1PERF-DIFF(1) perf Manual PERF-DIFF(1)
2
6 perf-diff - Read perf.data files and display the differential profile
7
9 perf diff [baseline file] [data file1] [[data file2] ... ]
10
12 This command displays the performance difference amongst two or more
13 perf.data files captured via perf record.
14 If no parameters are passed it will assume perf.data.old and perf.data.
16 The differential profile is displayed only for events matching both
18 specified perf.data files.
19 If no parameters are passed the samples will be sorted by dso and
21 symbol. As the perf.data files could come from different binaries, the
22 symbols addresses could vary. So perf diff is based on the comparison
23 of the files and symbols name.
24
26 -D, --dump-raw-trace
27 Dump raw trace in ASCII.
28 --kallsyms=<file>
30 kallsyms pathname
31 -m, --modules
33 Load module symbols. WARNING: use only with -k and LIVE kernel
34 -d, --dsos=
36 Only consider symbols in these dsos. CSV that understands
37 file://filename entries. This option will affect the percentage of
38 the Baseline/Delta column. See --percentage for more info.
39 -C, --comms=
41 Only consider symbols in these comms. CSV that understands
42 file://filename entries. This option will affect the percentage of
43 the Baseline/Delta column. See --percentage for more info.
44 -S, --symbols=
46 Only consider these symbols. CSV that understands file://filename
47 entries. This option will affect the percentage of the
48 Baseline/Delta column. See --percentage for more info.
49 -s, --sort=
51 Sort by key(s): pid, comm, dso, symbol, cpu, parent, srcline.
52 Please see description of --sort in the perf-report man page.
53 -t, --field-separator=
55 Use a special separator character and don’t pad with spaces,
56 replacing all occurrences of this separator in symbol names (and
57 other output) with a . character, that thus it’s the only non
58 valid separator.
59 -v, --verbose
61 Be verbose, for instance, show the raw counts in addition to the
62 diff.
63 -q, --quiet
65 Do not show any message. (Suppress -v)
66 -f, --force
68 Don’t do ownership validation.
69 --symfs=<directory>
71 Look for files with symbols relative to this directory.
72 -b, --baseline-only
74 Show only items with match in baseline.
75 -c, --compute
77 Differential computation selection - delta, ratio, wdiff, cycles,
78 delta-abs (default is delta-abs). Default can be changed using
79 diff.compute config option. See COMPARISON METHODS section for more
80 info.
81 --cycles-hist
83 Report a histogram and the standard deviation for cycles data. It
84 can help us to judge if the reported cycles data is noisy or not.
85 This option should be used with -c cycles.
86 -p, --period
88 Show period values for both compared hist entries.
89 -F, --formula
91 Show formula for given computation.
92 -o, --order
94 Specify compute sorting column number. 0 means sorting by baseline
95 overhead and 1 (default) means sorting by computed value of column
96 1 (data from the first file other base baseline). Values more than
97 1 can be used only if enough data files are provided. The default
98 value can be set using the diff.order config option.
99 --percentage
101 Determine how to display the overhead percentage of filtered
102 entries. Filters can be applied by --comms, --dsos and/or --symbols
103 options.
104 "relative" means it's relative to filtered entries only so that the
106 sum of shown entries will be always 100%. "absolute" means it retains
107 the original value before and after the filter is applied.
108 --time
110 Analyze samples within given time window. It supports time percent
111 with multiple time ranges. Time string is a%/n,b%/m,... or
112 a%-b%,c%-%d,....
113 For example:
115 Select the second 10% time slice to diff:
117 perf diff --time 10%/2
119 Select from 0% to 10% time slice to diff:
121 perf diff --time 0%-10%
123 Select the first and the second 10% time slices to diff:
125 perf diff --time 10%/1,10%/2
127 Select from 0% to 10% and 30% to 40% slices to diff:
129 perf diff --time 0%-10%,30%-40%
131 It also supports analyzing samples within a given time window
133 <start>,<stop>. Times have the format seconds.nanoseconds. If 'start'
134 is not given (i.e. time string is ',x.y') then analysis starts at
135 the beginning of the file. If stop time is not given (i.e. time
136 string is 'x.y,') then analysis goes to the end of the file.
137 Multiple ranges can be separated by spaces, which requires the argument
138 to be quoted e.g. --time "1234.567,1234.789 1235,"
139 Time string is'a1.b1,c1.d1:a2.b2,c2.d2'. Use ':' to separate timestamps
140 for different perf.data files.
141 For example, we get the timestamp information from 'perf script'.
143 perf script -i perf.data.old
145 mgen 13940 [000] 3946.361400: ...
146 perf script -i perf.data
148 mgen 13940 [000] 3971.150589 ...
149 perf diff --time 3946.361400,:3971.150589,
151 It analyzes the perf.data.old from the timestamp 3946.361400 to
153 the end of perf.data.old and analyzes the perf.data from the
154 timestamp 3971.150589 to the end of perf.data.
155 --cpu
157 Only diff samples for the list of CPUs provided. Multiple CPUs can
158 be provided as a comma-separated list with no space: 0,1. Ranges of
159 CPUs are specified with -: 0-2. Default is to report samples on all
160 CPUs.
161 --pid=
163 Only diff samples for given process ID (comma separated list).
164 --tid=
166 Only diff samples for given thread ID (comma separated list).
167 --stream
169 Enable hot streams comparison. Stream can be a callchain which is
170 aggregated by the branch records from samples.
171
173 The comparison is governed by the baseline file. The baseline perf.data
174 file is iterated for samples. All other perf.data files specified on
175 the command line are searched for the baseline sample pair. If the pair
176 is found, specified computation is made and result is displayed.
177 All samples from non-baseline perf.data files, that do not match any
179 baseline entry, are displayed with empty space within baseline column
180 and possible computation results (delta) in their related column.
181 Example files samples: - file A with samples f1, f2, f3, f4, f6 - file
183 B with samples f2, f4, f5 - file C with samples f1, f2, f5
184 Example output: x - computation takes place for pair b - baseline
186 sample percentage
187 • perf diff A B C
189 baseline/A compute/B compute/C samples
191 ---------------------------------------
192 b x f1
193 b x x f2
194 b f3
195 b x f4
196 b f6
197 x x f5
198 • perf diff B A C
200 baseline/B compute/A compute/C samples
202 ---------------------------------------
203 b x x f2
204 b x f4
205 b x f5
206 x x f1
207 x f3
208 x f6
209 • perf diff C B A
211 baseline/C compute/B compute/A samples
213 ---------------------------------------
214 b x f1
215 b x x f2
216 b x f5
217 x f3
218 x x f4
219 x f6
220
222 delta
223 If specified the Delta column is displayed with value d computed as:
224 d = A->period_percent - B->period_percent
226 with: - A/B being matching hist entry from data/baseline file specified
228 (or perf.data/perf.data.old) respectively.
229 • period_percent being the % of the hist entry period value within
231 single data file
232 • with filtering by -C, -d and/or -S, period_percent might be changed
234 relative to how entries are filtered. Use --percentage=absolute to
235 prevent such fluctuation.
236 delta-abs
238 Same as 'delta` method, but sort the result with the absolute values.
239 ratio
241 If specified the Ratio column is displayed with value r computed as:
242 r = A->period / B->period
244 with: - A/B being matching hist entry from data/baseline file specified
246 (or perf.data/perf.data.old) respectively.
247 • period being the hist entry period value
249 wdiff:WEIGHT-B,WEIGHT-A
251 If specified the Weighted diff column is displayed with value d
252 computed as:
253 d = B->period * WEIGHT-A - A->period * WEIGHT-B
255 • A/B being matching hist entry from data/baseline file specified (or
257 perf.data/perf.data.old) respectively.
258 • period being the hist entry period value
260 • WEIGHT-A/WEIGHT-B being user supplied weights in the the -c option
262 behind : separator like -c wdiff:1,2.
263 • WEIGHT-A being the weight of the data file
265 • WEIGHT-B being the weight of the baseline data file
267 cycles
269 If specified the [Program Block Range] Cycles Diff column is displayed.
270 It displays the cycles difference of same program basic block amongst
271 two perf.data. The program basic block is the code between two
272 branches.
273 [Program Block Range] indicates the range of a program basic block.
275 Source line is reported if it can be found otherwise uses symbol+offset
276 instead.
277
279 perf-record(1), perf-report(1)
280perf 11/22/2021 PERF-DIFF(1)