr .append(t3_dev[i][j])
+ bottom.append(t0_mean_i)
+
+ height_idle.append(t1_mean[i] - t3_mean[i][j])
+ bottom_idle.append(t0_mean_i + t3_mean[i][j])
+
+ if without_load:
+
+ p1 = ax.bar(x, height, yerr=yerr, bottom=bottom, width=width, color=PREFERED_COLORS["local"])
+ p2 = ax.bar(ind, t2_mean, yerr=t2_dev, bottom=t1_mean, color=PREFERED_COLORS["global"])
+
+ p_idle = ax.bar(x, height_idle, bottom=bottom_idle, width=width, color=PREFERED_COLORS["idle"])
+
+ bars = (p1, p2, p_idle)
+ steps = ("local", "global", "idle")
+
+ bars_with_error = (p1, p2)
+
+ else:
+
+ p0 = ax.bar(ind, t0_mean, yerr=t0_dev, color=PREFERED_COLORS["load"])
+ p1 = ax.bar(x, height, yerr=yerr, bottom=bottom, width=width, color=PREFERED_COLORS["local"])
+ p2 = ax.bar(ind, t2_mean, yerr=t2_dev, bottom=(np.array(t0_mean) + np.array(t1_mean)), color=PREFERED_COLORS["global"])
+
+ p_idle = ax.bar(x, height_idle, bottom=bottom_idle, width=width, color=PREFERED_COLORS["idle"])
+
+ bars = (p0, p1, p2, p_idle)
+ steps = ("load", "local", "global", "idle")
+
+ bars_with_error = (p0, p1, p2)
+
+ if without_error:
+ for e in itertools.chain(*[ b.errorbar[2] for b in bars_with_error ]):
+ e.remove()
+ else:
+ # PB: hide thread error
+ for e in p1.errorbar[2]:
+ e.remove()
+
+ ax.set_xticks(ind)
+ ax.set_xticklabels(methods)
+ ax.set_xlabel("method - threads")
+ ax.set_ylabel("duration [sec]")
+ ax.grid(True)
+
+ ax.legend([ b[0] for b in bars ], steps)
+
+ return ax
+
+def make_memory_bar(ax, data,
+ without_error=False):
+
+ ax = make_simple_bar(ax, data, data.keys(), "memory", without_error)
+ ax.set_xlabel("threads")
+ ax.set_ylabel("memory [bytes]")
+
+ return ax
+
+def make_traverse_bar(ax, data,
+ without_error=False):
+
+ methods = list(filter(lambda m: m != "cgal", data.keys()))
+ ax = make_simple_bar(ax, data, methods, "traverse", without_error)
+ ax.set_xlabel("threads")
+ ax.set_ylabel("traversal duration [sec]")
+
+ return ax
+
+def print_data(data, *args,
+ spacing=20):
+
+ unsupported = [ "thread" ]
+
+ methods = data.keys()
+ row_format = ("{:>" + str(spacing) + "}") * (len(methods) + 1)
+ print(row_format.format("", *methods))
+
+ for field in args:
+
+ if field in unsupported:
+ warnings.warn("'print_data' does not support '" + field + "' field")
+ continue
+
+ mean, _ = extract_field(data, methods, field)
+ print(row_format.format(field, *mean))
+
+if __name__ == "__main__":
+
+ parser = argparse.ArgumentParser()
+
+ parser.add_argument("input", nargs="*",
+ help="input result files")
+
+ parser.add_argument("--split", "-s", action="store_true",
+ help="one figure per graph instead of subplots")
+ parser.add_argument("--export", "-e", action="store_true",
+ help="export plots")
+ parser.add_argument("--ext", "-x", action="store", default="png",
+ help="set export extension, default is png")
+
+ parser.add_argument("--no-error", action="store_true",
+ help="hide error bar")
+ parser.add_argument("--no-load", action="store_true",
+ help="hide load bar")
+
+ parser.add_argument("--hide", action="store_true",
+ help="disable show")
+
+ args = parser.parse_args()
+
+ for input in args.input:
+
+ data = read_log(input)
+ # PB: hack to remove idle
+ data = remove_idle(data)
+
+ basename = os.path.splitext(os.path.basename(input))[0]
+
+ # if args.split:
+ if True:
+ #fig0 = plt.figure()
+ #fig1 = plt.figure()
+ fig2 = plt.figure()
+ #fig3 = plt.figure()
+ #fig4 = plt.figure()
+
+ #fig0.canvas.set_window_title("%s - total" % input)
+ #fig1.canvas.set_window_title("%s - step" % input)
+ fig2.canvas.set_window_title("%s - detail" % input)
+ #fig3.canvas.set_window_title("%s - memory" % input)
+ #fig4.canvas.set_window_title("%s - traverse" % input)
+
+ #ax0 = fig0.add_subplot()
+ #ax1 = fig1.add_subplot(sharex=ax0, sharey=ax0)
+ ax2 = fig2.add_subplot() #sharex=ax0, sharey=ax0)
+ #ax3 = fig3.add_subplot()
+ #ax4 = fig4.add_subplot()
+
+ #make_total_bar (ax0, data, without_error=args.no_error, without_load=args.no_load)
+ #make_step_bar (ax1, data, without_error=args.no_error, without_load=args.no_load)
+ make_detail_bar (ax2, data, without_error=args.no_error, without_load=args.no_load)
+ #make_memory_bar (ax3, data, without_error=args.no_error)
+ #make_traverse_bar(ax4, data, without_error=args.no_error)
+
+ if args.export:
+ #fig0.savefig("%s-total.%s" % (basename, args.ext))
+ #fig1.savefig("%s-step.%s" % (basename, args.ext))
+ fig2.savefig("%s-detail.%s" % (basename, args.ext))
+ #fig3.savefig("%s-memory.%s" % (basename, args.ext))
+ #fig4.savefig("%s-traverse.%s" % (basename, args.ext))
+
+ # else:
+
+ # fig = plt.figure()
+ # fig.canvas.set_window_title(input)
+
+ # ax0 = fig.add_subplot(1, 5, 1)
+ # ax1 = fig.add_subplot(1, 5, 2, sharex=ax0, sharey=ax0)
+ # ax2 = fig.add_subplot(1, 5, 3, sharex=ax0, sharey=ax0)
+ # ax3 = fig.add_subplot(1, 5, 4)
+ # ax4 = fig.add_subplot(1, 5, 5)
+
+ # make_total_bar (ax0, data, without_error=args.no_error, without_load=args.no_load)
+ # make_step_bar (ax1, data, without_error=args.no_error, without_load=args.no_load)
+ # make_detail_bar (ax2, data, without_error=args.no_error, without_load=args.no_load)
+ # make_memory_bar (ax3, data, without_error=args.no_error)
+ # make_traverse_bar(ax4, data, without_error=args.no_error)
+
+ # if args.export:
+ # fig.savefig("%s.%s" % (basename, args.ext))
+
+ print_data(data, "memory", "traverse", "cdarts")
+
+ if not args.hide:
+ plt.show()
diff --git a/python_scripts/run_benchmark.py b/python_scripts/run_benchmark.py
new file mode 100644
index 0000000..e8214dc
--- /dev/null
+++ b/python_scripts/run_benchmark.py
@@ -0,0 +1,173 @@
+import argparse
+import os
+import subprocess
+import time
+
+from enum import Enum
+
+from generate_data import bin_exist
+
+class Heuristic(Enum):
+
+ EDGE = "edge"
+ INTERSECTION = "intersection"
+ SEGMENT = "segment"
+ SEGMENT_OLD = "segment-old"
+
+ def __str__(self):
+ return self.value
+
+ def __repr__(self):
+ return str(self)
+
+PARALLEL_ARRANGEMENT_BIN = "./builds/release/parallel-arrangement"
+# COMPUTE_BAND_BIN = "./builds/release/tools/compute-band"
+
+def parallel_arrangement(t1, nbt,
+ xpos=None,
+ crop=False, optimized=False, traverse=False,
+ cgal=False):
+
+ args = [
+ PARALLEL_ARRANGEMENT_BIN,
+ "-t1", t1,
+ "-nbt", str(nbt),
+ "-nbs", str(nbt)
+ ]
+ if xpos:
+ args.append("-xpos")
+ args.extend(map(str,xpos))
+ if crop:
+ args.append("-crop")
+ if optimized:
+ args.append("-optimized-bands")
+ if traverse:
+ args.append("-traverse")
+ if cgal:
+ args.append("-cgal")
+
+ return subprocess.run(args, stdout=subprocess.PIPE)
+
+# def compute_band(filename, nbband, heuristic):
+
+# args = [
+# COMPUTE_BAND_BIN,
+# filename,
+# str(nbband),
+# "--" + str(heuristic)
+# ]
+
+# proc = subprocess.run(args, stdout=subprocess.PIPE)
+
+# lines = proc.stdout.decode("utf-8").splitlines()
+# pos = lines[-1].split()[1:]
+# band_pos = [ float(e) for e in pos ]
+
+# return proc, band_pos
+
+if __name__ == "__main__":
+
+ parser = argparse.ArgumentParser()
+
+ parser.add_argument("input", nargs="+",
+ help="set input, input can be a directory or list of file")
+
+ parser.add_argument('--verbose', '-v', action='count', default=0,
+ help="set verbosity level")
+ parser.add_argument("--log", "-l", default="log.txt",
+ help="set log filename, default is log.txt")
+
+ config = parser.add_argument_group("benchmark configuration")
+
+ config.add_argument("--thread", "-t", required=True, type=int, nargs="+",
+ help="set number of threads")
+ config.add_argument("--repeat", "-r", required=True, type=int, default=5,
+ help="set number of repeatition per data")
+ config.add_argument("--nocgal", action="store_true",
+ help="disable benchmark with cgal method")
+
+ optimi = parser.add_argument_group("optimization configuration")
+
+ optimi.add_argument("--optimized", action="store_true",
+ help="use optimized method")
+ optimi.add_argument("--crop", action="store_true",
+ help="use crop method")
+ optimi.add_argument("--heuristic", type=Heuristic,
+ choices=list(Heuristic),
+ help="set heuristic used to compute band for each thread")
+
+ binary = parser.add_argument_group("binaries configuration")
+
+ binary.add_argument("--arr-bin", default=PARALLEL_ARRANGEMENT_BIN,
+ help="set 'parallel-arrangement' binary path, default value is set line 9")
+ # binary.add_argument("--band-bin", default=COMPUTE_BAND_BIN,
+ # help="set 'compute-band' binary path, default value is set line 11")
+
+ args = parser.parse_args()
+
+ PARALLEL_ARRANGEMENT_BIN = args.arr_bin
+ # COMPUTE_BAND_BIN = args.band_bin
+
+ if not bin_exist(PARALLEL_ARRANGEMENT_BIN):
+ raise NameError("'" + PARALLEL_ARRANGEMENT_BIN + "' parallel-arrangement path does not exist. "
+ "Set variable 'PARALLEL_ARRANGEMENT_BIN' line 23 to a correct path.")
+ # if not bin_exist(COMPUTE_BAND_BIN):
+ # raise NameError("'" + COMPUTE_BAND_BIN + "' compute-band path does not exist. "
+ # "Set variable 'COMPUTE_BAND_BIN' line 24 to a correct path.")
+
+ if len(args.input) == 1 and os.path.isdir(args.input[0]):
+ dir = args.input[0]
+ args.input = [ os.path.join(dir, f) for f in os.listdir(dir) ]
+
+ iter = 0
+ total_iter = (len(args.thread) + (0 if args.nocgal else 1)) * len(args.input) * args.repeat
+
+ def advance_and_print():
+ global iter
+ iter += 1
+ if args.verbose > 1:
+ print("progress: %d / %d" % (iter, total_iter))
+
+ log_separator = ("#" * 80 + "\n").encode()
+
+ start = time.time()
+
+ with open(args.log, "wb") as log_file:
+
+ for input in args.input:
+
+ band_xpos = { t:None for t in args.thread }
+
+ # if args.heuristic:
+ # for thread in args.thread:
+ # if thread != 1:
+ # _, pos = compute_band(input, thread, args.heuristic)
+ # band_xpos[thread] = pos
+
+ for j in range(args.repeat):
+
+ if not args.nocgal:
+ proc = parallel_arrangement(input, 1,
+ cgal=True,
+ traverse=True)
+ log_file.write(log_separator)
+ log_file.write(("# \"%s\" repeat %d thread cgal\n" % (input, j)).encode())
+ log_file.write(proc.stdout)
+
+ advance_and_print()
+
+ for thread in args.thread:
+ proc = parallel_arrangement(input, thread,
+ crop=args.crop,
+ optimized=args.optimized,
+ xpos=band_xpos[thread],
+ traverse=True)
+ log_file.write(log_separator)
+ log_file.write(("# \"%s\" repeat %d thread %d\n" % (input, j, thread)).encode())
+ log_file.write(proc.stdout)
+
+ advance_and_print()
+
+ stop = time.time()
+ if args.verbose:
+ print("ellapsed time:", stop - start, "sec")
--
GitLab