-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun-figure.py
230 lines (187 loc) · 10.6 KB
/
run-figure.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
import argparse
import json
import pandas as pd
import numpy as np
import matplotlib
import matplotlib.pyplot as plt
import csv
import os
from pathlib import Path
matplotlib.rcParams['pdf.fonttype'] = 42
matplotlib.rcParams['ps.fonttype'] = 42
colors = {"tblis" : "olive", "gsl" : "green", "blas" : "red", "taco" : "blue", "gsl_tensor" : "cyan", \
"dot_blas" : "gold", "dot_gsl" : "grey", "gemv_blas" : "purple", "gemv_gsl" : "pink", "mkl" :"black", "cuda":"green",\
"dot_mkl" : "silver", "gemv_mkl" : "yellow", "stardust":"orange",
"taco_csr" : "green", "taco_coo" : "red", "row" : "olive", "col" : "green", "block" : "red",
"block_diagonal" : "blue", "random" : "cyan", "bug_mkl":"purple", "bug_gsl":"pink"}
markers = {"tblis" : "o", "gsl" : "p", "blas" : "*", "taco" : ".", "gsl_tensor" : ".", \
"dot_blas" : ".", "dot_gsl" : ".", "gemv_blas" : ".", "gemv_gsl" : "p", "mkl" :"v", "cuda":"s",\
"dot_mkl" : ".", "gemv_mkl" : ".", "stardust":"1", "taco_coo" : "p", "taco_csr" : ".", \
"row" : "o", "col" : "p", "random" : "*", "block" : ".", "block_diagonal" : ".", "bug_mkl": ".", "bug_gsl":"s"}
linestyles = {"tblis" : "-", "gsl" : "-", "blas" : "-", "taco" : "-", "gsl_tensor" : "-", \
"dot_blas" : "-", "dot_gsl" : "-", "gemv_blas" : "-", "gemv_gsl" : "-", "mkl" :"--", "cuda":"-",\
"dot_mkl" : "-", "gemv_mkl" : "-", "stardust":"-", "taco_coo" : "-", "taco_csr" : "--",\
"row" : "-", "col" : "-", "block" : "-", "block_diagonal" : "-", "random" : "-", "bug_mkl":"-", "bug_gsl":"--" }
def generate_dim_plot(name, directory, systems, expr, start, interval, stardust=None, filtered=None, unit="us", svg=False):
result = None
for system in systems:
data = json.load(open(f'{directory}/{system}'))
df = pd.DataFrame(data["benchmarks"])
try:
df = df[df['aggregate_name'] == "median"]['real_time']
except:
df = df['real_time']
df = df.reset_index(drop=True)
df = df.rename_axis('dimension').reset_index()
df['dimension'] = df['dimension']*interval + start
df.rename(columns = {'real_time': f'{system}_real_time'}, inplace = True)
if result is None:
result = df
else:
result = pd.merge(result, df, on='dimension', how='outer')
if stardust is not None:
data = pd.read_csv(os.getenv("PATH_TO_MOSAIC_ARTIFACT") + "/mosaic-benchmarks/stardust-runs/spmv_plus2.csv")
if stardust == "SpMV":
df = pd.DataFrame(data[["app", "cycles", "dim_0_2"]])
df.rename(columns = {'dim_0_2': f'dimension'}, inplace = True)
else:
df = pd.DataFrame(data[["app", "cycles", "dataset"]])
# Stardust cycles are in ns and call to it takes 0.0002 ms
df[df['app'] == stardust]
if unit == "ms":
df['cycles'] = df['cycles'] *1e-6 + 0.0002
elif unit == "us":
df['cycles'] = df['cycles'] * 1e-3 + 0.0002*1000
else:
raise NotImplemented
df.rename(columns = {'cycles': f'stardust_real_time'}, inplace = True)
result = pd.merge(result, df, on='dimension', how='outer')
systems.append("stardust")
if filtered is not None:
for system in filtered:
data = pd.read_csv(f'{directory}/{system}_filter')
df = data['real_time']
df = df.reset_index(drop=True)
df = df.rename_axis('dimension').reset_index()
df['dimension'] = df['dimension']*interval + start
df.rename(columns = {'real_time': f'{system}_real_time'}, inplace = True)
result = pd.merge(result, df, on='dimension', how='outer')
systems = systems + filtered
full_plt = result.plot(kind = 'line', x = 'dimension', y = [f'{i}_real_time' for i in systems],
color=[colors[system] for system in systems],
#marker='plot_marker',
title=expr)
for i, line in enumerate(full_plt.get_lines()):
line.set_marker(markers[systems[i]])
line.set_linestyle(linestyles[systems[i]])
plt.ylabel('Real Time (ms)')
plt.xlabel('Dimension')
plt.savefig(f'{directory}/raw_graph.svg', format="svg")
f, (ax, ax2) = plt.subplots(2, 1, sharex=True)
result.plot(kind = 'line', x = 'dimension', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], ax=ax)
result.plot(kind = 'line', x = 'dimension', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], ax=ax2)
full_plt = result.plot(kind = 'line', x = 'dimension', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], title=expr)
for i, line in enumerate(full_plt.get_lines()):
line.set_marker(markers[systems[i]])
line.set_linestyle(linestyles[systems[i]])
full_plt.legend(full_plt.get_lines(), [f'{i}_real_time' for i in systems])
plt.yscale("log")
plt.ylabel(f'Real Time ({unit})')
plt.xlabel('Dimension')
plt.savefig(f'{directory}/{name}.pdf', format="pdf")
if svg:
plt.savefig(f'{directory}/{name}.svg', format="svg")
def generate_sparsity_plots(name, directory, systems, expr, sparse, stardust=None, unit="us", svg=False):
print(stardust)
result = None
for system in systems:
data = json.load(open(f'{directory}/{system}'))
df = pd.DataFrame(data["benchmarks"])
try:
df = df[df['aggregate_name'] == "median"]['real_time']
except:
df = df['real_time']
df = df.reset_index(drop=True)
df = df.rename_axis('sparisty').reset_index()
df.rename(columns = {'real_time': f'{system}_real_time'}, inplace = True)
if result is None:
result = df
else:
result = pd.merge(result, df, on='sparisty', how='outer')
print(result)
if stardust is not None:
data = pd.read_csv(os.getenv("PATH_TO_MOSAIC_ARTIFACT") + "/mosaic-benchmarks/stardust-runs/spmv_plus2.csv")
if stardust == "SpMV":
df = pd.DataFrame(data[["app", "cycles", "dataset"]])
df.rename(columns = {'dim_0_2': f'dimension'}, inplace = True)
elif stardust == "Plus2CSR":
df = pd.DataFrame(data[["app", "cycles", "dataset", "par"]])
df.sort_values(by=['dataset'])
df = df[df['par'] == 8]
# Stardust cycles are in ns and call to it takes 0.0002 ms
df = df[df['app'] == stardust]
if unit == "ms":
df['cycles'] = df['cycles'] *1e-6 + 0.0002
elif unit == "us":
df['cycles'] = df['cycles'] *1e-3 + 0.0002*1000
df = df.reset_index(drop=True)
df = df.rename_axis('sparisty').reset_index()
print(df)
df.rename(columns = {'cycles': f'stardust_real_time'}, inplace = True)
result = pd.merge(result, df, on='sparisty', how='outer')
systems.append("stardust")
sparse = [str(i) for i in sparse]
result['sparisty'] = sparse
full_plt = result.plot(kind = 'line', x = 'sparisty', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], title=expr)
for i, line in enumerate(full_plt.get_lines()):
line.set_marker(markers[systems[i]])
line.set_linestyle(linestyles[systems[i]])
full_plt.legend(full_plt.get_lines(), [f'{i}_real_time' for i in systems])
f, (ax, ax2) = plt.subplots(2, 1, sharex=True)
result.plot(kind = 'line', x = 'sparisty', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], ax=ax)
result.plot(kind = 'line', x = 'sparisty', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], ax=ax2)
full_plt = result.plot(kind = 'line', x = 'sparisty', y = [f'{i}_real_time' for i in systems], color=[colors[system] for system in systems], title=expr)
for i, line in enumerate(full_plt.get_lines()):
line.set_marker(markers[systems[i]])
line.set_linestyle(linestyles[systems[i]])
full_plt.legend(full_plt.get_lines(), [f'{i}_real_time' for i in systems])
plt.yscale("log")
plt.ylabel(f'Real Time ({unit})')
plt.xlabel('sparisty')
plt.savefig(f'{directory}/{name}.pdf', format="pdf")
if svg:
plt.savefig(f'{directory}/{name}.svg', format="svg")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Generate Figures from Data")
parser.add_argument("--data_dir", type=str, default="./", help="Where To Find the Data")
parser.add_argument("--name", type=str, default="No Name", help="Name of the Figure")
parser.add_argument("--x_label", type=str, default="Dimension", help="Label of X axis")
parser.add_argument("--y_label", type=str, default="Runtime (us)", help="Label of Y axis")
parser.add_argument("--start_dim", type=int, default=0, help="Dimension to Start")
parser.add_argument("--step_dim", type=int, default=0, help="Step for dimension")
parser.add_argument("--type", type=str, default="", help="Which type of plot to generate")
parser.add_argument('--systems', type=str, default="")
parser.add_argument('--sparsity', type=str, default="")
parser.add_argument('--stardust', type=str, default="")
parser.add_argument('--unit', type=str, default="us")
parser.add_argument('--svg', action='store_true')
args = parser.parse_args()
print(args.stardust)
if args.type == "vary_sparse":
if (args.stardust == "Plus2CSR"):
generate_sparsity_plots(args.name, args.data_dir, args.systems.split(','),\
args.name, [float(item) for item in args.sparsity.split(',')],\
args.stardust, args.unit, args.svg)
else:
generate_sparsity_plots(args.name, args.data_dir, args.systems.split(','),\
args.name, [float(item) for item in args.sparsity.split(',')],\
None, args.unit, args.svg)
elif args.type == "vary_dim":
if (args.stardust == "SpMV"):
generate_dim_plot(args.name, args.data_dir, args.systems.split(','), args.name,\
args.start_dim, args.step_dim, args.stardust, None, args.unit, args.svg)
else:
generate_dim_plot(args.name, args.data_dir, args.systems.split(','), args.name,\
args.start_dim, args.step_dim, None, None, args.unit, args.svg)
else:
print("Type can only be of two types, vary_sparse and vary_dim")