1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <chapter id="profiling">
3 <title>Profiling</title>
4 <indexterm><primary>profiling</primary>
6 <indexterm><primary>cost-centre profiling</primary></indexterm>
8 <para> Glasgow Haskell comes with a time and space profiling
9 system. Its purpose is to help you improve your understanding of
10 your program's execution behaviour, so you can improve it.</para>
12 <para> Any comments, suggestions and/or improvements you have are
13 welcome. Recommended “profiling tricks” would be
14 especially cool! </para>
16 <para>Profiling a program is a three-step process:</para>
20 <para> Re-compile your program for profiling with the
21 <literal>-prof</literal> option, and probably one of the
22 <literal>-auto</literal> or <literal>-auto-all</literal>
23 options. These options are described in more detail in <xref
24 linkend="prof-compiler-options"/> </para>
25 <indexterm><primary><literal>-prof</literal></primary>
27 <indexterm><primary><literal>-auto</literal></primary>
29 <indexterm><primary><literal>-auto-all</literal></primary>
34 <para> Run your program with one of the profiling options, eg.
35 <literal>+RTS -p -RTS</literal>. This generates a file of
36 profiling information. Note that multi-processor execution
37 (e.g. <literal>+RTS -N2</literal>) is not supported while
39 <indexterm><primary><option>-p</option></primary><secondary>RTS
40 option</secondary></indexterm>
44 <para> Examine the generated profiling information, using one of
45 GHC's profiling tools. The tool to use will depend on the kind
46 of profiling information generated.</para>
51 <sect1 id="cost-centres">
52 <title>Cost centres and cost-centre stacks</title>
54 <para>GHC's profiling system assigns <firstterm>costs</firstterm>
55 to <firstterm>cost centres</firstterm>. A cost is simply the time
56 or space required to evaluate an expression. Cost centres are
57 program annotations around expressions; all costs incurred by the
58 annotated expression are assigned to the enclosing cost centre.
59 Furthermore, GHC will remember the stack of enclosing cost centres
60 for any given expression at run-time and generate a call-graph of
61 cost attributions.</para>
63 <para>Let's take a look at an example:</para>
66 main = print (nfib 25)
67 nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)
70 <para>Compile and run this program as follows:</para>
73 $ ghc -prof -auto-all -o Main Main.hs
79 <para>When a GHC-compiled program is run with the
80 <option>-p</option> RTS option, it generates a file called
81 <filename><prog>.prof</filename>. In this case, the file
82 will contain something like this:</para>
85 Fri May 12 14:06 2000 Time and Allocation Profiling Report (Final)
89 total time = 0.14 secs (7 ticks @ 20 ms)
90 total alloc = 8,741,204 bytes (excludes profiling overheads)
92 COST CENTRE MODULE %time %alloc
98 COST CENTRE MODULE entries %time %alloc %time %alloc
100 MAIN MAIN 0 0.0 0.0 100.0 100.0
101 main Main 0 0.0 0.0 0.0 0.0
102 CAF PrelHandle 3 0.0 0.0 0.0 0.0
103 CAF PrelAddr 1 0.0 0.0 0.0 0.0
104 CAF Main 6 0.0 0.0 100.0 100.0
105 main Main 1 0.0 0.0 100.0 100.0
106 nfib Main 242785 100.0 100.0 100.0 100.0
110 <para>The first part of the file gives the program name and
111 options, and the total time and total memory allocation measured
112 during the run of the program (note that the total memory
113 allocation figure isn't the same as the amount of
114 <emphasis>live</emphasis> memory needed by the program at any one
115 time; the latter can be determined using heap profiling, which we
116 will describe shortly).</para>
118 <para>The second part of the file is a break-down by cost centre
119 of the most costly functions in the program. In this case, there
120 was only one significant function in the program, namely
121 <function>nfib</function>, and it was responsible for 100%
122 of both the time and allocation costs of the program.</para>
124 <para>The third and final section of the file gives a profile
125 break-down by cost-centre stack. This is roughly a call-graph
126 profile of the program. In the example above, it is clear that
127 the costly call to <function>nfib</function> came from
128 <function>main</function>.</para>
130 <para>The time and allocation incurred by a given part of the
131 program is displayed in two ways: “individual”, which
132 are the costs incurred by the code covered by this cost centre
133 stack alone, and “inherited”, which includes the costs
134 incurred by all the children of this node.</para>
136 <para>The usefulness of cost-centre stacks is better demonstrated
137 by modifying the example slightly:</para>
140 main = print (f 25 + g 25)
142 g n = nfib (n `div` 2)
143 nfib n = if n < 2 then 1 else nfib (n-1) + nfib (n-2)
146 <para>Compile and run this program as before, and take a look at
147 the new profiling results:</para>
150 COST CENTRE MODULE scc %time %alloc %time %alloc
152 MAIN MAIN 0 0.0 0.0 100.0 100.0
153 main Main 0 0.0 0.0 0.0 0.0
154 CAF PrelHandle 3 0.0 0.0 0.0 0.0
155 CAF PrelAddr 1 0.0 0.0 0.0 0.0
156 CAF Main 9 0.0 0.0 100.0 100.0
157 main Main 1 0.0 0.0 100.0 100.0
158 g Main 1 0.0 0.0 0.0 0.2
159 nfib Main 465 0.0 0.2 0.0 0.2
160 f Main 1 0.0 0.0 100.0 99.8
161 nfib Main 242785 100.0 99.8 100.0 99.8
164 <para>Now although we had two calls to <function>nfib</function>
165 in the program, it is immediately clear that it was the call from
166 <function>f</function> which took all the time.</para>
168 <para>The actual meaning of the various columns in the output is:</para>
174 <para>The number of times this particular point in the call
175 graph was entered.</para>
180 <term>individual %time</term>
182 <para>The percentage of the total run time of the program
183 spent at this point in the call graph.</para>
188 <term>individual %alloc</term>
190 <para>The percentage of the total memory allocations
191 (excluding profiling overheads) of the program made by this
197 <term>inherited %time</term>
199 <para>The percentage of the total run time of the program
200 spent below this point in the call graph.</para>
205 <term>inherited %alloc</term>
207 <para>The percentage of the total memory allocations
208 (excluding profiling overheads) of the program made by this
209 call and all of its sub-calls.</para>
214 <para>In addition you can use the <option>-P</option> RTS option
215 <indexterm><primary><option>-P</option></primary></indexterm> to
216 get the following additional information:</para>
220 <term><literal>ticks</literal></term>
222 <para>The raw number of time “ticks” which were
223 attributed to this cost-centre; from this, we get the
224 <literal>%time</literal> figure mentioned
230 <term><literal>bytes</literal></term>
232 <para>Number of bytes allocated in the heap while in this
233 cost-centre; again, this is the raw number from which we get
234 the <literal>%alloc</literal> figure mentioned
240 <para>What about recursive functions, and mutually recursive
241 groups of functions? Where are the costs attributed? Well,
242 although GHC does keep information about which groups of functions
243 called each other recursively, this information isn't displayed in
244 the basic time and allocation profile, instead the call-graph is
245 flattened into a tree.</para>
247 <sect2><title>Inserting cost centres by hand</title>
249 <para>Cost centres are just program annotations. When you say
250 <option>-auto-all</option> to the compiler, it automatically
251 inserts a cost centre annotation around every top-level function
252 in your program, but you are entirely free to add the cost
253 centre annotations yourself.</para>
255 <para>The syntax of a cost centre annotation is</para>
258 {-# SCC "name" #-} <expression>
261 <para>where <literal>"name"</literal> is an arbitrary string,
262 that will become the name of your cost centre as it appears
263 in the profiling output, and
264 <literal><expression></literal> is any Haskell
265 expression. An <literal>SCC</literal> annotation extends as
266 far to the right as possible when parsing. (SCC stands for "Set
267 Cost Centre").</para>
269 <para>Here is an example of a program with a couple of SCCs:</para>
273 main = do let xs = {-# SCC "X" #-} [1..1000000]
274 let ys = {-# SCC "Y" #-} [1..2000000]
276 print $ last $ init xs
278 print $ last $ init ys
281 <para>which gives this heap profile when run:</para>
283 <!-- contentwidth/contentheight don't appear to have any effect
284 other than making the PS file generation work, rather than
285 falling over. The result seems to be broken PS on the page
287 <imagedata fileref="prof_scc" contentwidth="645px"
288 contentdepth="428px"/>
292 <sect2 id="prof-rules">
293 <title>Rules for attributing costs</title>
295 <para>The cost of evaluating any expression in your program is
296 attributed to a cost-centre stack using the following rules:</para>
300 <para>If the expression is part of the
301 <firstterm>one-off</firstterm> costs of evaluating the
302 enclosing top-level definition, then costs are attributed to
303 the stack of lexically enclosing <literal>SCC</literal>
304 annotations on top of the special <literal>CAF</literal>
309 <para>Otherwise, costs are attributed to the stack of
310 lexically-enclosing <literal>SCC</literal> annotations,
311 appended to the cost-centre stack in effect at the
312 <firstterm>call site</firstterm> of the current top-level
313 definition<footnote> <para>The call-site is just the place
314 in the source code which mentions the particular function or
315 variable.</para></footnote>. Notice that this is a recursive
320 <para>Time spent in foreign code (see <xref linkend="ffi"/>)
321 is always attributed to the cost centre in force at the
322 Haskell call-site of the foreign function.</para>
326 <para>What do we mean by one-off costs? Well, Haskell is a lazy
327 language, and certain expressions are only ever evaluated once.
328 For example, if we write:</para>
334 <para>then <varname>x</varname> will only be evaluated once (if
335 at all), and subsequent demands for <varname>x</varname> will
336 immediately get to see the cached result. The definition
337 <varname>x</varname> is called a CAF (Constant Applicative
338 Form), because it has no arguments.</para>
340 <para>For the purposes of profiling, we say that the expression
341 <literal>nfib 25</literal> belongs to the one-off costs of
342 evaluating <varname>x</varname>.</para>
344 <para>Since one-off costs aren't strictly speaking part of the
345 call-graph of the program, they are attributed to a special
346 top-level cost centre, <literal>CAF</literal>. There may be one
347 <literal>CAF</literal> cost centre for each module (the
348 default), or one for each top-level definition with any one-off
349 costs (this behaviour can be selected by giving GHC the
350 <option>-caf-all</option> flag).</para>
352 <indexterm><primary><literal>-caf-all</literal></primary>
355 <para>If you think you have a weird profile, or the call-graph
356 doesn't look like you expect it to, feel free to send it (and
357 your program) to us at
358 <email>glasgow-haskell-bugs@haskell.org</email>.</para>
362 <sect1 id="prof-compiler-options">
363 <title>Compiler options for profiling</title>
365 <indexterm><primary>profiling</primary><secondary>options</secondary></indexterm>
366 <indexterm><primary>options</primary><secondary>for profiling</secondary></indexterm>
371 <option>-prof</option>:
372 <indexterm><primary><option>-prof</option></primary></indexterm>
375 <para> To make use of the profiling system
376 <emphasis>all</emphasis> modules must be compiled and linked
377 with the <option>-prof</option> option. Any
378 <literal>SCC</literal> annotations you've put in your source
379 will spring to life.</para>
381 <para> Without a <option>-prof</option> option, your
382 <literal>SCC</literal>s are ignored; so you can compile
383 <literal>SCC</literal>-laden code without changing
389 <para>There are a few other profiling-related compilation options.
390 Use them <emphasis>in addition to</emphasis>
391 <option>-prof</option>. These do not have to be used consistently
392 for all modules in a program.</para>
397 <option>-auto</option>:
398 <indexterm><primary><option>-auto</option></primary></indexterm>
399 <indexterm><primary>cost centres</primary><secondary>automatically inserting</secondary></indexterm>
402 <para> GHC will automatically add
403 <function>_scc_</function> constructs for all
404 top-level, exported functions.</para>
410 <option>-auto-all</option>:
411 <indexterm><primary><option>-auto-all</option></primary></indexterm>
414 <para> <emphasis>All</emphasis> top-level functions,
415 exported or not, will be automatically
416 <function>_scc_</function>'d.</para>
422 <option>-caf-all</option>:
423 <indexterm><primary><option>-caf-all</option></primary></indexterm>
426 <para> The costs of all CAFs in a module are usually
427 attributed to one “big” CAF cost-centre. With
428 this option, all CAFs get their own cost-centre. An
429 “if all else fails” option…</para>
435 <option>-ignore-scc</option>:
436 <indexterm><primary><option>-ignore-scc</option></primary></indexterm>
439 <para>Ignore any <function>_scc_</function>
440 constructs, so a module which already has
441 <function>_scc_</function>s can be compiled
442 for profiling with the annotations ignored.</para>
450 <sect1 id="prof-time-options">
451 <title>Time and allocation profiling</title>
453 <para>To generate a time and allocation profile, give one of the
454 following RTS options to the compiled program when you run it (RTS
455 options should be enclosed between <literal>+RTS...-RTS</literal>
461 <option>-p</option> or <option>-P</option>:
462 <indexterm><primary><option>-p</option></primary></indexterm>
463 <indexterm><primary><option>-P</option></primary></indexterm>
464 <indexterm><primary>time profile</primary></indexterm>
467 <para>The <option>-p</option> option produces a standard
468 <emphasis>time profile</emphasis> report. It is written
470 <filename><replaceable>program</replaceable>.prof</filename>.</para>
472 <para>The <option>-P</option> option produces a more
473 detailed report containing the actual time and allocation
474 data as well. (Not used much.)</para>
481 <indexterm><primary><option>-xc</option></primary><secondary>RTS option</secondary></indexterm>
484 <para>This option makes use of the extra information
485 maintained by the cost-centre-stack profiler to provide
486 useful information about the location of runtime errors.
487 See <xref linkend="rts-options-debugging"/>.</para>
495 <sect1 id="prof-heap">
496 <title>Profiling memory usage</title>
498 <para>In addition to profiling the time and allocation behaviour
499 of your program, you can also generate a graph of its memory usage
500 over time. This is useful for detecting the causes of
501 <firstterm>space leaks</firstterm>, when your program holds on to
502 more memory at run-time that it needs to. Space leaks lead to
503 longer run-times due to heavy garbage collector activity, and may
504 even cause the program to run out of memory altogether.</para>
506 <para>To generate a heap profile from your program:</para>
510 <para>Compile the program for profiling (<xref
511 linkend="prof-compiler-options"/>).</para>
514 <para>Run it with one of the heap profiling options described
515 below (eg. <option>-hc</option> for a basic producer profile).
516 This generates the file
517 <filename><replaceable>prog</replaceable>.hp</filename>.</para>
520 <para>Run <command>hp2ps</command> to produce a Postscript
522 <filename><replaceable>prog</replaceable>.ps</filename>. The
523 <command>hp2ps</command> utility is described in detail in
524 <xref linkend="hp2ps"/>.</para>
527 <para>Display the heap profile using a postscript viewer such
528 as <application>Ghostview</application>, or print it out on a
529 Postscript-capable printer.</para>
533 <para>You might also want to take a look
534 at <ulink url="http://www.haskell.org/haskellwiki/Hp2any">hp2any</ulink>,
535 a more advanced suite of tools (not distributed with GHC) for
536 displaying heap profiles.</para>
538 <sect2 id="rts-options-heap-prof">
539 <title>RTS options for heap profiling</title>
541 <para>There are several different kinds of heap profile that can
542 be generated. All the different profile types yield a graph of
543 live heap against time, but they differ in how the live heap is
544 broken down into bands. The following RTS options select which
545 break-down to use:</para>
551 <indexterm><primary><option>-hc</option></primary><secondary>RTS option</secondary></indexterm>
554 <para>Breaks down the graph by the cost-centre stack which
555 produced the data.</para>
562 <indexterm><primary><option>-hm</option></primary><secondary>RTS option</secondary></indexterm>
565 <para>Break down the live heap by the module containing
566 the code which produced the data.</para>
573 <indexterm><primary><option>-hd</option></primary><secondary>RTS option</secondary></indexterm>
576 <para>Breaks down the graph by <firstterm>closure
577 description</firstterm>. For actual data, the description
578 is just the constructor name, for other closures it is a
579 compiler-generated string identifying the closure.</para>
586 <indexterm><primary><option>-hy</option></primary><secondary>RTS option</secondary></indexterm>
589 <para>Breaks down the graph by
590 <firstterm>type</firstterm>. For closures which have
591 function type or unknown/polymorphic type, the string will
592 represent an approximation to the actual type.</para>
599 <indexterm><primary><option>-hr</option></primary><secondary>RTS option</secondary></indexterm>
602 <para>Break down the graph by <firstterm>retainer
603 set</firstterm>. Retainer profiling is described in more
604 detail below (<xref linkend="retainer-prof"/>).</para>
611 <indexterm><primary><option>-hb</option></primary><secondary>RTS option</secondary></indexterm>
614 <para>Break down the graph by
615 <firstterm>biography</firstterm>. Biographical profiling
616 is described in more detail below (<xref
617 linkend="biography-prof"/>).</para>
622 <para>In addition, the profile can be restricted to heap data
623 which satisfies certain criteria - for example, you might want
624 to display a profile by type but only for data produced by a
625 certain module, or a profile by retainer for a certain type of
626 data. Restrictions are specified as follows:</para>
631 <option>-hc</option><replaceable>name</replaceable>,...
632 <indexterm><primary><option>-hc</option></primary><secondary>RTS option</secondary></indexterm>
635 <para>Restrict the profile to closures produced by
636 cost-centre stacks with one of the specified cost centres
643 <option>-hC</option><replaceable>name</replaceable>,...
644 <indexterm><primary><option>-hC</option></primary><secondary>RTS option</secondary></indexterm>
647 <para>Restrict the profile to closures produced by
648 cost-centre stacks with one of the specified cost centres
649 anywhere in the stack.</para>
655 <option>-hm</option><replaceable>module</replaceable>,...
656 <indexterm><primary><option>-hm</option></primary><secondary>RTS option</secondary></indexterm>
659 <para>Restrict the profile to closures produced by the
660 specified modules.</para>
666 <option>-hd</option><replaceable>desc</replaceable>,...
667 <indexterm><primary><option>-hd</option></primary><secondary>RTS option</secondary></indexterm>
670 <para>Restrict the profile to closures with the specified
671 description strings.</para>
677 <option>-hy</option><replaceable>type</replaceable>,...
678 <indexterm><primary><option>-hy</option></primary><secondary>RTS option</secondary></indexterm>
681 <para>Restrict the profile to closures with the specified
688 <option>-hr</option><replaceable>cc</replaceable>,...
689 <indexterm><primary><option>-hr</option></primary><secondary>RTS option</secondary></indexterm>
692 <para>Restrict the profile to closures with retainer sets
693 containing cost-centre stacks with one of the specified
694 cost centres at the top.</para>
700 <option>-hb</option><replaceable>bio</replaceable>,...
701 <indexterm><primary><option>-hb</option></primary><secondary>RTS option</secondary></indexterm>
704 <para>Restrict the profile to closures with one of the
705 specified biographies, where
706 <replaceable>bio</replaceable> is one of
707 <literal>lag</literal>, <literal>drag</literal>,
708 <literal>void</literal>, or <literal>use</literal>.</para>
713 <para>For example, the following options will generate a
714 retainer profile restricted to <literal>Branch</literal> and
715 <literal>Leaf</literal> constructors:</para>
718 <replaceable>prog</replaceable> +RTS -hr -hdBranch,Leaf
721 <para>There can only be one "break-down" option
722 (eg. <option>-hr</option> in the example above), but there is no
723 limit on the number of further restrictions that may be applied.
724 All the options may be combined, with one exception: GHC doesn't
725 currently support mixing the <option>-hr</option> and
726 <option>-hb</option> options.</para>
728 <para>There are three more options which relate to heap
734 <option>-i<replaceable>secs</replaceable></option>:
735 <indexterm><primary><option>-i</option></primary></indexterm>
738 <para>Set the profiling (sampling) interval to
739 <replaceable>secs</replaceable> seconds (the default is
740 0.1 second). Fractions are allowed: for example
741 <option>-i0.2</option> will get 5 samples per second.
742 This only affects heap profiling; time profiles are always
743 sampled on a 1/50 second frequency.</para>
750 <indexterm><primary><option>-xt</option></primary><secondary>RTS option</secondary></indexterm>
753 <para>Include the memory occupied by threads in a heap
754 profile. Each thread takes up a small area for its thread
755 state in addition to the space allocated for its stack
756 (stacks normally start small and then grow as
759 <para>This includes the main thread, so using
760 <option>-xt</option> is a good way to see how much stack
761 space the program is using.</para>
763 <para>Memory occupied by threads and their stacks is
764 labelled as “TSO” when displaying the profile
765 by closure description or type description.</para>
771 <option>-L<replaceable>num</replaceable></option>
772 <indexterm><primary><option>-L</option></primary><secondary>RTS option</secondary></indexterm>
776 Sets the maximum length of a cost-centre stack name in a
777 heap profile. Defaults to 25.
785 <sect2 id="retainer-prof">
786 <title>Retainer Profiling</title>
788 <para>Retainer profiling is designed to help answer questions
789 like <quote>why is this data being retained?</quote>. We start
790 by defining what we mean by a retainer:</para>
793 <para>A retainer is either the system stack, or an unevaluated
794 closure (thunk).</para>
797 <para>In particular, constructors are <emphasis>not</emphasis>
800 <para>An object B retains object A if (i) B is a retainer object and
801 (ii) object A can be reached by recursively following pointers
802 starting from object B, but not meeting any other retainer
803 objects on the way. Each live object is retained by one or more
804 retainer objects, collectively called its retainer set, or its
805 <firstterm>retainer set</firstterm>, or its
806 <firstterm>retainers</firstterm>.</para>
808 <para>When retainer profiling is requested by giving the program
809 the <option>-hr</option> option, a graph is generated which is
810 broken down by retainer set. A retainer set is displayed as a
811 set of cost-centre stacks; because this is usually too large to
812 fit on the profile graph, each retainer set is numbered and
813 shown abbreviated on the graph along with its number, and the
814 full list of retainer sets is dumped into the file
815 <filename><replaceable>prog</replaceable>.prof</filename>.</para>
817 <para>Retainer profiling requires multiple passes over the live
818 heap in order to discover the full retainer set for each
819 object, which can be quite slow. So we set a limit on the
820 maximum size of a retainer set, where all retainer sets larger
821 than the maximum retainer set size are replaced by the special
822 set <literal>MANY</literal>. The maximum set size defaults to 8
823 and can be altered with the <option>-R</option> RTS
828 <term><option>-R</option><replaceable>size</replaceable></term>
830 <para>Restrict the number of elements in a retainer set to
831 <replaceable>size</replaceable> (default 8).</para>
837 <title>Hints for using retainer profiling</title>
839 <para>The definition of retainers is designed to reflect a
840 common cause of space leaks: a large structure is retained by
841 an unevaluated computation, and will be released once the
842 computation is forced. A good example is looking up a value in
843 a finite map, where unless the lookup is forced in a timely
844 manner the unevaluated lookup will cause the whole mapping to
845 be retained. These kind of space leaks can often be
846 eliminated by forcing the relevant computations to be
847 performed eagerly, using <literal>seq</literal> or strictness
848 annotations on data constructor fields.</para>
850 <para>Often a particular data structure is being retained by a
851 chain of unevaluated closures, only the nearest of which will
852 be reported by retainer profiling - for example A retains B, B
853 retains C, and C retains a large structure. There might be a
854 large number of Bs but only a single A, so A is really the one
855 we're interested in eliminating. However, retainer profiling
856 will in this case report B as the retainer of the large
857 structure. To move further up the chain of retainers, we can
858 ask for another retainer profile but this time restrict the
859 profile to B objects, so we get a profile of the retainers of
863 <replaceable>prog</replaceable> +RTS -hr -hcB
866 <para>This trick isn't foolproof, because there might be other
867 B closures in the heap which aren't the retainers we are
868 interested in, but we've found this to be a useful technique
869 in most cases.</para>
873 <sect2 id="biography-prof">
874 <title>Biographical Profiling</title>
876 <para>A typical heap object may be in one of the following four
877 states at each point in its lifetime:</para>
881 <para>The <firstterm>lag</firstterm> stage, which is the
882 time between creation and the first use of the
886 <para>the <firstterm>use</firstterm> stage, which lasts from
887 the first use until the last use of the object, and</para>
890 <para>The <firstterm>drag</firstterm> stage, which lasts
891 from the final use until the last reference to the object
895 <para>An object which is never used is said to be in the
896 <firstterm>void</firstterm> state for its whole
901 <para>A biographical heap profile displays the portion of the
902 live heap in each of the four states listed above. Usually the
903 most interesting states are the void and drag states: live heap
904 in these states is more likely to be wasted space than heap in
905 the lag or use states.</para>
907 <para>It is also possible to break down the heap in one or more
908 of these states by a different criteria, by restricting a
909 profile by biography. For example, to show the portion of the
910 heap in the drag or void state by producer: </para>
913 <replaceable>prog</replaceable> +RTS -hc -hbdrag,void
916 <para>Once you know the producer or the type of the heap in the
917 drag or void states, the next step is usually to find the
921 <replaceable>prog</replaceable> +RTS -hr -hc<replaceable>cc</replaceable>...
924 <para>NOTE: this two stage process is required because GHC
925 cannot currently profile using both biographical and retainer
926 information simultaneously.</para>
929 <sect2 id="mem-residency">
930 <title>Actual memory residency</title>
932 <para>How does the heap residency reported by the heap profiler relate to
933 the actual memory residency of your program when you run it? You might
934 see a large discrepancy between the residency reported by the heap
935 profiler, and the residency reported by tools on your system
936 (eg. <literal>ps</literal> or <literal>top</literal> on Unix, or the
937 Task Manager on Windows). There are several reasons for this:</para>
941 <para>There is an overhead of profiling itself, which is subtracted
942 from the residency figures by the profiler. This overhead goes
943 away when compiling without profiling support, of course. The
944 space overhead is currently 2 extra
945 words per heap object, which probably results in
946 about a 30% overhead.</para>
950 <para>Garbage collection requires more memory than the actual
951 residency. The factor depends on the kind of garbage collection
952 algorithm in use: a major GC in the standard
953 generation copying collector will usually require 3L bytes of
954 memory, where L is the amount of live data. This is because by
955 default (see the <option>+RTS -F</option> option) we allow the old
956 generation to grow to twice its size (2L) before collecting it, and
957 we require additionally L bytes to copy the live data into. When
958 using compacting collection (see the <option>+RTS -c</option>
959 option), this is reduced to 2L, and can further be reduced by
960 tweaking the <option>-F</option> option. Also add the size of the
961 allocation area (currently a fixed 512Kb).</para>
965 <para>The stack isn't counted in the heap profile by default. See the
966 <option>+RTS -xt</option> option.</para>
970 <para>The program text itself, the C stack, any non-heap data (eg. data
971 allocated by foreign libraries, and data allocated by the RTS), and
972 <literal>mmap()</literal>'d memory are not counted in the heap profile.</para>
980 <title><command>hp2ps</command>––heap profile to PostScript</title>
982 <indexterm><primary><command>hp2ps</command></primary></indexterm>
983 <indexterm><primary>heap profiles</primary></indexterm>
984 <indexterm><primary>postscript, from heap profiles</primary></indexterm>
985 <indexterm><primary><option>-h<break-down></option></primary></indexterm>
990 hp2ps [flags] [<file>[.hp]]
994 <command>hp2ps</command><indexterm><primary>hp2ps
995 program</primary></indexterm> converts a heap profile as produced
996 by the <option>-h<break-down></option> runtime option into a
997 PostScript graph of the heap profile. By convention, the file to
998 be processed by <command>hp2ps</command> has a
999 <filename>.hp</filename> extension. The PostScript output is
1000 written to <filename><file>@.ps</filename>. If
1001 <filename><file></filename> is omitted entirely, then the
1002 program behaves as a filter.</para>
1004 <para><command>hp2ps</command> is distributed in
1005 <filename>ghc/utils/hp2ps</filename> in a GHC source
1006 distribution. It was originally developed by Dave Wakeling as part
1007 of the HBC/LML heap profiler.</para>
1009 <para>The flags are:</para>
1014 <term><option>-d</option></term>
1016 <para>In order to make graphs more readable,
1017 <command>hp2ps</command> sorts the shaded bands for each
1018 identifier. The default sort ordering is for the bands with
1019 the largest area to be stacked on top of the smaller ones.
1020 The <option>-d</option> option causes rougher bands (those
1021 representing series of values with the largest standard
1022 deviations) to be stacked on top of smoother ones.</para>
1027 <term><option>-b</option></term>
1029 <para>Normally, <command>hp2ps</command> puts the title of
1030 the graph in a small box at the top of the page. However, if
1031 the JOB string is too long to fit in a small box (more than
1032 35 characters), then <command>hp2ps</command> will choose to
1033 use a big box instead. The <option>-b</option> option
1034 forces <command>hp2ps</command> to use a big box.</para>
1039 <term><option>-e<float>[in|mm|pt]</option></term>
1041 <para>Generate encapsulated PostScript suitable for
1042 inclusion in LaTeX documents. Usually, the PostScript graph
1043 is drawn in landscape mode in an area 9 inches wide by 6
1044 inches high, and <command>hp2ps</command> arranges for this
1045 area to be approximately centred on a sheet of a4 paper.
1046 This format is convenient of studying the graph in detail,
1047 but it is unsuitable for inclusion in LaTeX documents. The
1048 <option>-e</option> option causes the graph to be drawn in
1049 portrait mode, with float specifying the width in inches,
1050 millimetres or points (the default). The resulting
1051 PostScript file conforms to the Encapsulated PostScript
1052 (EPS) convention, and it can be included in a LaTeX document
1053 using Rokicki's dvi-to-PostScript converter
1054 <command>dvips</command>.</para>
1059 <term><option>-g</option></term>
1061 <para>Create output suitable for the <command>gs</command>
1062 PostScript previewer (or similar). In this case the graph is
1063 printed in portrait mode without scaling. The output is
1064 unsuitable for a laser printer.</para>
1069 <term><option>-l</option></term>
1071 <para>Normally a profile is limited to 20 bands with
1072 additional identifiers being grouped into an
1073 <literal>OTHER</literal> band. The <option>-l</option> flag
1074 removes this 20 band and limit, producing as many bands as
1075 necessary. No key is produced as it won't fit!. It is useful
1076 for creation time profiles with many bands.</para>
1081 <term><option>-m<int></option></term>
1083 <para>Normally a profile is limited to 20 bands with
1084 additional identifiers being grouped into an
1085 <literal>OTHER</literal> band. The <option>-m</option> flag
1086 specifies an alternative band limit (the maximum is
1089 <para><option>-m0</option> requests the band limit to be
1090 removed. As many bands as necessary are produced. However no
1091 key is produced as it won't fit! It is useful for displaying
1092 creation time profiles with many bands.</para>
1097 <term><option>-p</option></term>
1099 <para>Use previous parameters. By default, the PostScript
1100 graph is automatically scaled both horizontally and
1101 vertically so that it fills the page. However, when
1102 preparing a series of graphs for use in a presentation, it
1103 is often useful to draw a new graph using the same scale,
1104 shading and ordering as a previous one. The
1105 <option>-p</option> flag causes the graph to be drawn using
1106 the parameters determined by a previous run of
1107 <command>hp2ps</command> on <filename>file</filename>. These
1108 are extracted from <filename>file@.aux</filename>.</para>
1113 <term><option>-s</option></term>
1115 <para>Use a small box for the title.</para>
1120 <term><option>-t<float></option></term>
1122 <para>Normally trace elements which sum to a total of less
1123 than 1% of the profile are removed from the
1124 profile. The <option>-t</option> option allows this
1125 percentage to be modified (maximum 5%).</para>
1127 <para><option>-t0</option> requests no trace elements to be
1128 removed from the profile, ensuring that all the data will be
1134 <term><option>-c</option></term>
1136 <para>Generate colour output.</para>
1141 <term><option>-y</option></term>
1143 <para>Ignore marks.</para>
1148 <term><option>-?</option></term>
1150 <para>Print out usage information.</para>
1156 <sect2 id="manipulating-hp">
1157 <title>Manipulating the hp file</title>
1159 <para>(Notes kindly offered by Jan-Willhem Maessen.)</para>
1162 The <filename>FOO.hp</filename> file produced when you ask for the
1163 heap profile of a program <filename>FOO</filename> is a text file with a particularly
1164 simple structure. Here's a representative example, with much of the
1165 actual data omitted:
1168 DATE "Thu Dec 26 18:17 2002"
1169 SAMPLE_UNIT "seconds"
1180 BEGIN_SAMPLE 11695.47
1183 The first four lines (<literal>JOB</literal>, <literal>DATE</literal>, <literal>SAMPLE_UNIT</literal>, <literal>VALUE_UNIT</literal>) form a
1184 header. Each block of lines starting with <literal>BEGIN_SAMPLE</literal> and ending
1185 with <literal>END_SAMPLE</literal> forms a single sample (you can think of this as a
1186 vertical slice of your heap profile). The hp2ps utility should accept
1187 any input with a properly-formatted header followed by a series of
1193 <title>Zooming in on regions of your profile</title>
1196 You can look at particular regions of your profile simply by loading a
1197 copy of the <filename>.hp</filename> file into a text editor and deleting the unwanted
1198 samples. The resulting <filename>.hp</filename> file can be run through <command>hp2ps</command> and viewed
1204 <title>Viewing the heap profile of a running program</title>
1207 The <filename>.hp</filename> file is generated incrementally as your
1208 program runs. In principle, running <command>hp2ps</command> on the incomplete file
1209 should produce a snapshot of your program's heap usage. However, the
1210 last sample in the file may be incomplete, causing <command>hp2ps</command> to fail. If
1211 you are using a machine with UNIX utilities installed, it's not too
1212 hard to work around this problem (though the resulting command line
1213 looks rather Byzantine):
1215 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1219 The command <command>fgrep -n END_SAMPLE FOO.hp</command> finds the
1220 end of every complete sample in <filename>FOO.hp</filename>, and labels each sample with
1221 its ending line number. We then select the line number of the last
1222 complete sample using <command>tail</command> and <command>cut</command>. This is used as a
1223 parameter to <command>head</command>; the result is as if we deleted the final
1224 incomplete sample from <filename>FOO.hp</filename>. This results in a properly-formatted
1225 .hp file which we feed directly to <command>hp2ps</command>.
1229 <title>Viewing a heap profile in real time</title>
1232 The <command>gv</command> and <command>ghostview</command> programs
1233 have a "watch file" option can be used to view an up-to-date heap
1234 profile of your program as it runs. Simply generate an incremental
1235 heap profile as described in the previous section. Run <command>gv</command> on your
1238 gv -watch -seascape FOO.ps
1240 If you forget the <literal>-watch</literal> flag you can still select
1241 "Watch file" from the "State" menu. Now each time you generate a new
1242 profile <filename>FOO.ps</filename> the view will update automatically.
1246 This can all be encapsulated in a little script:
1249 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1251 gv -watch -seascape FOO.ps &
1253 sleep 10 # We generate a new profile every 10 seconds.
1254 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1258 Occasionally <command>gv</command> will choke as it tries to read an incomplete copy of
1259 <filename>FOO.ps</filename> (because <command>hp2ps</command> is still running as an update
1260 occurs). A slightly more complicated script works around this
1261 problem, by using the fact that sending a SIGHUP to gv will cause it
1262 to re-read its input file:
1265 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1271 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1281 <title>Observing Code Coverage</title>
1282 <indexterm><primary>code coverage</primary></indexterm>
1283 <indexterm><primary>Haskell Program Coverage</primary></indexterm>
1284 <indexterm><primary>hpc</primary></indexterm>
1287 Code coverage tools allow a programmer to determine what parts of
1288 their code have been actually executed, and which parts have
1289 never actually been invoked. GHC has an option for generating
1290 instrumented code that records code coverage as part of the
1291 <ulink url="http://www.haskell.org/hpc">Haskell Program Coverage
1292 </ulink>(HPC) toolkit, which is included with GHC. HPC tools can
1293 be used to render the generated code coverage information into
1294 human understandable format. </para>
1297 Correctly instrumented code provides coverage information of two
1298 kinds: source coverage and boolean-control coverage. Source
1299 coverage is the extent to which every part of the program was
1300 used, measured at three different levels: declarations (both
1301 top-level and local), alternatives (among several equations or
1302 case branches) and expressions (at every level). Boolean
1303 coverage is the extent to which each of the values True and
1304 False is obtained in every syntactic boolean context (ie. guard,
1305 condition, qualifier). </para>
1308 HPC displays both kinds of information in two primary ways:
1309 textual reports with summary statistics (hpc report) and sources
1310 with color mark-up (hpc markup). For boolean coverage, there
1311 are four possible outcomes for each guard, condition or
1312 qualifier: both True and False values occur; only True; only
1313 False; never evaluated. In hpc-markup output, highlighting with
1314 a yellow background indicates a part of the program that was
1315 never evaluated; a green background indicates an always-True
1316 expression and a red background indicates an always-False one.
1319 <sect2><title>A small example: Reciprocation</title>
1322 For an example we have a program, called Recip.hs, which computes exact decimal
1323 representations of reciprocals, with recurring parts indicated in
1327 reciprocal :: Int -> (String, Int)
1328 reciprocal n | n > 1 = ('0' : '.' : digits, recur)
1330 "attempting to compute reciprocal of number <= 1"
1332 (digits, recur) = divide n 1 []
1333 divide :: Int -> Int -> [Int] -> (String, Int)
1334 divide n c cs | c `elem` cs = ([], position c cs)
1335 | r == 0 = (show q, 0)
1336 | r /= 0 = (show q ++ digits, recur)
1338 (q, r) = (c*10) `quotRem` n
1339 (digits, recur) = divide n r (c:cs)
1341 position :: Int -> [Int] -> Int
1342 position n (x:xs) | n==x = 1
1343 | otherwise = 1 + position n xs
1345 showRecip :: Int -> String
1347 "1/" ++ show n ++ " = " ++
1348 if r==0 then d else take p d ++ "(" ++ drop p d ++ ")"
1351 (d, r) = reciprocal n
1355 putStrLn (showRecip number)
1359 <para>The HPC instrumentation is enabled using the -fhpc flag.
1363 $ ghc -fhpc Recip.hs --make
1365 <para>HPC index (.mix) files are placed placed in .hpc subdirectory. These can be considered like
1366 the .hi files for HPC.
1373 <para>We can generate a textual summary of coverage:</para>
1376 80% expressions used (81/101)
1377 12% boolean coverage (1/8)
1378 14% guards (1/7), 3 always True,
1381 0% 'if' conditions (0/1), 1 always False
1382 100% qualifiers (0/0)
1383 55% alternatives used (5/9)
1384 100% local declarations used (9/9)
1385 100% top-level declarations used (5/5)
1387 <para>We can also generate a marked-up version of the source.</para>
1390 writing Recip.hs.html
1393 This generates one file per Haskell module, and 4 index files,
1394 hpc_index.html, hpc_index_alt.html, hpc_index_exp.html,
1399 <sect2><title>Options for instrumenting code for coverage</title>
1401 Turning on code coverage is easy, use the -fhpc flag.
1402 Instrumented and non-instrumented can be freely mixed.
1403 When compiling the Main module GHC automatically detects when there
1404 is an hpc compiled file, and adds the correct initialization code.
1409 <sect2><title>The hpc toolkit</title>
1412 The hpc toolkit uses a cvs/svn/darcs-like interface, where a
1413 single binary contains many function units.</para>
1416 Usage: hpc COMMAND ...
1419 help Display help for hpc or a single command
1421 report Output textual report about program coverage
1422 markup Markup Haskell source with program coverage
1423 Processing Coverage files:
1424 sum Sum multiple .tix files in a single .tix file
1425 combine Combine two .tix files in a single .tix file
1426 map Map a function over a single .tix file
1428 overlay Generate a .tix file from an overlay file
1429 draft Generate draft overlay that provides 100% coverage
1431 show Show .tix file in readable, verbose format
1432 version Display version for hpc
1435 <para>In general, these options act on .tix file after an
1436 instrumented binary has generated it, which hpc acting as a
1437 conduit between the raw .tix file, and the more detailed reports
1442 The hpc tool assumes you are in the top-level directory of
1443 the location where you built your application, and the .tix
1444 file is in the same top-level directory. You can use the
1445 flag --srcdir to use hpc for any other directory, and use
1446 --srcdir multiple times to analyse programs compiled from
1447 difference locations, as is typical for packages.
1451 We now explain in more details the major modes of hpc.
1454 <sect3><title>hpc report</title>
1455 <para>hpc report gives a textual report of coverage. By default,
1456 all modules and packages are considered in generating report,
1457 unless include or exclude are used. The report is a summary
1458 unless the --per-module flag is used. The --xml-output option
1459 allows for tools to use hpc to glean coverage.
1463 Usage: hpc report [OPTION] .. <TIX_FILE> [<MODULE> [<MODULE> ..]]
1467 --per-module show module level detail
1468 --decl-list show unused decls
1469 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1470 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1471 --srcdir=DIR path to source directory of .hs files
1472 multi-use of srcdir possible
1473 --hpcdir=DIR sub-directory that contains .mix files
1474 default .hpc [rarely used]
1475 --xml-output show output in XML
1478 <sect3><title>hpc markup</title>
1479 <para>hpc markup marks up source files into colored html.
1483 Usage: hpc markup [OPTION] .. <TIX_FILE> [<MODULE> [<MODULE> ..]]
1487 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1488 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1489 --srcdir=DIR path to source directory of .hs files
1490 multi-use of srcdir possible
1491 --hpcdir=DIR sub-directory that contains .mix files
1492 default .hpc [rarely used]
1493 --fun-entry-count show top-level function entry counts
1494 --highlight-covered highlight covered code, rather that code gaps
1495 --destdir=DIR path to write output to
1499 <sect3><title>hpc sum</title>
1500 <para>hpc sum adds together any number of .tix files into a single
1501 .tix file. hpc sum does not change the original .tix file; it generates a new .tix file.
1505 Usage: hpc sum [OPTION] .. <TIX_FILE> [<TIX_FILE> [<TIX_FILE> ..]]
1506 Sum multiple .tix files in a single .tix file
1510 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1511 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1512 --output=FILE output FILE
1513 --union use the union of the module namespace (default is intersection)
1516 <sect3><title>hpc combine</title>
1517 <para>hpc combine is the swiss army knife of hpc. It can be
1518 used to take the difference between .tix files, to subtract one
1519 .tix file from another, or to add two .tix files. hpc combine does not
1520 change the original .tix file; it generates a new .tix file.
1524 Usage: hpc combine [OPTION] .. <TIX_FILE> <TIX_FILE>
1525 Combine two .tix files in a single .tix file
1529 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1530 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1531 --output=FILE output FILE
1532 --function=FUNCTION combine .tix files with join function, default = ADD
1533 FUNCTION = ADD | DIFF | SUB
1534 --union use the union of the module namespace (default is intersection)
1537 <sect3><title>hpc map</title>
1538 <para>hpc map inverts or zeros a .tix file. hpc map does not
1539 change the original .tix file; it generates a new .tix file.
1543 Usage: hpc map [OPTION] .. <TIX_FILE>
1544 Map a function over a single .tix file
1548 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1549 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1550 --output=FILE output FILE
1551 --function=FUNCTION apply function to .tix files, default = ID
1552 FUNCTION = ID | INV | ZERO
1553 --union use the union of the module namespace (default is intersection)
1556 <sect3><title>hpc overlay and hpc draft</title>
1558 Overlays are an experimental feature of HPC, a textual description
1559 of coverage. hpc draft is used to generate a draft overlay from a .tix file,
1560 and hpc overlay generates a .tix files from an overlay.
1564 Usage: hpc overlay [OPTION] .. <OVERLAY_FILE> [<OVERLAY_FILE> [...]]
1568 --srcdir=DIR path to source directory of .hs files
1569 multi-use of srcdir possible
1570 --hpcdir=DIR sub-directory that contains .mix files
1571 default .hpc [rarely used]
1572 --output=FILE output FILE
1574 Usage: hpc draft [OPTION] .. <TIX_FILE>
1578 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1579 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1580 --srcdir=DIR path to source directory of .hs files
1581 multi-use of srcdir possible
1582 --hpcdir=DIR sub-directory that contains .mix files
1583 default .hpc [rarely used]
1584 --output=FILE output FILE
1588 <sect2><title>Caveats and Shortcomings of Haskell Program Coverage</title>
1590 HPC does not attempt to lock the .tix file, so multiple concurrently running
1591 binaries in the same directory will exhibit a race condition. There is no way
1592 to change the name of the .tix file generated, apart from renaming the binary.
1593 HPC does not work with GHCi.
1598 <sect1 id="ticky-ticky">
1599 <title>Using “ticky-ticky” profiling (for implementors)</title>
1600 <indexterm><primary>ticky-ticky profiling</primary></indexterm>
1602 <para>(ToDo: document properly.)</para>
1604 <para>It is possible to compile Haskell programs so that
1605 they will count lots and lots of interesting things, e.g., number
1606 of updates, number of data constructors entered, etc., etc. We
1607 call this “ticky-ticky”
1608 profiling,<indexterm><primary>ticky-ticky
1609 profiling</primary></indexterm> <indexterm><primary>profiling,
1610 ticky-ticky</primary></indexterm> because that's the sound a CPU
1611 makes when it is running up all those counters
1612 (<emphasis>slowly</emphasis>).</para>
1614 <para>Ticky-ticky profiling is mainly intended for implementors;
1615 it is quite separate from the main “cost-centre”
1616 profiling system, intended for all users everywhere.</para>
1619 You don't need to build GHC, the libraries, or the RTS a special
1620 way in order to use ticky-ticky profiling. You can decide on a
1621 module-by-module basis which parts of a program have the
1622 counters compiled in, using the
1623 compile-time <option>-ticky</option> option. Those modules that
1624 were not compiled with <option>-ticky</option> won't contribute
1625 to the ticky-ticky profiling results, and that will normally
1626 include all the pre-compiled packages that your program links
1631 To get your compiled program to spit out the ticky-ticky
1637 Link the program with <option>-debug</option>
1638 (<option>-ticky</option> is a synonym
1639 for <option>-debug</option> at link-time). This links in
1640 the debug version of the RTS, which includes the code for
1641 aggregating and reporting the results of ticky-ticky
1647 Run the program with the <option>-r</option> RTS
1648 option<indexterm><primary>-r RTS option</primary></indexterm>.
1649 See <xref linkend="runtime-control"/>.
1656 Here is a sample ticky-ticky statistics file, generated by
1658 <command>foo +RTS -rfoo.ticky</command>.
1662 foo +RTS -rfoo.ticky
1664 ALLOCATIONS: 3964631 (11330900 words total: 3999476 admin, 6098829 goods, 1232595 slop)
1665 total words: 2 3 4 5 6+
1666 69647 ( 1.8%) function values 50.0 50.0 0.0 0.0 0.0
1667 2382937 ( 60.1%) thunks 0.0 83.9 16.1 0.0 0.0
1668 1477218 ( 37.3%) data values 66.8 33.2 0.0 0.0 0.0
1669 0 ( 0.0%) big tuples
1670 2 ( 0.0%) black holes 0.0 100.0 0.0 0.0 0.0
1671 0 ( 0.0%) prim things
1672 34825 ( 0.9%) partial applications 0.0 0.0 0.0 100.0 0.0
1673 2 ( 0.0%) thread state objects 0.0 0.0 0.0 0.0 100.0
1675 Total storage-manager allocations: 3647137 (11882004 words)
1676 [551104 words lost to speculative heap-checks]
1680 ENTERS: 9400092 of which 2005772 (21.3%) direct to the entry code
1681 [the rest indirected via Node's info ptr]
1682 1860318 ( 19.8%) thunks
1683 3733184 ( 39.7%) data values
1684 3149544 ( 33.5%) function values
1685 [of which 1999880 (63.5%) bypassed arg-satisfaction chk]
1686 348140 ( 3.7%) partial applications
1687 308906 ( 3.3%) normal indirections
1688 0 ( 0.0%) permanent indirections
1691 2137257 ( 36.4%) from entering a new constructor
1692 [the rest from entering an existing constructor]
1693 2349219 ( 40.0%) vectored [the rest unvectored]
1695 RET_NEW: 2137257: 32.5% 46.2% 21.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1696 RET_OLD: 3733184: 2.8% 67.9% 29.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1697 RET_UNBOXED_TUP: 2: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1699 RET_VEC_RETURN : 2349219: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1701 UPDATE FRAMES: 2241725 (0 omitted from thunks)
1705 0 ( 0.0%) data values
1706 34827 ( 1.6%) partial applications
1707 [2 in place, 34825 allocated new space]
1708 2206898 ( 98.4%) updates to existing heap objects (46 by squeezing)
1709 UPD_CON_IN_NEW: 0: 0 0 0 0 0 0 0 0 0
1710 UPD_PAP_IN_NEW: 34825: 0 0 0 34825 0 0 0 0 0
1712 NEW GEN UPDATES: 2274700 ( 99.9%)
1714 OLD GEN UPDATES: 1852 ( 0.1%)
1716 Total bytes copied during GC: 190096
1718 **************************************************
1719 3647137 ALLOC_HEAP_ctr
1720 11882004 ALLOC_HEAP_tot
1725 34831 ALLOC_FUN_hst_0
1726 34816 ALLOC_FUN_hst_1
1730 2382937 ALLOC_UP_THK_ctr
1733 0 E!NT_PERM_IND_ctr requires +RTS -Z
1734 [... lots more info omitted ...]
1735 0 GC_SEL_ABANDONED_ctr
1738 0 GC_FAILED_PROMOTION_ctr
1739 47524 GC_WORDS_COPIED_ctr
1742 <para>The formatting of the information above the row of asterisks
1743 is subject to change, but hopefully provides a useful
1744 human-readable summary. Below the asterisks <emphasis>all
1745 counters</emphasis> maintained by the ticky-ticky system are
1746 dumped, in a format intended to be machine-readable: zero or more
1747 spaces, an integer, a space, the counter name, and a newline.</para>
1749 <para>In fact, not <emphasis>all</emphasis> counters are
1750 necessarily dumped; compile- or run-time flags can render certain
1751 counters invalid. In this case, either the counter will simply
1752 not appear, or it will appear with a modified counter name,
1753 possibly along with an explanation for the omission (notice
1754 <literal>ENT_PERM_IND_ctr</literal> appears
1755 with an inserted <literal>!</literal> above). Software analysing
1756 this output should always check that it has the counters it
1757 expects. Also, beware: some of the counters can have
1758 <emphasis>large</emphasis> values!</para>
1765 ;;; Local Variables: ***
1767 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***