1 <?xml version="1.0"?> <!-- -*- sgml -*- -->
2 <!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
3 "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd"
4 [ <!ENTITY % vg-entities SYSTEM "../../docs/xml/vg-entities.xml"> %vg-entities; ]>
6 <chapter id="cl-format" xreflabel="Callgrind Format Specification">
7 <title>Callgrind Format Specification</title>
9 <para>This chapter describes the Callgrind Profile Format, Version 1.</para>
11 <para>A synonymous name is "Calltree Profile Format". These names actually mean
12 the same since Callgrind was previously named Calltree.</para>
14 <para>The format description is meant for the user to be able to understand the
15 file contents; but more important, it is given for authors of measurement or
16 visualization tools to be able to write and read this format.</para>
18 <sect1 id="cl-format.overview" xreflabel="Overview">
19 <title>Overview</title>
21 <para>The profile data format is ASCII based.
22 It is written by Callgrind, and it is upwards compatible
23 to the format used by Cachegrind (ie. Cachegrind uses a subset). It can
24 be read by callgrind_annotate and KCachegrind.</para>
26 <para>This chapter gives on overview of format features and examples.
27 For detailed syntax, look at the format reference.</para>
29 <sect2 id="cl-format.overview.basics" xreflabel="Basic Structure">
30 <title>Basic Structure</title>
32 <para>Each file has a header part of an arbitrary number of lines of the
33 format "key: value". The lines with key "positions" and "events" define
34 the meaning of cost lines in the second part of the file: the value of
35 "positions" is a list of subpositions, and the value of "events" is a list
36 of event type names. Cost lines consist of subpositions followed by 64-bit
37 counters for the events, in the order specified by the "positions" and "events"
40 <para>The "events" header line is always required in contrast to the optional
41 line for "positions", which defaults to "line", i.e. a line number of some
42 source file. In addition, the second part of the file contains position
43 specifications of the form "spec=name". "spec" can be e.g. "fn" for a
44 function name or "fl" for a file name. Cost lines are always related to
45 the function/file specifications given directly before.</para>
49 <sect2 id="cl-format.overview.example1" xreflabel="Simple Example">
50 <title>Simple Example</title>
52 <para>The event names in the following example are quite arbitrary, and are not
53 related to event names used by Callgrind. Especially, cycle counts matching
54 real processors probably will never be generated by any Valgrind tools, as these
55 are bound to simulations of simple machine models for acceptable slowdown.
56 However, any profiling tool could use the format described in this chapter.</para>
59 <screen>events: Cycles Instructions Flops
63 16 20 12</screen></para>
65 <para>The above example gives profile information for event types "Cycles",
66 "Instructions", and "Flops". Thus, cost lines give the number of CPU cycles
67 passed by, number of executed instructions, and number of floating point
68 operations executed while running code corresponding to some source
69 position. As there is no line specifying the value of "positions", it defaults
70 to "line", which means that the first number of a cost line is always a line
73 <para>Thus, the first cost line specifies that in line 15 of source file
74 <filename>file.f</filename> there is code belonging to function
75 <function>main</function>. While running, 90 CPU cycles passed by, and 2 of
76 the 14 instructions executed were floating point operations. Similarly, the
77 next line specifies that there were 12 instructions executed in the context
78 of function <function>main</function> which can be related to line 16 in
79 file <filename>file.f</filename>, taking 20 CPU cycles. If a cost line
80 specifies less event counts than given in the "events" line, the rest is
81 assumed to be zero. I.e. there was no floating point instruction executed
82 relating to line 16.</para>
84 <para>Note that regular cost lines always give self (also called exclusive)
85 cost of code at a given position. If you specify multiple cost lines for the
86 same position, these will be summed up. On the other hand, in the example above
87 there is no specification of how many times function
88 <function>main</function> actually was
89 called: profile data only contains sums.</para>
94 <sect2 id="cl-format.overview.associations" xreflabel="Associations">
95 <title>Associations</title>
97 <para>The most important extension to the original format of Cachegrind is the
98 ability to specify call relationship among functions. More generally, you
99 specify associations among positions. For this, the second part of the
100 file also can contain association specifications. These look similar to
101 position specifications, but consist of 2 lines. For calls, the format
104 calls=(Call Count) (Destination position)
105 (Source position) (Inclusive cost of call)
108 <para>The destination only specifies subpositions like line number. Therefore,
109 to be able to specify a call to another function in another source file, you
110 have to precede the above lines with a "cfn=" specification for the name of the
111 called function, and a "cfl=" specification if the function is in another
112 source file. The 2nd line looks like a regular cost line with the difference
113 that inclusive cost spent inside of the function call has to be specified.</para>
115 <para>Other associations which or for example (conditional) jumps. See the
116 reference below for details.</para>
121 <sect2 id="cl-format.overview.example2" xreflabel="Extended Example">
122 <title>Extended Example</title>
124 <para>The following example shows 3 functions, <function>main</function>,
125 <function>func1</function>, and <function>func2</function>. Function
126 <function>main</function> calls <function>func1</function> once and
127 <function>func2</function> 3 times. <function>func1</function> calls
128 <function>func2</function> 2 times.
129 <screen>events: Instructions
151 20 700</screen></para>
153 <para>One can see that in <function>main</function> only code from line 16
154 is executed where also the other functions are called. Inclusive cost of
155 <function>main</function> is 820, which is the sum of self cost 20 and costs
156 spent in the calls: 400 for the single call to <function>func1</function>
157 and 400 as sum for the three calls to <function>func2</function>.</para>
159 <para>Function <function>func1</function> is located in
160 <filename>file1.c</filename>, the same as <function>main</function>.
161 Therefore, a "cfl=" specification for the call to <function>func1</function>
162 is not needed. The function <function>func1</function> only consists of code
163 at line 51 of <filename>file1.c</filename>, where <function>func2</function>
169 <sect2 id="cl-format.overview.compression1" xreflabel="Name Compression">
170 <title>Name Compression</title>
172 <para>With the introduction of association specifications like calls it is
173 needed to specify the same function or same file name multiple times. As
174 absolute filenames or symbol names in C++ can be quite long, it is advantageous
175 to be able to specify integer IDs for position specifications.
176 Here, the term "position" corresponds to a file name (source or object file)
177 or function name.</para>
179 <para>To support name compression, a position specification can be not only of
180 the format "spec=name", but also "spec=(ID) name" to specify a mapping of an
181 integer ID to a name, and "spec=(ID)" to reference a previously defined ID
182 mapping. There is a separate ID mapping for each position specification,
183 i.e. you can use ID 1 for both a file name and a symbol name.</para>
185 <para>With string compression, the example from 1.4 looks like this:
186 <screen>events: Instructions
208 20 700</screen></para>
210 <para>As position specifications carry no information themselves, but only change
211 the meaning of subsequent cost lines or associations, they can appear
212 everywhere in the file without any negative consequence. Especially, you can
213 define name compression mappings directly after the header, and before any cost
214 lines. Thus, the above example can also be written as
215 <screen>events: Instructions
217 # define file ID mapping
220 # define function ID mapping
233 <sect2 id="cl-format.overview.compression2" xreflabel="Subposition Compression">
234 <title>Subposition Compression</title>
236 <para>If a Callgrind data file should hold costs for each assembler instruction
237 of a program, you specify subposition "instr" in the "positions:" header line,
238 and each cost line has to include the address of some instruction. Addresses
239 are allowed to have a size of 64 bits to support 64-bit architectures. Thus,
240 repeating similar, long addresses for almost every line in the data file can
241 enlarge the file size quite significantly, and
242 motivates for subposition compression: instead of every cost line starting with
243 a 16 character long address, one is allowed to specify relative addresses.
244 This relative specification is not only allowed for instruction addresses, but
245 also for line numbers; both addresses and line numbers are called "subpositions".</para>
247 <para>A relative subposition always is based on the corresponding subposition
248 of the last cost line, and starts with a "+" to specify a positive difference,
249 a "-" to specify a negative difference, or consists of "*" to specify the same
250 subposition. Because absolute subpositions always are positive (ie. never
251 prefixed by "-"), any relative specification is non-ambiguous; additionally,
252 absolute and relative subposition specifications can be mixed freely.
253 Assume the following example (subpositions can always be specified
254 as hexadecimal numbers, beginning with "0x"):
255 <screen>positions: instr line
261 0x80001238 91 6</screen></para>
263 <para>With subposition compression, this looks like
264 <screen>positions: instr line
270 +1 +1 6</screen></para>
272 <para>Remark: For assembler annotation to work, instruction addresses have to
273 be corrected to correspond to addresses found in the original binary. I.e. for
274 relocatable shared objects, often a load offset has to be subtracted.</para>
279 <sect2 id="cl-format.overview.misc" xreflabel="Miscellaneous">
280 <title>Miscellaneous</title>
282 <sect3 id="cl-format.overview.misc.summary" xreflabel="Cost Summary Information">
283 <title>Cost Summary Information</title>
285 <para>For the visualization to be able to show cost percentage, a sum of the
286 cost of the full run has to be known. Usually, it is assumed that this is the
287 sum of all cost lines in a file. But sometimes, this is not correct. Thus, you
288 can specify a "summary:" line in the header giving the full cost for the
289 profile run. This has another effect: a import filter can show a progress bar
290 while loading a large data file if he knows to cost sum in advance.</para>
294 <sect3 id="cl-format.overview.misc.events" xreflabel="Long Names for Event Types and inherited Types">
295 <title>Long Names for Event Types and inherited Types</title>
297 <para>Event types for cost lines are specified in the "events:" line with an
298 abbreviated name. For visualization, it makes sense to be able to specify some
299 longer, more descriptive name. For an event type "Ir" which means "Instruction
300 Fetches", this can be specified the header line
301 <screen>event: Ir : Instruction Fetches
302 events: Ir Dr</screen></para>
304 <para>In this example, "Dr" itself has no long name associated. The order of
305 "event:" lines and the "events:" line is of no importance. Additionally,
306 inherited event types can be introduced for which no raw data is available, but
307 which are calculated from given types. Suppose the last example, you could add
308 <screen>event: Sum = Ir + Dr</screen>
309 to specify an additional event type "Sum", which is calculated by adding costs
310 for "Ir and "Dr".</para>
318 <sect1 id="cl-format.reference" xreflabel="Reference">
319 <title>Reference</title>
321 <sect2 id="cl-format.reference.grammar" xreflabel="Grammar">
322 <title>Grammar</title>
325 <screen>ProfileDataFile := FormatVersion? Creator? PartData*</screen>
326 <screen>FormatVersion := "version:" Space* Number "\n"</screen>
327 <screen>Creator := "creator:" NoNewLineChar* "\n"</screen>
328 <screen>PartData := (HeaderLine "\n")+ (BodyLine "\n")+</screen>
329 <screen>HeaderLine := (empty line)
330 | ('#' NoNewLineChar*)
334 | CostLineDef</screen>
335 <screen>PartDetail := TargetCommand | TargetID</screen>
336 <screen>TargetCommand := "cmd:" Space* NoNewLineChar*</screen>
337 <screen>TargetID := ("pid"|"thread"|"part") ":" Space* Number</screen>
338 <screen>Description := "desc:" Space* Name Space* ":" NoNewLineChar*</screen>
339 <screen>EventSpecification := "event:" Space* Name InheritedDef? LongNameDef?</screen>
340 <screen>InheritedDef := "=" InheritedExpr</screen>
341 <screen>InheritedExpr := Name
342 | Number Space* ("*" Space*)? Name
343 | InheritedExpr Space* "+" Space* InheritedExpr</screen>
344 <screen>LongNameDef := ":" NoNewLineChar*</screen>
345 <screen>CostLineDef := "events:" Space* Name (Space+ Name)*
346 | "positions:" "instr"? (Space+ "line")?</screen>
347 <screen>BodyLine := (empty line)
348 | ('#' NoNewLineChar*)
350 | PositionSpecification
351 | AssociationSpecification</screen>
352 <screen>CostLine := SubPositionList Costs?</screen>
353 <screen>SubPositionList := (SubPosition+ Space+)+</screen>
354 <screen>SubPosition := Number | "+" Number | "-" Number | "*"</screen>
355 <screen>Costs := (Number Space+)+</screen>
356 <screen>PositionSpecification := Position "=" Space* PositionName</screen>
357 <screen>Position := CostPosition | CalledPosition</screen>
358 <screen>CostPosition := "ob" | "fl" | "fi" | "fe" | "fn"</screen>
359 <screen>CalledPosition := " "cob" | "cfl" | "cfn"</screen>
360 <screen>PositionName := ( "(" Number ")" )? (Space* NoNewLineChar* )?</screen>
361 <screen>AssociationSpecification := CallSpecification
362 | JumpSpecification</screen>
363 <screen>CallSpecification := CallLine "\n" CostLine</screen>
364 <screen>CallLine := "calls=" Space* Number Space+ SubPositionList</screen>
365 <screen>JumpSpecification := ...</screen>
366 <screen>Space := " " | "\t"</screen>
367 <screen>Number := HexNumber | (Digit)+</screen>
368 <screen>Digit := "0" | ... | "9"</screen>
369 <screen>HexNumber := "0x" (Digit | HexChar)+</screen>
370 <screen>HexChar := "a" | ... | "f" | "A" | ... | "F"</screen>
371 <screen>Name = Alpha (Digit | Alpha)*</screen>
372 <screen>Alpha = "a" | ... | "z" | "A" | ... | "Z"</screen>
373 <screen>NoNewLineChar := all characters without "\n"</screen>
378 <sect2 id="cl-format.reference.header" xreflabel="Description of Header Lines">
379 <title>Description of Header Lines</title>
381 <para>The header has an arbitrary number of lines of the format
382 "key: value". Possible <emphasis>key</emphasis> values for the header are:</para>
387 <para><computeroutput>version: number</computeroutput> [Callgrind]</para>
388 <para>This is used to distinguish future profile data formats. A
389 major version of 0 or 1 is supposed to be upwards compatible with
390 Cachegrind's format. It is optional; if not appearing, version 1
391 is supposed. Otherwise, this has to be the first header line.</para>
395 <para><computeroutput>pid: process id</computeroutput> [Callgrind]</para>
396 <para>This specifies the process ID of the supervised application
397 for which this profile was generated.</para>
401 <para><computeroutput>cmd: program name + args</computeroutput> [Cachegrind]</para>
402 <para>This specifies the full command line of the supervised
403 application for which this profile was generated.</para>
407 <para><computeroutput>part: number</computeroutput> [Callgrind]</para>
408 <para>This specifies a sequentially incremented number for each dump
409 generated, starting at 1.</para>
413 <para><computeroutput>desc: type: value</computeroutput> [Cachegrind]</para>
414 <para>This specifies various information for this dump. For some
415 types, the semantic is defined, but any description type is allowed.
416 Unknown types should be ignored.</para>
417 <para>There are the types "I1 cache", "D1 cache", "L2 cache", which
418 specify parameters used for the cache simulator. These are the only
419 types originally used by Cachegrind. Additionally, Callgrind uses
420 the following types: "Timerange" gives a rough range of the basic
421 block counter, for which the cost of this dump was collected.
422 Type "Trigger" states the reason of why this trace was generated.
423 E.g. program termination or forced interactive dump.</para>
427 <para><computeroutput>positions: [instr] [line]</computeroutput> [Callgrind]</para>
428 <para>For cost lines, this defines the semantic of the first numbers.
429 Any combination of "instr", "bb" and "line" is allowed, but has to be
430 in this order which corresponds to position numbers at the start of
431 the cost lines later in the file.</para>
432 <para>If "instr" is specified, the position is the address of an
433 instruction whose execution raised the events given later on the
434 line. This address is relative to the offset of the binary/shared
435 library file to not have to specify relocation info. For "line",
436 the position is the line number of a source file, which is
437 responsible for the events raised. Note that the mapping of "instr"
438 and "line" positions are given by the debugging line information
439 produced by the compiler.</para>
440 <para>This field is optional. If not specified, "line" is supposed
445 <para><computeroutput>events: event type abbreviations</computeroutput> [Cachegrind]</para>
446 <para>A list of short names of the event types logged in this file.
447 The order is the same as in cost lines. The first event type is the
448 second or third number in a cost line, depending on the value of
449 "positions". Callgrind does not add additional cost types. Specify
451 <para>Cost types from original Cachegrind are:
454 <para><command>Ir</command>: Instruction read access</para>
457 <para><command>I1mr</command>: Instruction Level 1 read cache miss</para>
460 <para><command>I2mr</command>: Instruction Level 2 read cache miss</para>
470 <para><computeroutput>summary: costs</computeroutput> [Callgrind]</para>
471 <para><computeroutput>totals: costs</computeroutput> [Cachegrind]</para>
472 <para>The value or the total number of events covered by this trace
473 file. Both keys have the same meaning, but the "totals:" line
474 happens to be at the end of the file, while "summary:" appears in
475 the header. This was added to allow postprocessing tools to know
476 in advance to total cost. The two lines always give the same cost
484 <sect2 id="cl-format.reference.body" xreflabel="Description of Body Lines">
485 <title>Description of Body Lines</title>
487 <para>There exist lines
488 <computeroutput>spec=position</computeroutput>. The values for position
489 specifications are arbitrary strings. When starting with "(" and a
490 digit, it's a string in compressed format. Otherwise it's the real
491 position string. This allows for file and symbol names as position
492 strings, as these never start with "(" + <emphasis>digit</emphasis>.
493 The compressed format is either "(" <emphasis>number</emphasis> ")"
494 <emphasis>space</emphasis> <emphasis>position</emphasis> or only
495 "(" <emphasis>number</emphasis> ")". The first relates
496 <emphasis>position</emphasis> to <emphasis>number</emphasis> in the
497 context of the given format specification from this line to the end of
498 the file; it makes the (<emphasis>number</emphasis>) an alias for
499 <emphasis>position</emphasis>. Compressed format is always
502 <para>Position specifications allowed:</para>
506 <para><computeroutput>ob=</computeroutput> [Callgrind]</para>
507 <para>The ELF object where the cost of next cost lines happens.</para>
511 <para><computeroutput>fl=</computeroutput> [Cachegrind]</para>
515 <para><computeroutput>fi=</computeroutput> [Cachegrind]</para>
519 <para><computeroutput>fe=</computeroutput> [Cachegrind]</para>
520 <para>The source file including the code which is responsible for
521 the cost of next cost lines. "fi="/"fe=" is used when the source
522 file changes inside of a function, i.e. for inlined code.</para>
526 <para><computeroutput>fn=</computeroutput> [Cachegrind]</para>
527 <para>The name of the function where the cost of next cost lines
532 <para><computeroutput>cob=</computeroutput> [Callgrind]</para>
533 <para>The ELF object of the target of the next call cost lines.</para>
537 <para><computeroutput>cfl=</computeroutput> [Callgrind]</para>
538 <para>The source file including the code of the target of the
539 next call cost lines.</para>
543 <para><computeroutput>cfn=</computeroutput> [Callgrind]</para>
544 <para>The name of the target function of the next call cost
549 <para><computeroutput>calls=</computeroutput> [Callgrind]</para>
550 <para>The number of nonrecursive calls which are responsible for the
551 cost specified by the next call cost line. This is the cost spent
552 inside of the called function.</para>
553 <para>After "calls=" there MUST be a cost line. This is the cost
554 spent in the called function. The first number is the source line
555 from where the call happened.</para>
559 <para><computeroutput>jump=count target position</computeroutput> [Callgrind]</para>
560 <para>Unconditional jump, executed count times, to the given target
565 <para><computeroutput>jcnd=exe.count jumpcount target position</computeroutput> [Callgrind]</para>
566 <para>Conditional jump, executed exe.count times with jumpcount
567 jumps to the given target position.</para>