| 1 | perf-diff(1) |
| 2 | ============ |
| 3 | |
| 4 | NAME |
| 5 | ---- |
| 6 | perf-diff - Read perf.data files and display the differential profile |
| 7 | |
| 8 | SYNOPSIS |
| 9 | -------- |
| 10 | [verse] |
| 11 | 'perf diff' [baseline file] [data file1] [[data file2] ... ] |
| 12 | |
| 13 | DESCRIPTION |
| 14 | ----------- |
| 15 | This command displays the performance difference amongst two or more perf.data |
| 16 | files captured via perf record. |
| 17 | |
| 18 | If no parameters are passed it will assume perf.data.old and perf.data. |
| 19 | |
| 20 | The differential profile is displayed only for events matching both |
| 21 | specified perf.data files. |
| 22 | |
| 23 | If no parameters are passed the samples will be sorted by dso and symbol. |
| 24 | As the perf.data files could come from different binaries, the symbols addresses |
| 25 | could vary. So perf diff is based on the comparison of the files and |
| 26 | symbols name. |
| 27 | |
| 28 | OPTIONS |
| 29 | ------- |
| 30 | -D:: |
| 31 | --dump-raw-trace:: |
| 32 | Dump raw trace in ASCII. |
| 33 | |
| 34 | --kallsyms=<file>:: |
| 35 | kallsyms pathname |
| 36 | |
| 37 | -m:: |
| 38 | --modules:: |
| 39 | Load module symbols. WARNING: use only with -k and LIVE kernel |
| 40 | |
| 41 | -d:: |
| 42 | --dsos=:: |
| 43 | Only consider symbols in these dsos. CSV that understands |
| 44 | file://filename entries. This option will affect the percentage |
| 45 | of the Baseline/Delta column. See --percentage for more info. |
| 46 | |
| 47 | -C:: |
| 48 | --comms=:: |
| 49 | Only consider symbols in these comms. CSV that understands |
| 50 | file://filename entries. This option will affect the percentage |
| 51 | of the Baseline/Delta column. See --percentage for more info. |
| 52 | |
| 53 | -S:: |
| 54 | --symbols=:: |
| 55 | Only consider these symbols. CSV that understands |
| 56 | file://filename entries. This option will affect the percentage |
| 57 | of the Baseline/Delta column. See --percentage for more info. |
| 58 | |
| 59 | -s:: |
| 60 | --sort=:: |
| 61 | Sort by key(s): pid, comm, dso, symbol, cpu, parent, srcline. |
| 62 | Please see description of --sort in the perf-report man page. |
| 63 | |
| 64 | -t:: |
| 65 | --field-separator=:: |
| 66 | |
| 67 | Use a special separator character and don't pad with spaces, replacing |
| 68 | all occurrences of this separator in symbol names (and other output) |
| 69 | with a '.' character, that thus it's the only non valid separator. |
| 70 | |
| 71 | -v:: |
| 72 | --verbose:: |
| 73 | Be verbose, for instance, show the raw counts in addition to the |
| 74 | diff. |
| 75 | |
| 76 | -q:: |
| 77 | --quiet:: |
| 78 | Do not show any warnings or messages. (Suppress -v) |
| 79 | |
| 80 | -f:: |
| 81 | --force:: |
| 82 | Don't do ownership validation. |
| 83 | |
| 84 | --symfs=<directory[,layout]>:: |
| 85 | Look for files with symbols relative to this directory. The optional |
| 86 | layout can be 'hierarchy' (default, matches full path) or 'flat' |
| 87 | (only matches base name). This is useful when debug files are stored |
| 88 | in a flat directory structure. |
| 89 | |
| 90 | -b:: |
| 91 | --baseline-only:: |
| 92 | Show only items with match in baseline. |
| 93 | |
| 94 | -c:: |
| 95 | --compute:: |
| 96 | Differential computation selection - delta, ratio, wdiff, cycles, |
| 97 | delta-abs (default is delta-abs). Default can be changed using |
| 98 | diff.compute config option. See COMPARISON METHODS section for |
| 99 | more info. |
| 100 | |
| 101 | --cycles-hist:: |
| 102 | Report a histogram and the standard deviation for cycles data. |
| 103 | It can help us to judge if the reported cycles data is noisy or |
| 104 | not. This option should be used with '-c cycles'. |
| 105 | |
| 106 | -p:: |
| 107 | --period:: |
| 108 | Show period values for both compared hist entries. |
| 109 | |
| 110 | -F:: |
| 111 | --formula:: |
| 112 | Show formula for given computation. |
| 113 | |
| 114 | -o:: |
| 115 | --order:: |
| 116 | Specify compute sorting column number. 0 means sorting by baseline |
| 117 | overhead and 1 (default) means sorting by computed value of column 1 |
| 118 | (data from the first file other base baseline). Values more than 1 |
| 119 | can be used only if enough data files are provided. |
| 120 | The default value can be set using the diff.order config option. |
| 121 | |
| 122 | --percentage:: |
| 123 | Determine how to display the overhead percentage of filtered entries. |
| 124 | Filters can be applied by --comms, --dsos and/or --symbols options. |
| 125 | |
| 126 | "relative" means it's relative to filtered entries only so that the |
| 127 | sum of shown entries will be always 100%. "absolute" means it retains |
| 128 | the original value before and after the filter is applied. |
| 129 | |
| 130 | --time:: |
| 131 | Analyze samples within given time window. It supports time |
| 132 | percent with multiple time ranges. Time string is 'a%/n,b%/m,...' |
| 133 | or 'a%-b%,c%-%d,...'. |
| 134 | |
| 135 | For example: |
| 136 | |
| 137 | Select the second 10% time slice to diff: |
| 138 | |
| 139 | perf diff --time 10%/2 |
| 140 | |
| 141 | Select from 0% to 10% time slice to diff: |
| 142 | |
| 143 | perf diff --time 0%-10% |
| 144 | |
| 145 | Select the first and the second 10% time slices to diff: |
| 146 | |
| 147 | perf diff --time 10%/1,10%/2 |
| 148 | |
| 149 | Select from 0% to 10% and 30% to 40% slices to diff: |
| 150 | |
| 151 | perf diff --time 0%-10%,30%-40% |
| 152 | |
| 153 | It also supports analyzing samples within a given time window |
| 154 | <start>,<stop>. Times have the format seconds.nanoseconds. If 'start' |
| 155 | is not given (i.e. time string is ',x.y') then analysis starts at |
| 156 | the beginning of the file. If stop time is not given (i.e. time |
| 157 | string is 'x.y,') then analysis goes to the end of the file. |
| 158 | Multiple ranges can be separated by spaces, which requires the argument |
| 159 | to be quoted e.g. --time "1234.567,1234.789 1235," |
| 160 | Time string is'a1.b1,c1.d1:a2.b2,c2.d2'. Use ':' to separate timestamps |
| 161 | for different perf.data files. |
| 162 | |
| 163 | For example, we get the timestamp information from 'perf script'. |
| 164 | |
| 165 | perf script -i perf.data.old |
| 166 | mgen 13940 [000] 3946.361400: ... |
| 167 | |
| 168 | perf script -i perf.data |
| 169 | mgen 13940 [000] 3971.150589 ... |
| 170 | |
| 171 | perf diff --time 3946.361400,:3971.150589, |
| 172 | |
| 173 | It analyzes the perf.data.old from the timestamp 3946.361400 to |
| 174 | the end of perf.data.old and analyzes the perf.data from the |
| 175 | timestamp 3971.150589 to the end of perf.data. |
| 176 | |
| 177 | --cpu:: Only diff samples for the list of CPUs provided. Multiple CPUs can |
| 178 | be provided as a comma-separated list with no space: 0,1. Ranges of |
| 179 | CPUs are specified with -: 0-2. Default is to report samples on all |
| 180 | CPUs. |
| 181 | |
| 182 | --pid=:: |
| 183 | Only diff samples for given process ID (comma separated list). |
| 184 | |
| 185 | --tid=:: |
| 186 | Only diff samples for given thread ID (comma separated list). |
| 187 | |
| 188 | --stream:: |
| 189 | Enable hot streams comparison. Stream can be a callchain which is |
| 190 | aggregated by the branch records from samples. |
| 191 | |
| 192 | COMPARISON |
| 193 | ---------- |
| 194 | The comparison is governed by the baseline file. The baseline perf.data |
| 195 | file is iterated for samples. All other perf.data files specified on |
| 196 | the command line are searched for the baseline sample pair. If the pair |
| 197 | is found, specified computation is made and result is displayed. |
| 198 | |
| 199 | All samples from non-baseline perf.data files, that do not match any |
| 200 | baseline entry, are displayed with empty space within baseline column |
| 201 | and possible computation results (delta) in their related column. |
| 202 | |
| 203 | Example files samples: |
| 204 | - file A with samples f1, f2, f3, f4, f6 |
| 205 | - file B with samples f2, f4, f5 |
| 206 | - file C with samples f1, f2, f5 |
| 207 | |
| 208 | Example output: |
| 209 | x - computation takes place for pair |
| 210 | b - baseline sample percentage |
| 211 | |
| 212 | - perf diff A B C |
| 213 | |
| 214 | baseline/A compute/B compute/C samples |
| 215 | --------------------------------------- |
| 216 | b x f1 |
| 217 | b x x f2 |
| 218 | b f3 |
| 219 | b x f4 |
| 220 | b f6 |
| 221 | x x f5 |
| 222 | |
| 223 | - perf diff B A C |
| 224 | |
| 225 | baseline/B compute/A compute/C samples |
| 226 | --------------------------------------- |
| 227 | b x x f2 |
| 228 | b x f4 |
| 229 | b x f5 |
| 230 | x x f1 |
| 231 | x f3 |
| 232 | x f6 |
| 233 | |
| 234 | - perf diff C B A |
| 235 | |
| 236 | baseline/C compute/B compute/A samples |
| 237 | --------------------------------------- |
| 238 | b x f1 |
| 239 | b x x f2 |
| 240 | b x f5 |
| 241 | x f3 |
| 242 | x x f4 |
| 243 | x f6 |
| 244 | |
| 245 | COMPARISON METHODS |
| 246 | ------------------ |
| 247 | delta |
| 248 | ~~~~~ |
| 249 | If specified the 'Delta' column is displayed with value 'd' computed as: |
| 250 | |
| 251 | d = A->period_percent - B->period_percent |
| 252 | |
| 253 | with: |
| 254 | - A/B being matching hist entry from data/baseline file specified |
| 255 | (or perf.data/perf.data.old) respectively. |
| 256 | |
| 257 | - period_percent being the % of the hist entry period value within |
| 258 | single data file |
| 259 | |
| 260 | - with filtering by -C, -d and/or -S, period_percent might be changed |
| 261 | relative to how entries are filtered. Use --percentage=absolute to |
| 262 | prevent such fluctuation. |
| 263 | |
| 264 | delta-abs |
| 265 | ~~~~~~~~~ |
| 266 | Same as 'delta` method, but sort the result with the absolute values. |
| 267 | |
| 268 | ratio |
| 269 | ~~~~~ |
| 270 | If specified the 'Ratio' column is displayed with value 'r' computed as: |
| 271 | |
| 272 | r = A->period / B->period |
| 273 | |
| 274 | with: |
| 275 | - A/B being matching hist entry from data/baseline file specified |
| 276 | (or perf.data/perf.data.old) respectively. |
| 277 | |
| 278 | - period being the hist entry period value |
| 279 | |
| 280 | wdiff:WEIGHT-B,WEIGHT-A |
| 281 | ~~~~~~~~~~~~~~~~~~~~~~~ |
| 282 | If specified the 'Weighted diff' column is displayed with value 'd' computed as: |
| 283 | |
| 284 | d = B->period * WEIGHT-A - A->period * WEIGHT-B |
| 285 | |
| 286 | - A/B being matching hist entry from data/baseline file specified |
| 287 | (or perf.data/perf.data.old) respectively. |
| 288 | |
| 289 | - period being the hist entry period value |
| 290 | |
| 291 | - WEIGHT-A/WEIGHT-B being user supplied weights in the '-c' option |
| 292 | behind ':' separator like '-c wdiff:1,2'. |
| 293 | - WEIGHT-A being the weight of the data file |
| 294 | - WEIGHT-B being the weight of the baseline data file |
| 295 | |
| 296 | cycles |
| 297 | ~~~~~~ |
| 298 | If specified the '[Program Block Range] Cycles Diff' column is displayed. |
| 299 | It displays the cycles difference of same program basic block amongst |
| 300 | two perf.data. The program basic block is the code between two branches. |
| 301 | |
| 302 | '[Program Block Range]' indicates the range of a program basic block. |
| 303 | Source line is reported if it can be found otherwise uses symbol+offset |
| 304 | instead. |
| 305 | |
| 306 | SEE ALSO |
| 307 | -------- |
| 308 | linkperf:perf-record[1], linkperf:perf-report[1] |