Austin Schuh | 745610d | 2015-09-06 18:19:50 -0700 | [diff] [blame] | 1 | <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> |
| 2 | <HTML> |
| 3 | |
| 4 | <HEAD> |
| 5 | <link rel="stylesheet" href="designstyle.css"> |
| 6 | <title>Google CPU Profiler Binary Data File Format</title> |
| 7 | </HEAD> |
| 8 | |
| 9 | <BODY> |
| 10 | |
| 11 | <h1>Google CPU Profiler Binary Data File Format</h1> |
| 12 | |
| 13 | <p align=right> |
| 14 | <i>Last modified |
| 15 | <script type=text/javascript> |
| 16 | var lm = new Date(document.lastModified); |
| 17 | document.write(lm.toDateString()); |
| 18 | </script></i> |
| 19 | </p> |
| 20 | |
| 21 | <p>This file documents the binary data file format produced by the |
| 22 | Google CPU Profiler. For information about using the CPU Profiler, |
| 23 | see <a href="cpuprofile.html">its user guide</a>. |
| 24 | |
| 25 | <p>The profiler source code, which generates files using this format, is at |
| 26 | <code>src/profiler.cc</code></a>. |
| 27 | |
| 28 | |
| 29 | <h2>CPU Profile Data File Structure</h2> |
| 30 | |
| 31 | <p>CPU profile data files each consist of four parts, in order: |
| 32 | |
| 33 | <ul> |
| 34 | <li> Binary header |
| 35 | <li> Binary profile records |
| 36 | <li> Binary trailer |
| 37 | <li> Text list of mapped objects |
| 38 | </ul> |
| 39 | |
| 40 | <p>The binary data is expressed in terms of "slots." These are words |
| 41 | large enough to hold the program's pointer type, i.e., for 32-bit |
| 42 | programs they are 4 bytes in size, and for 64-bit programs they are 8 |
| 43 | bytes. They are stored in the profile data file in the native byte |
| 44 | order (i.e., little-endian for x86 and x86_64). |
| 45 | |
| 46 | |
| 47 | <h2>Binary Header</h2> |
| 48 | |
| 49 | <p>The binary header format is show below. Values written by the |
| 50 | profiler, along with requirements currently enforced by the analysis |
| 51 | tools, are shown in parentheses. |
| 52 | |
| 53 | <p> |
| 54 | <table summary="Header Format" |
| 55 | frame="box" rules="sides" cellpadding="5" width="50%"> |
| 56 | <tr> |
| 57 | <th width="30%">slot</th> |
| 58 | <th width="70%">data</th> |
| 59 | </tr> |
| 60 | |
| 61 | <tr> |
| 62 | <td>0</td> |
| 63 | <td>header count (0; must be 0)</td> |
| 64 | </tr> |
| 65 | |
| 66 | <tr> |
| 67 | <td>1</td> |
| 68 | <td>header slots after this one (3; must be >= 3)</td> |
| 69 | </tr> |
| 70 | |
| 71 | <tr> |
| 72 | <td>2</td> |
| 73 | <td>format version (0; must be 0)</td> |
| 74 | </tr> |
| 75 | |
| 76 | <tr> |
| 77 | <td>3</td> |
| 78 | <td>sampling period, in microseconds</td> |
| 79 | </tr> |
| 80 | |
| 81 | <tr> |
| 82 | <td>4</td> |
| 83 | <td>padding (0)</td> |
| 84 | </tr> |
| 85 | </table> |
| 86 | |
| 87 | <p>The headers currently generated for 32-bit and 64-bit little-endian |
| 88 | (x86 and x86_64) profiles are shown below, for comparison. |
| 89 | |
| 90 | <p> |
| 91 | <table summary="Header Example" frame="box" rules="sides" cellpadding="5"> |
| 92 | <tr> |
| 93 | <th></th> |
| 94 | <th>hdr count</th> |
| 95 | <th>hdr words</th> |
| 96 | <th>version</th> |
| 97 | <th>sampling period</th> |
| 98 | <th>pad</th> |
| 99 | </tr> |
| 100 | <tr> |
| 101 | <td>32-bit or 64-bit (slots)</td> |
| 102 | <td>0</td> |
| 103 | <td>3</td> |
| 104 | <td>0</td> |
| 105 | <td>10000</td> |
| 106 | <td>0</td> |
| 107 | </tr> |
| 108 | <tr> |
| 109 | <td>32-bit (4-byte words in file)</td> |
| 110 | <td><tt>0x00000</tt></td> |
| 111 | <td><tt>0x00003</tt></td> |
| 112 | <td><tt>0x00000</tt></td> |
| 113 | <td><tt>0x02710</tt></td> |
| 114 | <td><tt>0x00000</tt></td> |
| 115 | </tr> |
| 116 | <tr> |
| 117 | <td>64-bit LE (4-byte words in file)</td> |
| 118 | <td><tt>0x00000 0x00000</tt></td> |
| 119 | <td><tt>0x00003 0x00000</tt></td> |
| 120 | <td><tt>0x00000 0x00000</tt></td> |
| 121 | <td><tt>0x02710 0x00000</tt></td> |
| 122 | <td><tt>0x00000 0x00000</tt></td> |
| 123 | </tr> |
| 124 | </table> |
| 125 | |
| 126 | <p>The contents are shown in terms of slots, and in terms of 4-byte |
| 127 | words in the profile data file. The slot contents for 32-bit and |
| 128 | 64-bit headers are identical. For 32-bit profiles, the 4-byte word |
| 129 | view matches the slot view. For 64-bit profiles, each (8-byte) slot |
| 130 | is shown as two 4-byte words, ordered as they would appear in the |
| 131 | file. |
| 132 | |
| 133 | <p>The profiling tools examine the contents of the file and use the |
| 134 | expected locations and values of the header words field to detect |
| 135 | whether the file is 32-bit or 64-bit. |
| 136 | |
| 137 | |
| 138 | <h2>Binary Profile Records</h2> |
| 139 | |
| 140 | <p>The binary profile record format is shown below. |
| 141 | |
| 142 | <p> |
| 143 | <table summary="Profile Record Format" |
| 144 | frame="box" rules="sides" cellpadding="5" width="50%"> |
| 145 | <tr> |
| 146 | <th width="30%">slot</th> |
| 147 | <th width="70%">data</th> |
| 148 | </tr> |
| 149 | |
| 150 | <tr> |
| 151 | <td>0</td> |
| 152 | <td>sample count, must be >= 1</td> |
| 153 | </tr> |
| 154 | |
| 155 | <tr> |
| 156 | <td>1</td> |
| 157 | <td>number of call chain PCs (num_pcs), must be >= 1</td> |
| 158 | </tr> |
| 159 | |
| 160 | <tr> |
| 161 | <td>2 .. (num_pcs + 1)</td> |
| 162 | <td>call chain PCs, most-recently-called function first. |
| 163 | </tr> |
| 164 | </table> |
| 165 | |
| 166 | <p>The total length of a given record is 2 + num_pcs. |
| 167 | |
| 168 | <p>Note that multiple profile records can be emitted by the profiler |
| 169 | having an identical call chain. In that case, analysis tools should |
| 170 | sum the counts of all records having identical call chains. |
| 171 | |
| 172 | <p><b>Note:</b> Some profile analysis tools terminate if they see |
| 173 | <em>any</em> profile record with a call chain with its first entry |
| 174 | having the address 0. (This is similar to the binary trailer.) |
| 175 | |
| 176 | <h3>Example</h3> |
| 177 | |
| 178 | This example shows the slots contained in a sample profile record. |
| 179 | |
| 180 | <p> |
| 181 | <table summary="Profile Record Example" |
| 182 | frame="box" rules="sides" cellpadding="5"> |
| 183 | <tr> |
| 184 | <td>5</td> |
| 185 | <td>3</td> |
| 186 | <td>0xa0000</td> |
| 187 | <td>0xc0000</td> |
| 188 | <td>0xe0000</td> |
| 189 | </tr> |
| 190 | </table> |
| 191 | |
| 192 | <p>In this example, 5 ticks were received at PC 0xa0000, whose |
| 193 | function had been called by the function containing 0xc0000, which had |
| 194 | been called from the function containing 0xe0000. |
| 195 | |
| 196 | |
| 197 | <h2>Binary Trailer</h2> |
| 198 | |
| 199 | <p>The binary trailer consists of three slots of data with fixed |
| 200 | values, shown below. |
| 201 | |
| 202 | <p> |
| 203 | <table summary="Trailer Format" |
| 204 | frame="box" rules="sides" cellpadding="5" width="50%"> |
| 205 | <tr> |
| 206 | <th width="30%">slot</th> |
| 207 | <th width="70%">value</th> |
| 208 | </tr> |
| 209 | |
| 210 | <tr> |
| 211 | <td>0</td> |
| 212 | <td>0</td> |
| 213 | </tr> |
| 214 | |
| 215 | <tr> |
| 216 | <td>1</td> |
| 217 | <td>1</td> |
| 218 | </tr> |
| 219 | |
| 220 | <tr> |
| 221 | <td>2</td> |
| 222 | <td>0</td> |
| 223 | </tr> |
| 224 | </table> |
| 225 | |
| 226 | <p>Note that this is the same data that would contained in a profile |
| 227 | record with sample count = 0, num_pcs = 1, and a one-element call |
| 228 | chain containing the address 0. |
| 229 | |
| 230 | |
| 231 | <h2>Text List of Mapped Objects</h2> |
| 232 | |
| 233 | <p>The binary data in the file is followed immediately by a list of |
| 234 | mapped objects. This list consists of lines of text separated by |
| 235 | newline characters. |
| 236 | |
| 237 | <p>Each line is one of the following types: |
| 238 | |
| 239 | <ul> |
| 240 | <li>Build specifier, starting with "<tt>build=</tt>". For example: |
| 241 | <pre> build=/path/to/binary</pre> |
| 242 | Leading spaces on the line are ignored. |
| 243 | |
| 244 | <li>Mapping line from ProcMapsIterator::FormatLine. For example: |
| 245 | <pre> 40000000-40015000 r-xp 00000000 03:01 12845071 /lib/ld-2.3.2.so</pre> |
| 246 | The first address must start at the beginning of the line. |
| 247 | </ul> |
| 248 | |
| 249 | <p>Unrecognized lines should be ignored by analysis tools. |
| 250 | |
| 251 | <p>When processing the paths see in mapping lines, occurrences of |
| 252 | <tt>$build</tt> followed by a non-word character (i.e., characters |
| 253 | other than underscore or alphanumeric characters), should be replaced |
| 254 | by the path given on the last build specifier line. |
| 255 | |
| 256 | <hr> |
| 257 | <address>Chris Demetriou<br> |
| 258 | <!-- Created: Mon Aug 27 12:18:26 PDT 2007 --> |
| 259 | <!-- hhmts start --> |
| 260 | Last modified: Mon Aug 27 12:18:26 PDT 2007 (cgd) |
| 261 | <!-- hhmts end --> |
| 262 | </address> |
| 263 | </BODY> |
| 264 | </HTML> |