1 <chapter id="profiling">
2 <title>Profiling</Title>
3 <indexterm><primary>profiling</primary>
5 <indexterm><primary>cost-centre profiling</primary></indexterm>
7 <para> Glasgow Haskell comes with a time and space profiling
8 system. Its purpose is to help you improve your understanding of
9 your program's execution behaviour, so you can improve it.</para>
11 <para> Any comments, suggestions and/or improvements you have are
12 welcome. Recommended “profiling tricks” would be
13 especially cool! </para>
15 <para>Profiling a program is a three-step process:</para>
19 <para> Re-compile your program for profiling with the
20 <literal>-prof</literal> option, and probably one of the
21 <literal>-auto</literal> or <literal>-auto-all</literal>
22 options. These options are described in more detail in <xref
23 linkend="prof-compiler-options"> </para>
24 <indexterm><primary><literal>-prof</literal></primary>
26 <indexterm><primary><literal>-auto</literal></primary>
28 <indexterm><primary><literal>-auto-all</literal></primary>
33 <para> Run your program with one of the profiling options, eg.
34 <literal>+RTS -p -RTS</literal>. This generates a file of
35 profiling information.</para>
36 <indexterm><primary><option>-p</option></primary><secondary>RTS
37 option</secondary></indexterm>
41 <para> Examine the generated profiling information, using one of
42 GHC's profiling tools. The tool to use will depend on the kind
43 of profiling information generated.</para>
49 <title>Cost centres and cost-centre stacks</title>
51 <para>GHC's profiling system assigns <firstterm>costs</firstterm>
52 to <firstterm>cost centres</firstterm>. A cost is simply the time
53 or space required to evaluate an expression. Cost centres are
54 program annotations around expressions; all costs incurred by the
55 annotated expression are assigned to the enclosing cost centre.
56 Furthermore, GHC will remember the stack of enclosing cost centres
57 for any given expression at run-time and generate a call-graph of
58 cost attributions.</para>
60 <para>Let's take a look at an example:</para>
63 main = print (nfib 25)
64 nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)
67 <para>Compile and run this program as follows:</para>
70 $ ghc -prof -auto-all -o Main Main.hs
76 <para>When a GHC-compiled program is run with the
77 <option>-p</option> RTS option, it generates a file called
78 <filename><prog>.prof</filename>. In this case, the file
79 will contain something like this:</para>
82 Fri May 12 14:06 2000 Time and Allocation Profiling Report (Final)
86 total time = 0.14 secs (7 ticks @ 20 ms)
87 total alloc = 8,741,204 bytes (excludes profiling overheads)
89 COST CENTRE MODULE %time %alloc
95 COST CENTRE MODULE entries %time %alloc %time %alloc
97 MAIN MAIN 0 0.0 0.0 100.0 100.0
98 main Main 0 0.0 0.0 0.0 0.0
99 CAF PrelHandle 3 0.0 0.0 0.0 0.0
100 CAF PrelAddr 1 0.0 0.0 0.0 0.0
101 CAF Main 6 0.0 0.0 100.0 100.0
102 main Main 1 0.0 0.0 100.0 100.0
103 nfib Main 242785 100.0 100.0 100.0 100.0
107 <para>The first part of the file gives the program name and
108 options, and the total time and total memory allocation measured
109 during the run of the program (note that the total memory
110 allocation figure isn't the same as the amount of
111 <emphasis>live</emphasis> memory needed by the program at any one
112 time; the latter can be determined using heap profiling, which we
113 will describe shortly).</para>
115 <para>The second part of the file is a break-down by cost centre
116 of the most costly functions in the program. In this case, there
117 was only one significant function in the program, namely
118 <function>nfib</function>, and it was responsible for 100%
119 of both the time and allocation costs of the program.</para>
121 <para>The third and final section of the file gives a profile
122 break-down by cost-centre stack. This is roughly a call-graph
123 profile of the program. In the example above, it is clear that
124 the costly call to <function>nfib</function> came from
125 <function>main</function>.</para>
127 <para>The time and allocation incurred by a given part of the
128 program is displayed in two ways: “individual”, which
129 are the costs incurred by the code covered by this cost centre
130 stack alone, and “inherited”, which includes the costs
131 incurred by all the children of this node.</para>
133 <para>The usefulness of cost-centre stacks is better demonstrated
134 by modifying the example slightly:</para>
137 main = print (f 25 + g 25)
139 g n = nfib (n `div` 2)
140 nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)
143 <para>Compile and run this program as before, and take a look at
144 the new profiling results:</para>
147 COST CENTRE MODULE scc %time %alloc %time %alloc
149 MAIN MAIN 0 0.0 0.0 100.0 100.0
150 main Main 0 0.0 0.0 0.0 0.0
151 CAF PrelHandle 3 0.0 0.0 0.0 0.0
152 CAF PrelAddr 1 0.0 0.0 0.0 0.0
153 CAF Main 9 0.0 0.0 100.0 100.0
154 main Main 1 0.0 0.0 100.0 100.0
155 g Main 1 0.0 0.0 0.0 0.2
156 nfib Main 465 0.0 0.2 0.0 0.2
157 f Main 1 0.0 0.0 100.0 99.8
158 nfib Main 242785 100.0 99.8 100.0 99.8
161 <para>Now although we had two calls to <function>nfib</function>
162 in the program, it is immediately clear that it was the call from
163 <function>f</function> which took all the time.</para>
165 <para>The actual meaning of the various columns in the output is:</para>
171 <para>The number of times this particular point in the call
172 graph was entered.</para>
177 <term>individual %time</term>
179 <para>The percentage of the total run time of the program
180 spent at this point in the call graph.</para>
185 <term>individual %alloc</term>
187 <para>The percentage of the total memory allocations
188 (excluding profiling overheads) of the program made by this
194 <term>inherited %time</term>
196 <para>The percentage of the total run time of the program
197 spent below this point in the call graph.</para>
202 <term>inherited %alloc</term>
204 <para>The percentage of the total memory allocations
205 (excluding profiling overheads) of the program made by this
206 call and all of its sub-calls.</para>
211 <para>In addition you can use the <Option>-P</Option> RTS option
212 <indexterm><primary><option>-P</option></primary></indexterm> to
213 get the following additional information:</para>
217 <term><literal>ticks</literal></term>
219 <para>The raw number of time “ticks” which were
220 attributed to this cost-centre; from this, we get the
221 <literal>%time</literal> figure mentioned
227 <term><literal>bytes</literal></term>
229 <para>Number of bytes allocated in the heap while in this
230 cost-centre; again, this is the raw number from which we get
231 the <literal>%alloc</literal> figure mentioned
237 <para>What about recursive functions, and mutually recursive
238 groups of functions? Where are the costs attributed? Well,
239 although GHC does keep information about which groups of functions
240 called each other recursively, this information isn't displayed in
241 the basic time and allocation profile, instead the call-graph is
242 flattened into a tree. The XML profiling tool (described in <xref
243 linkend="prof-xml-tool">) will be able to display real loops in
244 the call-graph.</para>
246 <sect2><title>Inserting cost centres by hand</title>
248 <para>Cost centres are just program annotations. When you say
249 <option>-auto-all</option> to the compiler, it automatically
250 inserts a cost centre annotation around every top-level function
251 in your program, but you are entirely free to add the cost
252 centre annotations yourself.</para>
254 <para>The syntax of a cost centre annotation is</para>
257 {-# SCC "name" #-} <expression>
260 <para>where <literal>"name"</literal> is an aribrary string,
261 that will become the name of your cost centre as it appears
262 in the profiling output, and
263 <literal><expression></literal> is any Haskell
264 expression. An <literal>SCC</literal> annotation extends as
265 far to the right as possible when parsing.</para>
269 <sect2 id="prof-rules">
270 <title>Rules for attributing costs</title>
272 <para>The cost of evaluating any expression in your program is
273 attributed to a cost-centre stack using the following rules:</para>
277 <para>If the expression is part of the
278 <firstterm>one-off</firstterm> costs of evaluating the
279 enclosing top-level definition, then costs are attributed to
280 the stack of lexically enclosing <literal>SCC</literal>
281 annotations on top of the special <literal>CAF</literal>
286 <para>Otherwise, costs are attributed to the stack of
287 lexically-enclosing <literal>SCC</literal> annotations,
288 appended to the cost-centre stack in effect at the
289 <firstterm>call site</firstterm> of the current top-level
290 definition<footnote> <para>The call-site is just the place
291 in the source code which mentions the particular function or
292 variable.</para></footnote>. Notice that this is a recursive
297 <para>What do we mean by one-off costs? Well, Haskell is a lazy
298 language, and certain expressions are only ever evaluated once.
299 For example, if we write:</para>
305 <para>then <varname>x</varname> will only be evaluated once (if
306 at all), and subsequent demands for <varname>x</varname> will
307 immediately get to see the cached result. The definition
308 <varname>x</varname> is called a CAF (Constant Applicative
309 Form), because it has no arguments.</para>
311 <para>For the purposes of profiling, we say that the expression
312 <literal>nfib 25</literal> belongs to the one-off costs of
313 evaluating <varname>x</varname>.</para>
315 <para>Since one-off costs aren't strictly speaking part of the
316 call-graph of the program, they are attributed to a special
317 top-level cost centre, <literal>CAF</literal>. There may be one
318 <literal>CAF</literal> cost centre for each module (the
319 default), or one for each top-level definition with any one-off
320 costs (this behaviour can be selected by giving GHC the
321 <option>-caf-all</option> flag).</para>
323 <indexterm><primary><literal>-caf-all</literal></primary>
326 <para>If you think you have a weird profile, or the call-graph
327 doesn't look like you expect it to, feel free to send it (and
328 your program) to us at
329 <email>glasgow-haskell-bugs@haskell.org</email>.</para>
334 <sect1 id="prof-compiler-options">
335 <title>Compiler options for profiling</title>
337 <indexterm><primary>profiling</primary><secondary>options</secondary></indexterm>
338 <indexterm><primary>options</primary><secondary>for profiling</secondary></indexterm>
342 <term><Option>-prof</Option>:</Term>
343 <indexterm><primary><option>-prof</option></primary></indexterm>
345 <para> To make use of the profiling system
346 <emphasis>all</emphasis> modules must be compiled and linked
347 with the <option>-prof</option> option. Any
348 <literal>SCC</literal> annotations you've put in your source
349 will spring to life.</para>
351 <para> Without a <option>-prof</option> option, your
352 <literal>SCC</literal>s are ignored; so you can compile
353 <literal>SCC</literal>-laden code without changing
359 <para>There are a few other profiling-related compilation options.
360 Use them <emphasis>in addition to</emphasis>
361 <option>-prof</option>. These do not have to be used consistently
362 for all modules in a program.</para>
366 <term><option>-auto</option>:</Term>
367 <indexterm><primary><option>-auto</option></primary></indexterm>
368 <indexterm><primary>cost centres</primary><secondary>automatically inserting</secondary></indexterm>
370 <para> GHC will automatically add
371 <Function>_scc_</Function> constructs for all
372 top-level, exported functions.</para>
377 <term><option>-auto-all</option>:</Term>
378 <indexterm><primary><option>-auto-all</option></primary></indexterm>
380 <para> <Emphasis>All</Emphasis> top-level functions,
381 exported or not, will be automatically
382 <Function>_scc_</Function>'d.</para>
387 <term><option>-caf-all</option>:</Term>
388 <indexterm><primary><option>-caf-all</option></primary></indexterm>
390 <para> The costs of all CAFs in a module are usually
391 attributed to one “big” CAF cost-centre. With
392 this option, all CAFs get their own cost-centre. An
393 “if all else fails” option…</para>
398 <term><option>-ignore-scc</option>:</Term>
399 <indexterm><primary><option>-ignore-scc</option></primary></indexterm>
401 <para>Ignore any <Function>_scc_</Function>
402 constructs, so a module which already has
403 <Function>_scc_</Function>s can be compiled
404 for profiling with the annotations ignored.</para>
412 <sect1 id="prof-time-options">
413 <title>Time and allocation profiling</Title>
415 <para>To generate a time and allocation profile, give one of the
416 following RTS options to the compiled program when you run it (RTS
417 options should be enclosed between <literal>+RTS...-RTS</literal>
422 <term><Option>-p</Option> or <Option>-P</Option>:</Term>
423 <indexterm><primary><option>-p</option></primary></indexterm>
424 <indexterm><primary><option>-P</option></primary></indexterm>
425 <indexterm><primary>time profile</primary></indexterm>
427 <para>The <Option>-p</Option> option produces a standard
428 <Emphasis>time profile</Emphasis> report. It is written
430 <Filename><replaceable>program</replaceable>.prof</Filename>.</para>
432 <para>The <Option>-P</Option> option produces a more
433 detailed report containing the actual time and allocation
434 data as well. (Not used much.)</para>
439 <term><option>-px</option>:</term>
440 <indexterm><primary><option>-px</option></primary></indexterm>
442 <para>The <option>-px</option> option generates profiling
443 information in the XML format understood by our new
444 profiling tool, see <xref linkend="prof-xml-tool">.</para>
449 <term><option>-xc</option></term>
450 <indexterm><primary><option>-xc</option></primary><secondary>RTS
451 option</secondary></indexterm>
453 <para>This option makes use of the extra information
454 maintained by the cost-centre-stack profiler to provide
455 useful information about the location of runtime errors.
456 See <xref linkend="rts-options-debugging">.</para>
464 <sect1 id="prof-heap">
465 <title>Profiling memory usage</title>
467 <para>In addition to profiling the time and allocation behaviour
468 of your program, you can also generate a graph of its memory usage
469 over time. This is useful for detecting the causes of
470 <firstterm>space leaks</firstterm>, when your program holds on to
471 more memory at run-time that it needs to. Space leaks lead to
472 longer run-times due to heavy garbage collector ativity, and may
473 even cause the program to run out of memory altogether.</para>
475 <para>To generate a heap profile from your program:</para>
479 <para>Compile the program for profiling (<xref
480 linkend="prof-compiler-options">).</para>
483 <para>Run it with one of the heap profiling options described
484 below (eg. <option>-hc</option> for a basic producer profile).
485 This generates the file
486 <filename><replaceable>prog</replaceable>.hp</filename>.</para>
489 <para>Run <command>hp2ps</command> to produce a Postscript
491 <filename><replaceable>prog</replaceable>.ps</filename>. The
492 <command>hp2ps</command> utility is described in detail in
493 <xref linkend="hp2ps">.</para>
496 <para>Display the heap profile using a postscript viewer such
497 as <application>Ghostview</application>, or print it out on a
498 Postscript-capable printer.</para>
502 <sect2 id="rts-options-heap-prof">
503 <title>RTS options for heap profiling</title>
505 <para>There are several different kinds of heap profile that can
506 be generated. All the different profile types yield a graph of
507 live heap against time, but they differ in how the live heap is
508 broken down into bands. The following RTS options select which
509 break-down to use:</para>
513 <term><option>-hc</option></term>
514 <indexterm><primary><option>-hc</option></primary><secondary>RTS
515 option</secondary></indexterm>
517 <para>Breaks down the graph by the cost-centre stack which
518 produced the data.</para>
523 <term><option>-hm</option></term>
524 <indexterm><primary><option>-hm</option></primary><secondary>RTS
525 option</secondary></indexterm>
527 <para>Break down the live heap by the module containing
528 the code which produced the data.</para>
533 <term><option>-hd</option></term>
534 <indexterm><primary><option>-hd</option></primary><secondary>RTS
535 option</secondary></indexterm>
537 <para>Breaks down the graph by <firstterm>closure
538 description</firstterm>. For actual data, the description
539 is just the constructor name, for other closures it is a
540 compiler-generated string identifying the closure.</para>
545 <term><option>-hy</option></term>
546 <indexterm><primary><option>-hy</option></primary><secondary>RTS
547 option</secondary></indexterm>
549 <para>Breaks down the graph by
550 <firstterm>type</firstterm>. For closures which have
551 function type or unknown/polymorphic type, the string will
552 represent an approximation to the actual type.</para>
557 <term><option>-hr</option></term>
558 <indexterm><primary><option>-hr</option></primary><secondary>RTS
559 option</secondary></indexterm>
561 <para>Break down the graph by <firstterm>retainer
562 set</firstterm>. Retainer profiling is described in more
563 detail below (<xref linkend="retainer-prof">).</para>
568 <term><option>-hb</option></term>
569 <indexterm><primary><option>-hb</option></primary><secondary>RTS
570 option</secondary></indexterm>
572 <para>Break down the graph by
573 <firstterm>biography</firstterm>. Biographical profiling
574 is described in more detail below (<xref
575 linkend="biography-prof">).</para>
580 <para>In addition, the profile can be restricted to heap data
581 which satisfies certain criteria - for example, you might want
582 to display a profile by type but only for data produced by a
583 certain module, or a profile by retainer for a certain type of
584 data. Restrictions are specified as follows:</para>
588 <term><option>-hc</option><replaceable>name</replaceable>,...</term>
589 <indexterm><primary><option>-hc</option></primary><secondary>RTS
590 option</secondary></indexterm>
592 <para>Restrict the profile to closures produced by
593 cost-centre stacks with one of the specified cost centres
599 <term><option>-hC</option><replaceable>name</replaceable>,...</term>
600 <indexterm><primary><option>-hC</option></primary><secondary>RTS
601 option</secondary></indexterm>
603 <para>Restrict the profile to closures produced by
604 cost-centre stacks with one of the specified cost centres
605 anywhere in the stack.</para>
610 <term><option>-hm</option><replaceable>module</replaceable>,...</term>
611 <indexterm><primary><option>-hm</option></primary><secondary>RTS
612 option</secondary></indexterm>
614 <para>Restrict the profile to closures produced by the
615 specified modules.</para>
620 <term><option>-hd</option><replaceable>desc</replaceable>,...</term>
621 <indexterm><primary><option>-hd</option></primary><secondary>RTS
622 option</secondary></indexterm>
624 <para>Restrict the profile to closures with the specified
625 description strings.</para>
630 <term><option>-hy</option><replaceable>type</replaceable>,...</term>
631 <indexterm><primary><option>-hy</option></primary><secondary>RTS
632 option</secondary></indexterm>
634 <para>Restrict the profile to closures with the specified
640 <term><option>-hr</option><replaceable>cc</replaceable>,...</term>
641 <indexterm><primary><option>-hr</option></primary><secondary>RTS
642 option</secondary></indexterm>
644 <para>Restrict the profile to closures with retainer sets
645 containing cost-centre stacks with one of the specified
646 cost centres at the top.</para>
651 <term><option>-hb</option><replaceable>bio</replaceable>,...</term>
652 <indexterm><primary><option>-hb</option></primary><secondary>RTS
653 option</secondary></indexterm>
655 <para>Restrict the profile to closures with one of the
656 specified biographies, where
657 <replaceable>bio</replaceable> is one of
658 <literal>lag</literal>, <literal>drag</literal>,
659 <literal>void</literal>, or <literal>use</literal>.</para>
664 <para>For example, the following options will generate a
665 retainer profile restricted to <literal>Branch</literal> and
666 <literal>Leaf</literal> constructors:</para>
669 <replaceable>prog</replaceable> +RTS -hr -hdBranch,Leaf
672 <para>There can only be one "break-down" option
673 (eg. <option>-hr</option> in the example above), but there is no
674 limit on the number of further restrictions that may be applied.
675 All the options may be combined, with one exception: GHC doesn't
676 currently support mixing the <option>-hr</option> and
677 <option>-hb</option> options.</para>
679 <para>There's one more option which relates to heap
684 <term><Option>-i<replaceable>secs</replaceable></Option>:</Term>
685 <indexterm><primary><option>-i</option></primary></indexterm>
687 <para> Set the profiling (sampling) interval to
688 <replaceable>secs</replaceable> seconds (the default is
689 0.1 second). Fractions are allowed: for example
690 <Option>-i0.2</Option> will get 5 samples per second.
691 This only affects heap profiling; time profiles are always
692 sampled on a 1/50 second frequency.</para>
699 <sect2 id="retainer-prof">
700 <title>Retainer Profiling</title>
702 <para>Retainer profiling is designed to help answer questions
703 like <quote>why is this data being retained?</quote>. We start
704 by defining what we mean by a retainer:</para>
707 <para>A retainer is either the system stack, or an unevaluated
708 closure (thunk).</para>
711 <para>In particular, constructors are <emphasis>not</emphasis>
714 <para>An object A is retained by an object B if object A can be
715 reached by recursively following pointers starting from object
716 B but not meeting any other retainers on the way. Each object
717 has one or more retainers, collectively called its
718 <firstterm>retainer set</firstterm>.</para>
720 <para>When retainer profiling is requested by giving the program
721 the <option>-hr</option> option, a graph is generated which is
722 broken down by retainer set. A retainer set is displayed as a
723 set of cost-centre stacks; because this is usually too large to
724 fit on the profile graph, each retainer set is numbered and
725 shown abbreviated on the graph along with its number, and the
726 full list of retainer sets is dumped into the file
727 <filename><replaceable>prog</replaceable>.prof</filename>.</para>
729 <para>Retainer profiling requires multiple passes over the live
730 heap in order to discover the full retainer set for each
731 object, which can be quite slow. So we set a limit on the
732 maximum size of a retainer set, where all retainer sets larger
733 than the maximum retainer set size are replaced by the special
734 set <literal>MANY</literal>. The maximum set size defaults to 8
735 and can be altered with the <option>-R</option> RTS
740 <term><option>-R</option><replaceable>size</replaceable></term>
742 <para>Restrict the number of elements in a retainer set to
743 <replaceable>size</replaceable> (default 8).</para>
749 <title>Hints for using retainer profiling</title>
751 <para>The definition of retainers is designed to reflect a
752 common cause of space leaks: a large structure is retained by
753 an unevaluated computation, and will be released once the
754 compuation is forced. A good example is looking up a value in
755 a finite map, where unless the lookup is forced in a timely
756 manner the unevaluated lookup will cause the whole mapping to
757 be retained. These kind of space leaks can often be
758 eliminated by forcing the relevant computations to be
759 performed eagerly, using <literal>seq</literal> or strictness
760 annotations on data constructor fields.</para>
762 <para>Often a particular data structure is being retained by a
763 chain of unevaluated closures, only the nearest of which will
764 be reported by retainer profiling - for example A retains B, B
765 retains C, and C retains a large structure. There might be a
766 large number of Bs but only a single A, so A is really the one
767 we're interested in eliminating. However, retainer profiling
768 will in this case report B as the retainer of the large
769 structure. To move further up the chain of retainers, we can
770 ask for another retainer profile but this time restrict the
771 profile to B objects, so we get a profile of the retainers of
775 <replaceable>prog</replaceable> +RTS -hr -hcB
778 <para>This trick isn't foolproof, because there might be other
779 B closures in the heap which aren't the retainers we are
780 interested in, but we've found this to be a useful technique
781 in most cases.</para>
785 <sect2 id="biography-prof">
786 <title>Biographical Profiling</title>
788 <para>A typical heap object may be in one of the following four
789 states at each point in its lifetime:</para>
793 <para>The <firstterm>lag</firstterm> stage, which is the
794 time between creation and the first use of the
798 <para>the <firstterm>use</firstterm> stage, which lasts from
799 the first use until the last use of the object, and</para>
802 <para>The <firstterm>drag</firstterm> stage, which lasts
803 from the final use until the last reference to the object
807 <para>An object which is never used is said to be in the
808 <firstterm>void</firstterm> state for its whole
813 <para>A biographical heap profile displays the portion of the
814 live heap in each of the four states listed above. Usually the
815 most interesting states are the void and drag states: live heap
816 in these states is more likely to be wasted space than heap in
817 the lag or use states.</para>
819 <para>It is also possible to break down the heap in one or more
820 of these states by a different criteria, by restricting a
821 profile by biography. For example, to show the portion of the
822 heap in the drag or void state by producer: </para>
825 <replaceable>prog</replaceable> +RTS -hc -hbdrag,void
828 <para>Once you know the producer or the type of the heap in the
829 drag or void states, the next step is usually to find the
833 <replaceable>prog</replaceable> +RTS -hr -hc<replaceable>cc</replaceable>...
836 <para>NOTE: this two stage process is required because GHC
837 cannot currently profile using both biographical and retainer
838 information simultaneously.</para>
843 <sect1 id="prof-xml-tool">
844 <title>Graphical time/allocation profile</title>
846 <para>You can view the time and allocation profiling graph of your
847 program graphically, using <command>ghcprof</command>. This is a
848 new tool with GHC 4.08, and will eventually be the de-facto
849 standard way of viewing GHC profiles<footnote><para>Actually this
850 isn't true any more, we are working on a new tool for
851 displaying heap profiles using Gtk+HS, so
852 <command>ghcprof</command> may go away at some point in the future.</para>
855 <para>To run <command>ghcprof</command>, you need
856 <productname>daVinci</productname> installed, which can be
858 url="http://www.informatik.uni-bremen.de/daVinci/"><citetitle>The Graph
859 Visualisation Tool daVinci</citetitle></ulink>. Install one of
861 distributions<footnote><para><productname>daVinci</productname> is
862 sadly not open-source :-(.</para></footnote>, and set your
863 <envar>DAVINCIHOME</envar> environment variable to point to the
864 installation directory.</para>
866 <para><command>ghcprof</command> uses an XML-based profiling log
867 format, and you therefore need to run your program with a
868 different option: <option>-px</option>. The file generated is
869 still called <filename><prog>.prof</filename>. To see the
870 profile, run <command>ghcprof</command> like this:</para>
872 <indexterm><primary><option>-px</option></primary></indexterm>
875 $ ghcprof <prog>.prof
878 <para>which should pop up a window showing the call-graph of your
879 program in glorious detail. More information on using
880 <command>ghcprof</command> can be found at <ulink
881 url="http://www.dcs.warwick.ac.uk/people/academic/Stephen.Jarvis/profiler/index.html"><citetitle>The
882 Cost-Centre Stack Profiling Tool for
883 GHC</citetitle></ulink>.</para>
888 <title><command>hp2ps</command>--heap profile to PostScript</title>
890 <indexterm><primary><command>hp2ps</command></primary></indexterm>
891 <indexterm><primary>heap profiles</primary></indexterm>
892 <indexterm><primary>postscript, from heap profiles</primary></indexterm>
893 <indexterm><primary><option>-h<break-down></option></primary></indexterm>
898 hp2ps [flags] [<file>[.hp]]
902 <command>hp2ps</command><indexterm><primary>hp2ps
903 program</primary></indexterm> converts a heap profile as produced
904 by the <Option>-h<break-down></Option> runtime option into a
905 PostScript graph of the heap profile. By convention, the file to
906 be processed by <command>hp2ps</command> has a
907 <filename>.hp</filename> extension. The PostScript output is
908 written to <filename><file>@.ps</filename>. If
909 <filename><file></filename> is omitted entirely, then the
910 program behaves as a filter.</para>
912 <para><command>hp2ps</command> is distributed in
913 <filename>ghc/utils/hp2ps</filename> in a GHC source
914 distribution. It was originally developed by Dave Wakeling as part
915 of the HBC/LML heap profiler.</para>
917 <para>The flags are:</para>
922 <term><Option>-d</Option></Term>
924 <para>In order to make graphs more readable,
925 <command>hp2ps</command> sorts the shaded bands for each
926 identifier. The default sort ordering is for the bands with
927 the largest area to be stacked on top of the smaller ones.
928 The <Option>-d</Option> option causes rougher bands (those
929 representing series of values with the largest standard
930 deviations) to be stacked on top of smoother ones.</para>
935 <term><Option>-b</Option></Term>
937 <para>Normally, <command>hp2ps</command> puts the title of
938 the graph in a small box at the top of the page. However, if
939 the JOB string is too long to fit in a small box (more than
940 35 characters), then <command>hp2ps</command> will choose to
941 use a big box instead. The <Option>-b</Option> option
942 forces <command>hp2ps</command> to use a big box.</para>
947 <term><Option>-e<float>[in|mm|pt]</Option></Term>
949 <para>Generate encapsulated PostScript suitable for
950 inclusion in LaTeX documents. Usually, the PostScript graph
951 is drawn in landscape mode in an area 9 inches wide by 6
952 inches high, and <command>hp2ps</command> arranges for this
953 area to be approximately centred on a sheet of a4 paper.
954 This format is convenient of studying the graph in detail,
955 but it is unsuitable for inclusion in LaTeX documents. The
956 <Option>-e</Option> option causes the graph to be drawn in
957 portrait mode, with float specifying the width in inches,
958 millimetres or points (the default). The resulting
959 PostScript file conforms to the Encapsulated PostScript
960 (EPS) convention, and it can be included in a LaTeX document
961 using Rokicki's dvi-to-PostScript converter
962 <command>dvips</command>.</para>
967 <term><Option>-g</Option></Term>
969 <para>Create output suitable for the <command>gs</command>
970 PostScript previewer (or similar). In this case the graph is
971 printed in portrait mode without scaling. The output is
972 unsuitable for a laser printer.</para>
977 <term><Option>-l</Option></Term>
979 <para>Normally a profile is limited to 20 bands with
980 additional identifiers being grouped into an
981 <literal>OTHER</literal> band. The <Option>-l</Option> flag
982 removes this 20 band and limit, producing as many bands as
983 necessary. No key is produced as it won't fit!. It is useful
984 for creation time profiles with many bands.</para>
989 <term><Option>-m<int></Option></Term>
991 <para>Normally a profile is limited to 20 bands with
992 additional identifiers being grouped into an
993 <literal>OTHER</literal> band. The <Option>-m</Option> flag
994 specifies an alternative band limit (the maximum is
997 <para><Option>-m0</Option> requests the band limit to be
998 removed. As many bands as necessary are produced. However no
999 key is produced as it won't fit! It is useful for displaying
1000 creation time profiles with many bands.</para>
1005 <term><Option>-p</Option></Term>
1007 <para>Use previous parameters. By default, the PostScript
1008 graph is automatically scaled both horizontally and
1009 vertically so that it fills the page. However, when
1010 preparing a series of graphs for use in a presentation, it
1011 is often useful to draw a new graph using the same scale,
1012 shading and ordering as a previous one. The
1013 <Option>-p</Option> flag causes the graph to be drawn using
1014 the parameters determined by a previous run of
1015 <command>hp2ps</command> on <filename>file</filename>. These
1016 are extracted from <filename>file@.aux</filename>.</para>
1021 <term><Option>-s</Option></Term>
1023 <para>Use a small box for the title.</para>
1028 <term><Option>-t<float></Option></Term>
1030 <para>Normally trace elements which sum to a total of less
1031 than 1% of the profile are removed from the
1032 profile. The <option>-t</option> option allows this
1033 percentage to be modified (maximum 5%).</para>
1035 <para><Option>-t0</Option> requests no trace elements to be
1036 removed from the profile, ensuring that all the data will be
1042 <term><Option>-c</Option></Term>
1044 <para>Generate colour output.</para>
1049 <term><Option>-y</Option></Term>
1051 <para>Ignore marks.</para>
1056 <term><Option>-?</Option></Term>
1058 <para>Print out usage information.</para>
1064 <sect1 id="ticky-ticky">
1065 <title>Using “ticky-ticky” profiling (for implementors)</Title>
1066 <indexterm><primary>ticky-ticky profiling</primary></indexterm>
1068 <para>(ToDo: document properly.)</para>
1070 <para>It is possible to compile Glasgow Haskell programs so that
1071 they will count lots and lots of interesting things, e.g., number
1072 of updates, number of data constructors entered, etc., etc. We
1073 call this “ticky-ticky”
1074 profiling,<indexterm><primary>ticky-ticky
1075 profiling</primary></indexterm> <indexterm><primary>profiling,
1076 ticky-ticky</primary></indexterm> because that's the sound a Sun4
1077 makes when it is running up all those counters
1078 (<Emphasis>slowly</Emphasis>).</para>
1080 <para>Ticky-ticky profiling is mainly intended for implementors;
1081 it is quite separate from the main “cost-centre”
1082 profiling system, intended for all users everywhere.</para>
1084 <para>To be able to use ticky-ticky profiling, you will need to
1085 have built appropriate libraries and things when you made the
1086 system. See “Customising what libraries to build,” in
1087 the installation guide.</para>
1089 <para>To get your compiled program to spit out the ticky-ticky
1090 numbers, use a <Option>-r</Option> RTS
1091 option<indexterm><primary>-r RTS option</primary></indexterm>.
1092 See <XRef LinkEnd="runtime-control">.</para>
1094 <para>Compiling your program with the <Option>-ticky</Option>
1095 switch yields an executable that performs these counts. Here is a
1096 sample ticky-ticky statistics file, generated by the invocation
1097 <command>foo +RTS -rfoo.ticky</command>.</para>
1100 foo +RTS -rfoo.ticky
1103 ALLOCATIONS: 3964631 (11330900 words total: 3999476 admin, 6098829 goods, 1232595 slop)
1104 total words: 2 3 4 5 6+
1105 69647 ( 1.8%) function values 50.0 50.0 0.0 0.0 0.0
1106 2382937 ( 60.1%) thunks 0.0 83.9 16.1 0.0 0.0
1107 1477218 ( 37.3%) data values 66.8 33.2 0.0 0.0 0.0
1108 0 ( 0.0%) big tuples
1109 2 ( 0.0%) black holes 0.0 100.0 0.0 0.0 0.0
1110 0 ( 0.0%) prim things
1111 34825 ( 0.9%) partial applications 0.0 0.0 0.0 100.0 0.0
1112 2 ( 0.0%) thread state objects 0.0 0.0 0.0 0.0 100.0
1114 Total storage-manager allocations: 3647137 (11882004 words)
1115 [551104 words lost to speculative heap-checks]
1119 ENTERS: 9400092 of which 2005772 (21.3%) direct to the entry code
1120 [the rest indirected via Node's info ptr]
1121 1860318 ( 19.8%) thunks
1122 3733184 ( 39.7%) data values
1123 3149544 ( 33.5%) function values
1124 [of which 1999880 (63.5%) bypassed arg-satisfaction chk]
1125 348140 ( 3.7%) partial applications
1126 308906 ( 3.3%) normal indirections
1127 0 ( 0.0%) permanent indirections
1130 2137257 ( 36.4%) from entering a new constructor
1131 [the rest from entering an existing constructor]
1132 2349219 ( 40.0%) vectored [the rest unvectored]
1134 RET_NEW: 2137257: 32.5% 46.2% 21.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1135 RET_OLD: 3733184: 2.8% 67.9% 29.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1136 RET_UNBOXED_TUP: 2: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1138 RET_VEC_RETURN : 2349219: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1140 UPDATE FRAMES: 2241725 (0 omitted from thunks)
1144 0 ( 0.0%) data values
1145 34827 ( 1.6%) partial applications
1146 [2 in place, 34825 allocated new space]
1147 2206898 ( 98.4%) updates to existing heap objects (46 by squeezing)
1148 UPD_CON_IN_NEW: 0: 0 0 0 0 0 0 0 0 0
1149 UPD_PAP_IN_NEW: 34825: 0 0 0 34825 0 0 0 0 0
1151 NEW GEN UPDATES: 2274700 ( 99.9%)
1153 OLD GEN UPDATES: 1852 ( 0.1%)
1155 Total bytes copied during GC: 190096
1157 **************************************************
1158 3647137 ALLOC_HEAP_ctr
1159 11882004 ALLOC_HEAP_tot
1164 34831 ALLOC_FUN_hst_0
1165 34816 ALLOC_FUN_hst_1
1169 2382937 ALLOC_UP_THK_ctr
1172 0 E!NT_PERM_IND_ctr requires +RTS -Z
1173 [... lots more info omitted ...]
1174 0 GC_SEL_ABANDONED_ctr
1177 0 GC_FAILED_PROMOTION_ctr
1178 47524 GC_WORDS_COPIED_ctr
1181 <para>The formatting of the information above the row of asterisks
1182 is subject to change, but hopefully provides a useful
1183 human-readable summary. Below the asterisks <Emphasis>all
1184 counters</Emphasis> maintained by the ticky-ticky system are
1185 dumped, in a format intended to be machine-readable: zero or more
1186 spaces, an integer, a space, the counter name, and a newline.</para>
1188 <para>In fact, not <Emphasis>all</Emphasis> counters are
1189 necessarily dumped; compile- or run-time flags can render certain
1190 counters invalid. In this case, either the counter will simply
1191 not appear, or it will appear with a modified counter name,
1192 possibly along with an explanation for the omission (notice
1193 <literal>ENT_PERM_IND_ctr</literal> appears
1194 with an inserted <literal>!</literal> above). Software analysing
1195 this output should always check that it has the counters it
1196 expects. Also, beware: some of the counters can have
1197 <Emphasis>large</Emphasis> values!</para>
1204 ;;; Local Variables: ***
1206 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter") ***