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
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
|
#!/usr/bin/python
#
# Intel GPU Top data analyser
#
import sys
import os
import getopt
import tempfile
HEADER="""
<html>
<head>
<title>Analysis for %(title)s</title>
<style>
body {
font-family: arial;
}
p {
margin-left: 10px;
}
</style>
</head>
<body>
<h1>Analysis for %(title)s</h1>
<h2>Summary:</h2>
<p>
<h3>Execution lasted %(duration).2f seconds</h3>
"""
PERF_REFERENCE="""
<p>
<a href="%(perf)s">Kernel perf results analysis</a>
</p>
"""
MINMAXAVG="""
<p>
<h3>%(descr)s:</h3>
<b>Minimum:</b> %(min)s <br/>
<b>Maximum:</b> %(max)s <br/>
<b>Average:</b> %(avg)s <br/>
</p>
"""
FIGURE="""
<p>
<hr/>
<a href="%(img)s"><img src="%(img)s" width=800 height=600></a>
<br/>
<b>%(title)s</b>
</p>
"""
PERF_ITERATION="""
<hr/>
<a name="%(sec)d">
<h3>Second %(sec)d</h3>
</a>
Processes in execution:
"""
PERF_SECOND_TITLE="""
<a href="#%(sec)d">Second %(sec)d</a>
<br/>
"""
PERF_PROCESS_REF="""
<a href="#%(sec)d.%(process)s">%(process)s</a>
"""
PERF_PROCESS="""
<a name="%(sec)d.%(process)s">
<h3>Process %(process)s</h3>
</a>
"""
PERF_TOP="""
<pre>
%(top)s
</pre>
"""
TAIL="""
</body>
</html>
"""
def print_header(output):
"""Prints the header"""
print >>output, HEADER
def collect(fd):
"""Collects statistics for a given command"""
columns={}
results = {}
while True:
line = fd.readline()
if not line:
break
line.strip()
if line[0] == "#":
# detecting column names
cols = line[1:].strip().split('\t')
for pos in range(len(cols)):
columns[cols[pos]] = pos
for col in columns:
results[col] = []
continue
data = line.split()
# fill in results from the headers
for item in columns.keys():
pos = columns[item]
results[item].append(float(data[pos]))
return results
def analyse_perf(logfile, out_dir, perf="perf.html"):
"""Analyses perf results"""
if not os.access(logfile, os.R_OK):
print "Error: unable to access perf log file %s" % logfile
return
perf_data = os.popen("perf script -f comm,pid,time,ip,sym -i %s 2>/dev/null" % logfile)
if not perf_data:
print "Error: unable to process perf log %s" % logfile
return
results = {}
time_start = -1
time_prev = -1
while 1:
line = perf_data.readline()
if not line:
break
fields = line.strip().split()
process = fields[0]
pid = int(fields[1])
time = float(fields[2][:-1]) # remove the trailing ':'
ip = fields[3]
if len(fields) > 4:
function = fields[4]
else:
function = "unknown"
# calculate time
if time_start == -1:
time_start = int(time) # floor down to the corresponding second
# are we on the next second already?
if time - time_start > 1.0:
time_start = int(time)
if time_start not in results:
results[time_start] = {}
# is the current process being tracked?
if process not in results[time_start]:
results[time_start][process]={}
# calculate times for functions
if function not in results[time_start][process]:
results[time_start][process][function] = 0
# ok, so now calculate per-function per-process stats
results[time_start][process][function] += (time - time_prev)
time_prev = time
# all done, now process the results
seconds = results.keys()
seconds.sort()
output = open("%s/%s" % (out_dir, perf), "w")
print >>output, HEADER % {'title': 'Perf results for %s' % logfile,
'duration': len(seconds)
}
for sec in seconds:
# print TOC
print >>output, PERF_SECOND_TITLE % {'sec': sec - seconds[0]}
for sec in seconds:
print >>output, PERF_ITERATION % {'sec': sec - seconds[0], 'processes': ', '.join(results[sec].keys())}
for process in results[sec]:
print >>output, PERF_PROCESS_REF % {'sec': sec - seconds[0], 'process': process}
for process in results[sec]:
print >>output, PERF_PROCESS % {'sec': sec - seconds[0], 'process': process}
# let's sort functions
functions_by_time = sorted(results[sec][process], key=lambda key: results[sec][process][key], reverse=True)
top = ""
for function in functions_by_time:
top += "%.6f\t\t%s\n" % (results[sec][process][function], function)
print >>output, PERF_TOP % { 'top': top }
print >>output, TAIL
output.close()
def plot_series(timeline, values, results, axis, title, linestyle='solid'):
"""A helper function to simplify plotting of a function
from the array of results. It checks if the column does, in fact,
exists; and if it does, carries on with plotting"""
for column in [timeline, values]:
if column not in results:
print "Column %s not available for plotting" % values
return
axis.plot(results[timeline], results[values], label=title, linestyle=linestyle)
def analyse(results, title, out_dir, perf_logfile=None, summary="index.html"):
"""Analyses intel_gpu_top results"""
# calculate min/max/avg values
keys = results.keys()
for iter in keys:
if not results[iter]:
print "ERROR: no results collected for '%s', skipping" % iter
continue
results["%s_min" % iter] = min(results[iter])
results["%s_max" % iter] = max(results[iter])
results["%s_avg" % iter] = sum(results[iter]) / len(results[iter])
# start composing the output
output = open("%s/%s" % (out_dir, summary), "w")
print >>output, HEADER % {'title': title,
'duration': results['time'][-1]
}
# print summaries
for iter, descr in [('utime', 'User time % of CPU'),
('stime', 'System time % of CPU'),
('power', 'Power consumption (W)'),
('render', 'Render engine usage % of GPU'),
('render.ops', 'Render engine ops per interval'),
('bitstream', 'Bitstream engine usage % of GPU'),
('bitstream.ops', 'Bitstream engine ops per interval'),
('bitstream6', 'Bitstream 6 engine usage % of GPU'),
('bitstream6.ops', 'Bitstream 6 engine ops per interval'),
('blitter', 'Blitter engine usage % of GPU'),
('blitter.ops', 'Blitter engine ops per interval'),
]:
if iter not in results:
print "Column %s not present in results, skipping" % iter
continue
minval = results['%s_min' % iter]
maxval = results['%s_max' % iter]
avgval = results['%s_avg' % iter]
if minval == maxval == avgval:
# No variation in data, skipping
print "No variations in %s, skipping" % iter
continue
minval_s = "%.2f" % minval
maxval_s = "%.2f" % maxval
avgval_s = "%.2f" % avgval
print >>output, MINMAXAVG % {
'descr': descr,
'min': minval_s,
'max': maxval_s,
'avg': avgval_s,
}
# Do we have perf results?
if perf_logfile:
print >>output, PERF_REFERENCE % {'perf': 'perf.html'}
# graphics
try:
import pylab
except:
print "Error: unable to import pylab: %s" % sys.exc_value
return
fig = pylab.figure()
ax = pylab.subplot(111)
box = ax.get_position()
pylab.title("Summary of CPU/GPU/Power usage")
pylab.ylabel("Usage (%)")
pylab.xlabel("Time (s)")
plot_series('time', 'utime', results, ax, "User time")
plot_series('time', 'stime', results, ax, "System time")
num_axis = 2
for ring in ["render", "bitstream", "bitstream6", "blitter"]:
if results.get("%s_avg" % ring, -1) == -1:
print "No data for %s, skipping" % ring
continue
plot_series('time', ring, results, ax, ring)
num_axis += 1
# Do we have power?
if results.get("power_avg", -1) != -1:
# plotting power
ax2 = ax.twinx()
plot_series('time', 'power', results, ax2, "Power", "dotted")
ax2.set_ylabel('Watts')
ax2.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax2.legend(loc = 'upper center', ncol=3, fancybox=True, shadow=True, bbox_to_anchor = (0.5, 0.0))
pylab.grid()
# Shink current axis's height by 10% on the bottom
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax.legend(loc = 'upper center', ncol=num_axis, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
pylab.savefig("%s/plot_summary.svg" % out_dir, format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_summary.svg',
'title': 'Summary of the execution, outlining the CPU usage (in user and system mode), per-ring GPU usage, and power usage (if available) '
}
# graphics ops/per/second
fig = pylab.figure()
ax = pylab.subplot(111)
pylab.title("GPU rings ops per interval")
pylab.ylabel("Ops (n)")
pylab.xlabel("Time (s)")
num_axis = 0
for ring in ["render.ops", "bitstream.ops", "bitstream6.ops", "blitter.ops"]:
plot_series('time', ring, results, ax, ring)
num_axis += 1
pylab.grid()
# Shink current axis's height by 10% on the bottom
box = ax.get_position()
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax.legend(loc = 'upper center', ncol=num_axis, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
pylab.savefig("%s/plot_gpu_ops.svg" % out_dir, format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_gpu_ops.svg',
'title': 'Ops per interval for each GPU ring during the execution'
}
# power
fig = pylab.figure()
pylab.title("Power usage summary")
pylab.ylabel("Power (W)")
pylab.xlabel("Time (s)")
ax = pylab.subplot(111)
plot_series('time', 'power', results, ax, "Power")
plot_series('time', 'power.chipset', results, ax, "Chipset")
plot_series('time', 'power.gfx', results, ax, "GFX")
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax.legend(loc = 'upper center', ncol=3, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
pylab.grid()
# Shink current axis's height by 10% on the bottom
pylab.savefig("%s/plot_power.svg" % out_dir, format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_power.svg',
'title': 'Power usage over the course of execution'
}
# power vs CPU
fig = pylab.figure()
ax = pylab.subplot(111)
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
pylab.title("CPU vs Power usage")
pylab.ylabel("Usage (%)")
pylab.xlabel("Time (s)")
plot_series('time', 'utime', results, ax, "User time")
plot_series('time', 'stime', results, ax, "System time")
# plotting power
if 'power' in results:
ax2 = ax.twinx()
plot_series('time', 'power', results, ax2, "Power", "dashed")
ax2.set_ylabel('Watts')
ax2.legend(loc = 'upper center', ncol=1, fancybox=True, shadow=True, bbox_to_anchor = (0.5, 0.0))
pylab.grid()
# Shink current axis's height by 10% on the bottom
box = ax.get_position()
ax.legend(loc = 'upper center', ncol=3, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
pylab.savefig("%s/plot_power_cpu.svg" % out_dir, format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_power_cpu.svg',
'title': 'Power utilization co-related to CPU'
}
# power vs GPU
fig = pylab.figure()
ax = pylab.subplot(111)
pylab.title("GPU vs Power usage")
pylab.ylabel("Usage (%)")
pylab.xlabel("Time (s)")
num_axis = 0
for ring in ["render", "bitstream", "bitstream6", "blitter"]:
plot_series('time', ring, results, ax, ring, "dashed")
num_axis += 1
# Do we have power?
# plotting power
ax2 = ax.twinx()
plot_series('time', 'power', results, ax2, "Power")
plot_series('time', 'power.chipset', results, ax2, "Chipset")
plot_series('time', 'power.gfx', results, ax2, "GFX")
ax2.set_ylabel('Watts')
num_axis += 1
pylab.grid()
# Shink current axis's height by 10% on the bottom
box = ax.get_position()
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax2.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax.legend(loc = 'upper center', ncol=num_axis, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
ax2.legend(loc = 'upper center', ncol=3, fancybox=True, shadow=True, bbox_to_anchor = (0.5, 0.0))
pylab.savefig("%s/plot_power_gpu.svg" % out_dir, format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_power_gpu.svg',
'title': 'Power utilization co-related to all GPU rings'
}
# per-ring power
for ring in ["render", "bitstream", "bitstream6", "blitter"]:
fig = pylab.figure()
ax = pylab.subplot(111)
box = ax.get_position()
ax.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
pylab.title("Power usage vs %s ring" % ring)
pylab.ylabel("Usage (%)")
pylab.xlabel("Time (s)")
plot_series('time', ring, results, ax, ring, "dashed")
# plotting power
ax2 = ax.twinx()
plot_series('time', 'power', results, ax2, "Power")
plot_series('time', 'power.chipset', results, ax2, "Chipset")
plot_series('time', 'power.gfx', results, ax2, "GFX")
ax2.set_ylabel('Watts')
pylab.grid()
# Shink current axis's height by 10% on the bottom
ax2.set_position([box.x0, box.y0 + box.height * 0.1, box.width, box.height * 0.9])
ax.legend(loc = 'upper center', ncol=1, fancybox=True, shadow=True, bbox_to_anchor = (0.5, -0.1))
ax2.legend(loc = 'upper center', ncol=3, fancybox=True, shadow=True, bbox_to_anchor = (0.5, 0.0))
pylab.savefig("%s/plot_power_gpu_%s.svg" % (out_dir, ring), format="svg", dpi=200)
print >>output, FIGURE % {'img': 'plot_power_gpu_%s.svg' % ring,
'title': 'Power utilization co-related to the %s ring' % ring
}
pass
print >>output, TAIL
output.close()
def usage(cmd):
"""Prints usage message"""
print """Intel-gpu-analyser: intel GPU data analyser
Usage: %s [parameters].
The following arguments are accepted:
-h, --help display help message
-l, --logfile intel-gpu-top log file to analyse
-c, --command command to profile
-t, --title description of the command analysed
-o, --output output directory for the results
-p, --perf kernel perf log file to analyse.
You need to specify either a log file to analyse, or a command to profile,
or a perf log file to process.
When you specify both -l and -c, the last one takes predence.
If you do not give a log file to analyse, or a command to profile, this
tool will become sad, depressed and suicide itself with exit code of -1, and
you will feel bad about it.
""" % cmd
if __name__ == "__main__":
# parse command line
try:
opt, args = getopt.getopt(sys.argv[1:], 'hl:c:t:o:p:', ['help', 'logfile=', 'command=', 'title=', 'output=', 'perf='])
except getopt.error:
usage(sys.argv[0])
sys.exit(1)
logfile = None
perf_logfile = None
title = None
output = os.curdir
for o in opt:
# help
if o[0] == '-h' or o[0] == '--help':
usage(sys.argv[0])
sys.exit(0)
# list
elif o[0] == '-p' or o[0] == '--perf':
perf_logfile = o[1]
print "Analysing perf log file: %s" % perf_logfile
elif o[0] == '-l' or o[0] == '--logfile':
logfile = open(o[1], "r")
title = "Log file '%s'" % o[1]
elif o[0] == '-c' or o[0] == '--command':
logfile = os.popen("intel_gpu_top -o - -e \"%s\"" % o[1])
title = "Execution of '%s'" % o[1]
# force new level
elif o[0] == '-t' or o[0] == '--title':
title = o[1]
elif o[0] == '-o' or o[0] == '--output':
output = o[1]
if not logfile and not perf_logfile:
usage(sys.argv[0])
print "Error: no log file and no command to profile, don't know what to do"
sys.exit(1)
try:
os.makedirs(output)
except:
pass
if logfile:
results = collect(logfile)
analyse(results, title, output, perf_logfile=perf_logfile)
if perf_logfile:
analyse_perf(perf_logfile, output)
|