energy_perf_bar.py
18.8 KB
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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
import os, sys, argparse
import numpy
import matplotlib.pyplot as plt
import json
from pylab import *
from mpl_toolkits.mplot3d import Axes3D
from matplotlib.patches import Rectangle
from matplotlib.backends.backend_qt4agg import FigureCanvasQTAgg as FigureCanvas
from sets import Set
from common import *
def get_energy(frontiers_data, cpu_freq, mem_freq, sample, bmark):
for point in frontiers_data["data"][sample]:
if point["cpu_freq"] == cpu_freq:
if point["mem_freq"] == mem_freq:
return point["energy"]
def get_performance(frontiers_data, cpu_freq, mem_freq, sample, bmark):
for point in frontiers_data["data"][sample]:
if point["cpu_freq"] == cpu_freq:
if point["mem_freq"] == mem_freq:
return point["performance"]
def plot_energy_perf_bar(args, threshold_list, budget):
dir_path = args.input_dir
output_dir_path = args.output_dir
perf_cost = 500000
energy_cost = 30000
benchmarks, labels = get_benchmarks(args)
#plot constants
cpuflist = get_cpu_freq_plot_list(args)
memflist = get_mem_freq_plot_list(args)
# finding points with 3% threshold of target budget/inefficiency
thresh = 3
energy_bar_data = [ [] for thr in threshold_list]
performance_bar_data = [ [] for thr in threshold_list]
energy_bar_data_nocost = [ [] for thr in threshold_list]
performance_bar_data_nocost = [ [] for thr in threshold_list]
print budget
for threshold_index, cluster_thresh in enumerate(threshold_list):
print cluster_thresh
data = []
for bmark in benchmarks:
print bmark
bmarkDirPath = os.path.join(os.path.join(dir_path, "per_sample_data"), bmark)
frontiers_file = os.path.join(bmarkDirPath, "per_sample_frontiers.json")
frontiers_data = json.loads(open(frontiers_file).read())
data_to_plot=[]
sample_points=[]
cpu_rectangle_points=[]
for data in frontiers_data["data"]:
filtered_data=[]
#filtering all those points which have inefficiency with error% of the budget
for point in data:
if point["inefficiency"] <= budget:
filtered_data.append(point)
elif (point["inefficiency"] - budget) * 100 / budget <= thresh:
filtered_data.append(point)
#finding the point with highest performance among filtered data
optimal_point = filtered_data[0]
for point in filtered_data:
if point["speedup"] > optimal_point["speedup"]:
optimal_point = point
local_rect_points=[]
for point in filtered_data:
if (optimal_point["speedup"] - point["speedup"]) < 0:
print "something is wrong!"
if (optimal_point["speedup"] - point["speedup"]) * 100 / optimal_point["speedup"] <= cluster_thresh:
data_to_plot.append(point)
sample_points.append(frontiers_data["data"].index(data))
local_rect_points.append(point)
cpufpoints = [cpuf for cpuf in [data_to_plot[sample]["cpu_freq"] for sample in range(len(data_to_plot))]]
memfpoints = [memf for memf in [data_to_plot[sample]["mem_freq"] for sample in range(len(data_to_plot))]]
samplepoints = sample_points
#Total number of transitions
num_transitions = 0.0
#Available settings for (CPU, MEM)
settings_available = Set()
index = 0
current_sample = -1
length = 1
lengths = []
prev_tr_sample = 0
energy = 0
performance = 0
while index < len( samplepoints ):
current_sample = samplepoints[index]
#Construct the current settings
current_settings = Set()
while index < len( samplepoints ) and samplepoints[index] == current_sample:
current_settings.add((cpufpoints[index],memfpoints[index]))
index = index + 1
#Compute the common points between current and what is available
common_points = current_settings.intersection(settings_available)
if (len(common_points) == 0 or index == len( samplepoints )):
# find the point with highest cpu and mem freq
if (current_sample !=0 ):
optimal_point = settings_available.pop()
settings_available.add(optimal_point)
for point in settings_available:
if point[0] > optimal_point[0]:
optimal_point = point
elif point[0] == optimal_point[0]:
if point[1] > optimal_point[1]:
optimal_point = point
done = 0
idx = prev_tr_sample+1
while done == 0:
energy += get_energy(frontiers_data, optimal_point[0], optimal_point[1], idx, bmark)
performance += get_performance(frontiers_data, optimal_point[0], optimal_point[1], idx, bmark)
if idx == samplepoints[index-1]:
done = 1
else:
idx += 1
prev_tr_sample = samplepoints[index-1]
# When there are no common points, transition
if (len(common_points) == 0):
settings_available = current_settings #all current settings are now available
#Ignore the transition if it's the first sample
if (current_sample!=0):
num_transitions += 1
else: # Continue with the common points
settings_available = common_points
# Record the length if we need to transition or we are at the end
if (len(common_points) == 0 or index == len( samplepoints )):
#Ignore the length if the transition is for the first sample
if (current_sample!=0):
lengths.append(length)
#reset the length
length = 1
else: #otherwise, just increment the length
length += 1
energy_bar_data[threshold_index].append((energy + (num_transitions * energy_cost))/1e6)
performance_bar_data[threshold_index].append((performance +(num_transitions * perf_cost))/1e6 )
energy_bar_data_nocost[threshold_index].append((energy)/1e6)
performance_bar_data_nocost[threshold_index].append((performance )/1e6 )
for threshold_index, cluster_thresh in enumerate(threshold_list):
for bmark_index, bmark in enumerate(benchmarks):
if threshold_index != 0:
energy_bar_data[threshold_index][bmark_index] = (energy_bar_data[0][bmark_index] - energy_bar_data[threshold_index][bmark_index]) * 100 / energy_bar_data[0][bmark_index]
performance_bar_data[threshold_index][bmark_index] = (performance_bar_data[threshold_index][bmark_index] - performance_bar_data[0][bmark_index]) * 100 / performance_bar_data[0][bmark_index]
energy_bar_data_nocost[threshold_index][bmark_index] = (energy_bar_data_nocost[0][bmark_index] - energy_bar_data_nocost[threshold_index][bmark_index]) * 100 / energy_bar_data_nocost[0][bmark_index]
performance_bar_data_nocost[threshold_index][bmark_index] = (performance_bar_data_nocost[threshold_index][bmark_index] - performance_bar_data_nocost[0][bmark_index]) * 100 / performance_bar_data_nocost[0][bmark_index]
energy_bar_data = energy_bar_data[1:]
performance_bar_data = performance_bar_data[1:]
energy_bar_data_nocost = energy_bar_data_nocost[1:]
performance_bar_data_nocost = performance_bar_data_nocost[1:]
threshold_list = threshold_list[1:]
bar_colors=['r', 'b', 'y', 'c', 'm', 'g', 'k', 'w']
bar_width = 1.0 / (len(threshold_list)) #Number of thresholds times metrics
offset = 0.1 #10% offset
bar_width = bar_width *0.8 #fill 80% of the space
index = numpy.arange(len(benchmarks))
fig = plt.figure(figsize=(7.5,1.5))
# energy bar
ax1 = fig.add_subplot(1, 4, 2)
ax2 = fig.add_subplot(1, 4, 4)
for threshold_index, cluster_thresh in enumerate(threshold_list):
ax1.bar(index + offset + threshold_index*bar_width, [-1*x for x in energy_bar_data_nocost[threshold_index]], bar_width, label = str(cluster_thresh)+"\%" , color = bar_colors[threshold_index], edgecolor='none', linewidth=0)
ax1.plot([0, len(benchmarks)], [0, 0], linewidth=1, color = 'k')
ax1.set_ylabel("Energy (\%)", labelpad=0)
ax1.set_xlim([0, len(benchmarks)])
ax1.set_ylim([-2.5, 0.25])
ax1.yaxis.grid('on')
ax1.set_axisbelow(True)
x_ticks = [i + offset*3 for i in range(len(benchmarks))]
ax1.set_xticks(x_ticks)
ax1.set_xticklabels(labels, rotation=45)
for threshold_index, cluster_thresh in enumerate(threshold_list):
ax2.bar(index + offset + threshold_index*bar_width, [-1*x for x in energy_bar_data[threshold_index]], bar_width, label = str(cluster_thresh)+"\%" , color = bar_colors[threshold_index], edgecolor='none', linewidth=0)
ax2.plot([0, len(benchmarks)], [0, 0], linewidth=1, color = 'k')
ax2.legend(fontsize=legend_size, bbox_to_anchor=(1.02, 1.), handlelength=1.0, loc=2, borderaxespad=0.,handletextpad=0.1)
ax2.set_ylabel("Energy (\%)", labelpad=0)
ax2.set_xlim([0, len(benchmarks)])
ax2.set_ylim([-2.5, 0.25])
ax2.yaxis.grid('on')
ax2.set_axisbelow(True)
x_ticks = [i + offset*3 for i in range(len(benchmarks))]
ax2.set_xticks(x_ticks)
ax2.set_xticklabels(labels, rotation=45)
ax1.xaxis.set_ticks_position('bottom')
ax1.yaxis.set_ticks_position('left')
ax2.xaxis.set_ticks_position('bottom')
ax2.yaxis.set_ticks_position('left')
# performance bar
ax1 = fig.add_subplot(1, 4, 1)
ax2 = fig.add_subplot(1, 4, 3)
for threshold_index, cluster_thresh in enumerate(threshold_list):
ax1.bar(index + offset + threshold_index*bar_width, [-1*x for x in performance_bar_data_nocost[threshold_index]], bar_width, label = str(cluster_thresh)+"\%" , color = bar_colors[threshold_index], edgecolor='none', linewidth=0)
ax1.plot([0, len(benchmarks)], [-1*x for x in [cluster_thresh, cluster_thresh]], linewidth=1, color = bar_colors[threshold_index], linestyle='--')
ax1.plot([0, len(benchmarks)], [0, 0], linewidth=1, color = 'k')
ax1.set_ylabel("Performance (\%)", labelpad=-1)
ax1.set_xlim([0, len(benchmarks)])
if budget == 1.0:
ax1.set_ylim([-5, 1.5])
elif budget == 1.3:
ax1.set_ylim([-3.5, 1.25])
else:
ax1.set_ylim([-5, 5])
ax1.yaxis.grid('on')
ax1.set_axisbelow(True)
x_ticks = [i + offset*3 for i in range(len(benchmarks))]
ax1.set_xticks(x_ticks)
ax1.set_xticklabels(labels, rotation=45)
for threshold_index, cluster_thresh in enumerate(threshold_list):
ax2.bar(index + offset + threshold_index*bar_width, [-1*x for x in performance_bar_data[threshold_index]], bar_width, label = str(cluster_thresh)+"\%" , color = bar_colors[threshold_index], edgecolor='none', linewidth=0)
ax2.plot([0, len(benchmarks)], [-1*x for x in [cluster_thresh, cluster_thresh]], linewidth=1, color = bar_colors[threshold_index], linestyle='--')
ax2.plot([0, len(benchmarks)], [0, 0], linewidth=1, color = 'k')
ax2.set_ylabel("Performance (\%)", labelpad=-1)
ax2.set_xlim([0, len(benchmarks)])
if budget == 1.0:
ax2.set_ylim([-5, 1.5])
elif budget == 1.3:
ax2.set_ylim([-3.5, 4])
else:
ax2.set_ylim([-5, 5])
ax2.yaxis.grid('on')
ax2.set_axisbelow(True)
x_ticks = [i + offset*3 for i in range(len(benchmarks))]
ax2.set_xticks(x_ticks)
ax2.set_xticklabels(labels, rotation=45)
ax1.xaxis.set_ticks_position('bottom')
ax1.yaxis.set_ticks_position('left')
ax2.xaxis.set_ticks_position('bottom')
ax2.yaxis.set_ticks_position('left')
ax1.text(4.2, -6.5, '(a) No Tuning Overhead')
ax2.text(4, -8.2, '(b) With Tuning Overhead')
outputf = os.path.join(os.path.join(output_dir_path, "energy_perf_bar"), "energy_perf_bar_%s" % (str(budget)))
fig.subplots_adjust(top=0.9, right=0.92, left=0.05, bottom=0.4, wspace=0.38, hspace=0.0)
plt.savefig("%s.pdf" % (outputf))
def plot_abs_energy_time_bar(args, budget_list, cluster_thresh, perf_cost, energy_cost):
dir_path = args.input_dir
output_dir_path = args.output_dir
benchmarks, labels = get_benchmarks(args)
#plot constants
cpuflist = get_cpu_freq_plot_list(args)
memflist = get_mem_freq_plot_list(args)
# dir_path = 'data/70/'
# finding points with 3% threshold of target budget/inefficiency
thresh = 3
energy_bar_data = [ [] for thr in budget_list]
performance_bar_data = [ [] for thr in budget_list]
for budget_index, budget in enumerate(budget_list):
data = []
for bmark in benchmarks:
bmarkDirPath = os.path.join(os.path.join(dir_path, "per_sample_data"), bmark)
frontiers_file = os.path.join(bmarkDirPath, "per_sample_frontiers.json")
frontiers_data = json.loads(open(frontiers_file).read())
data_to_plot=[]
sample_points=[]
cpu_rectangle_points=[]
for data in frontiers_data["data"]:
filtered_data=[]
#filtering all those points which have inefficiency with error% of the budget
for point in data:
if point["inefficiency"] <= budget:
filtered_data.append(point)
elif (point["inefficiency"] - budget) * 100 / budget <= thresh:
filtered_data.append(point)
#finding the point with highest performance among filtered data
optimal_point = filtered_data[0]
for point in filtered_data:
if point["speedup"] > optimal_point["speedup"]:
optimal_point = point
local_rect_points=[]
for point in filtered_data:
if (optimal_point["speedup"] - point["speedup"]) < 0:
print "something is wrong!"
if (optimal_point["speedup"] - point["speedup"]) * 100 / optimal_point["speedup"] <= cluster_thresh:
data_to_plot.append(point)
sample_points.append(frontiers_data["data"].index(data))
local_rect_points.append(point)
cpufpoints = [cpuf for cpuf in [data_to_plot[sample]["cpu_freq"] for sample in range(len(data_to_plot))]]
memfpoints = [memf for memf in [data_to_plot[sample]["mem_freq"] for sample in range(len(data_to_plot))]]
samplepoints = sample_points
#Total number of transitions
num_transitions = 0.0
#Available settings for (CPU, MEM)
settings_available = Set()
index = 0
current_sample = -1
length = 1
lengths = []
prev_tr_sample = 0
energy = 0
performance = 0
while index < len( samplepoints ):
current_sample = samplepoints[index]
#Construct the current settings
current_settings = Set()
while index < len( samplepoints ) and samplepoints[index] == current_sample:
current_settings.add((cpufpoints[index],memfpoints[index]))
index = index + 1
#Compute the common points between current and what is available
common_points = current_settings.intersection(settings_available)
if (len(common_points) == 0 or index == len( samplepoints )):
# find the point with highest cpu and mem freq
if (current_sample !=0 ):
optimal_point = settings_available.pop()
settings_available.add(optimal_point)
for point in settings_available:
if point[0] > optimal_point[0]:
optimal_point = point
elif point[0] == optimal_point[0]:
if point[1] > optimal_point[1]:
optimal_point = point
done = 0
idx = prev_tr_sample+1
while done == 0:
energy += get_energy(frontiers_data, optimal_point[0], optimal_point[1], idx, bmark)
performance += get_performance(frontiers_data, optimal_point[0], optimal_point[1], idx, bmark)
if idx == samplepoints[index-1]:
done = 1
else:
idx += 1
prev_tr_sample = samplepoints[index-1]
# When there are no common points, transition
if (len(common_points) == 0):
settings_available = current_settings #all current settings are now available
#Ignore the transition if it's the first sample
if (current_sample!=0):
num_transitions += 1
else: # Continue with the common points
settings_available = common_points
# Record the length if we need to transition or we are at the end
if (len(common_points) == 0 or index == len( samplepoints )):
#Ignore the length if the transition is for the first sample
if (current_sample!=0):
lengths.append(length)
#reset the length
length = 1
else: #otherwise, just increment the length
length += 1
energy_bar_data[budget_index].append((energy + (num_transitions * energy_cost))/1e6)
performance_bar_data[budget_index].append((performance +(num_transitions * perf_cost))/1e6 )
#print performance
#print energy
for bmark_index, bmark in enumerate(benchmarks):
for budget_index, budget in enumerate(budget_list):
if budget_index != 0:
energy_bar_data[budget_index][bmark_index] = energy_bar_data[budget_index][bmark_index] / energy_bar_data[0][bmark_index]
performance_bar_data[budget_index][bmark_index] = (performance_bar_data[budget_index][bmark_index] / performance_bar_data[0][bmark_index])
energy_bar_data[0][bmark_index] = 1
performance_bar_data[0][bmark_index] = 1
bar_colors=['r', 'b', 'y', 'c', 'm', 'g', 'k', 'w']
bar_width = 1.0 / (len(budget_list)) #Number of thresholds times metrics
offset = 0.1 #10% offset
bar_width = bar_width *0.8 #fill 80% of the space
index = numpy.arange(len(benchmarks))
# energy bar
fig = plt.figure(figsize=(3, 1))
canvas = FigureCanvas(fig)
fig.set_canvas(canvas)
ax = fig.add_subplot('111')
bars = []
for budget_index, budget in enumerate(budget_list):
bar = ax.bar(index + offset + budget_index*bar_width, energy_bar_data[budget_index], bar_width, label = str(budget) if budget < 2 else "\infty" , color = bar_colors[budget_index], edgecolor='none', linewidth=0)
bars.append(bar)
ax.legend(fontsize=legend_size, ncol = 6, bbox_to_anchor=(0., 1.02, 1., .102), handlelength=1.0, handletextpad=0.1, loc=3, borderaxespad=0., mode="expand")
ax.set_ylabel("Consumed \nInefficiency")
ax.set_xlim([0, len(benchmarks)])
ax.set_ylim([0, 2])
ax.grid('on')
ax.set_axisbelow(True)
x_ticks = [i + offset*5 for i in range(len(benchmarks))]
ax.set_xticks(x_ticks)
ax.set_xticklabels(labels, rotation=45)
#fig.legend(bars, budget_list, fontsize=legend_size, bbox_to_anchor=(1.19,1.49), labelspacing=0.15, handletextpad=0.7)
outputf = os.path.join(os.path.join(output_dir_path, "energy_perf_bar"), "energy_bar_normalized_%s_%s_%s" % (str(cluster_thresh), perf_cost, energy_cost))
save_plot(ax, outputf)
canvas.close()
# performance bar
fig = plt.figure(figsize=(3, 1))
canvas = FigureCanvas(fig)
fig.set_canvas(canvas)
ax = fig.add_subplot('111')
bars = []
for budget_index, budget in enumerate(budget_list):
bar = ax.bar(index + offset + budget_index*bar_width, performance_bar_data[budget_index], bar_width, label = str(budget) if budget < 2 else "\infty" , color = bar_colors[budget_index], edgecolor='none', linewidth=0)
bars.append(bar)
ax.legend(fontsize=legend_size, ncol = 6, bbox_to_anchor=(0., 1.02, 1., .102), handlelength=1.0, handletextpad=0.1, loc=3, borderaxespad=0., mode="expand")
ax.set_ylabel("Execution Time \n (Normalized)")
ax.set_xlim([0, len(benchmarks)])
ax.set_ylim([0, 1.2])
ax.grid('on')
ax.set_axisbelow(True)
x_ticks = [i + offset*5 for i in range(len(benchmarks))]
ax.set_xticks(x_ticks)
ax.set_xticklabels(labels, rotation=45)
#fig.legend(bars, budget_list, fontsize=legend_size, bbox_to_anchor=(1.19,1.49), labelspacing=0.15, handletextpad=0.7)
outputf = os.path.join(os.path.join(output_dir_path, "energy_perf_bar"), "performance_bar_normalized_%s_%s_%s" % (str(cluster_thresh), perf_cost, energy_cost))
save_plot(ax, outputf)
canvas.close()
def main(argv):
args = parse(argv)
for thresh in [0.0, 1.0, 5.0]:
# for thresh in [0.0]:
plot_abs_energy_time_bar(args, [1.0,1.1, 1.2, 1.3,1.6,3.0], thresh, 0, 0)
#no tuning overhead
#for ineff in [1.0, 1.3, 1.6]:
for ineff in [1.3]:
plot_energy_perf_bar(args, [0,1,3,5], ineff)
#with tuning overhead - 500us, 30uJ (scaled bzip2 energy of first sample at maxmax to 30us)
# for ineff in [1.0, 1.3, 1.6]:
# for ineff in [1.3]:
# plot_energy_perf_bar(args, [0,1,3,5], ineff)
if __name__ == "__main__":
main(sys.argv)