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 <imagedata fileref="prof_scc"/>
287 <sect2 id="prof-rules">
288 <title>Rules for attributing costs</title>
290 <para>The cost of evaluating any expression in your program is
291 attributed to a cost-centre stack using the following rules:</para>
295 <para>If the expression is part of the
296 <firstterm>one-off</firstterm> costs of evaluating the
297 enclosing top-level definition, then costs are attributed to
298 the stack of lexically enclosing <literal>SCC</literal>
299 annotations on top of the special <literal>CAF</literal>
304 <para>Otherwise, costs are attributed to the stack of
305 lexically-enclosing <literal>SCC</literal> annotations,
306 appended to the cost-centre stack in effect at the
307 <firstterm>call site</firstterm> of the current top-level
308 definition<footnote> <para>The call-site is just the place
309 in the source code which mentions the particular function or
310 variable.</para></footnote>. Notice that this is a recursive
315 <para>Time spent in foreign code (see <xref linkend="ffi"/>)
316 is always attributed to the cost centre in force at the
317 Haskell call-site of the foreign function.</para>
321 <para>What do we mean by one-off costs? Well, Haskell is a lazy
322 language, and certain expressions are only ever evaluated once.
323 For example, if we write:</para>
329 <para>then <varname>x</varname> will only be evaluated once (if
330 at all), and subsequent demands for <varname>x</varname> will
331 immediately get to see the cached result. The definition
332 <varname>x</varname> is called a CAF (Constant Applicative
333 Form), because it has no arguments.</para>
335 <para>For the purposes of profiling, we say that the expression
336 <literal>nfib 25</literal> belongs to the one-off costs of
337 evaluating <varname>x</varname>.</para>
339 <para>Since one-off costs aren't strictly speaking part of the
340 call-graph of the program, they are attributed to a special
341 top-level cost centre, <literal>CAF</literal>. There may be one
342 <literal>CAF</literal> cost centre for each module (the
343 default), or one for each top-level definition with any one-off
344 costs (this behaviour can be selected by giving GHC the
345 <option>-caf-all</option> flag).</para>
347 <indexterm><primary><literal>-caf-all</literal></primary>
350 <para>If you think you have a weird profile, or the call-graph
351 doesn't look like you expect it to, feel free to send it (and
352 your program) to us at
353 <email>glasgow-haskell-bugs@haskell.org</email>.</para>
357 <sect1 id="prof-compiler-options">
358 <title>Compiler options for profiling</title>
360 <indexterm><primary>profiling</primary><secondary>options</secondary></indexterm>
361 <indexterm><primary>options</primary><secondary>for profiling</secondary></indexterm>
366 <option>-prof</option>:
367 <indexterm><primary><option>-prof</option></primary></indexterm>
370 <para> To make use of the profiling system
371 <emphasis>all</emphasis> modules must be compiled and linked
372 with the <option>-prof</option> option. Any
373 <literal>SCC</literal> annotations you've put in your source
374 will spring to life.</para>
376 <para> Without a <option>-prof</option> option, your
377 <literal>SCC</literal>s are ignored; so you can compile
378 <literal>SCC</literal>-laden code without changing
384 <para>There are a few other profiling-related compilation options.
385 Use them <emphasis>in addition to</emphasis>
386 <option>-prof</option>. These do not have to be used consistently
387 for all modules in a program.</para>
392 <option>-auto</option>:
393 <indexterm><primary><option>-auto</option></primary></indexterm>
394 <indexterm><primary>cost centres</primary><secondary>automatically inserting</secondary></indexterm>
397 <para> GHC will automatically add
398 <function>_scc_</function> constructs for all
399 top-level, exported functions.</para>
405 <option>-auto-all</option>:
406 <indexterm><primary><option>-auto-all</option></primary></indexterm>
409 <para> <emphasis>All</emphasis> top-level functions,
410 exported or not, will be automatically
411 <function>_scc_</function>'d.</para>
417 <option>-caf-all</option>:
418 <indexterm><primary><option>-caf-all</option></primary></indexterm>
421 <para> The costs of all CAFs in a module are usually
422 attributed to one “big” CAF cost-centre. With
423 this option, all CAFs get their own cost-centre. An
424 “if all else fails” option…</para>
430 <option>-ignore-scc</option>:
431 <indexterm><primary><option>-ignore-scc</option></primary></indexterm>
434 <para>Ignore any <function>_scc_</function>
435 constructs, so a module which already has
436 <function>_scc_</function>s can be compiled
437 for profiling with the annotations ignored.</para>
445 <sect1 id="prof-time-options">
446 <title>Time and allocation profiling</title>
448 <para>To generate a time and allocation profile, give one of the
449 following RTS options to the compiled program when you run it (RTS
450 options should be enclosed between <literal>+RTS...-RTS</literal>
456 <option>-p</option> or <option>-P</option>:
457 <indexterm><primary><option>-p</option></primary></indexterm>
458 <indexterm><primary><option>-P</option></primary></indexterm>
459 <indexterm><primary>time profile</primary></indexterm>
462 <para>The <option>-p</option> option produces a standard
463 <emphasis>time profile</emphasis> report. It is written
465 <filename><replaceable>program</replaceable>.prof</filename>.</para>
467 <para>The <option>-P</option> option produces a more
468 detailed report containing the actual time and allocation
469 data as well. (Not used much.)</para>
476 <indexterm><primary><option>-xc</option></primary><secondary>RTS option</secondary></indexterm>
479 <para>This option makes use of the extra information
480 maintained by the cost-centre-stack profiler to provide
481 useful information about the location of runtime errors.
482 See <xref linkend="rts-options-debugging"/>.</para>
490 <sect1 id="prof-heap">
491 <title>Profiling memory usage</title>
493 <para>In addition to profiling the time and allocation behaviour
494 of your program, you can also generate a graph of its memory usage
495 over time. This is useful for detecting the causes of
496 <firstterm>space leaks</firstterm>, when your program holds on to
497 more memory at run-time that it needs to. Space leaks lead to
498 longer run-times due to heavy garbage collector activity, and may
499 even cause the program to run out of memory altogether.</para>
501 <para>To generate a heap profile from your program:</para>
505 <para>Compile the program for profiling (<xref
506 linkend="prof-compiler-options"/>).</para>
509 <para>Run it with one of the heap profiling options described
510 below (eg. <option>-hc</option> for a basic producer profile).
511 This generates the file
512 <filename><replaceable>prog</replaceable>.hp</filename>.</para>
515 <para>Run <command>hp2ps</command> to produce a Postscript
517 <filename><replaceable>prog</replaceable>.ps</filename>. The
518 <command>hp2ps</command> utility is described in detail in
519 <xref linkend="hp2ps"/>.</para>
522 <para>Display the heap profile using a postscript viewer such
523 as <application>Ghostview</application>, or print it out on a
524 Postscript-capable printer.</para>
528 <para>You might also want to take a look
529 at <ulink url="http://www.haskell.org/haskellwiki/Hp2any">hp2any</ulink>,
530 a more advanced suite of tools (not distributed with GHC) for
531 displaying heap profiles.</para>
533 <sect2 id="rts-options-heap-prof">
534 <title>RTS options for heap profiling</title>
536 <para>There are several different kinds of heap profile that can
537 be generated. All the different profile types yield a graph of
538 live heap against time, but they differ in how the live heap is
539 broken down into bands. The following RTS options select which
540 break-down to use:</para>
546 <indexterm><primary><option>-hc</option></primary><secondary>RTS option</secondary></indexterm>
549 <para>Breaks down the graph by the cost-centre stack which
550 produced the data.</para>
557 <indexterm><primary><option>-hm</option></primary><secondary>RTS option</secondary></indexterm>
560 <para>Break down the live heap by the module containing
561 the code which produced the data.</para>
568 <indexterm><primary><option>-hd</option></primary><secondary>RTS option</secondary></indexterm>
571 <para>Breaks down the graph by <firstterm>closure
572 description</firstterm>. For actual data, the description
573 is just the constructor name, for other closures it is a
574 compiler-generated string identifying the closure.</para>
581 <indexterm><primary><option>-hy</option></primary><secondary>RTS option</secondary></indexterm>
584 <para>Breaks down the graph by
585 <firstterm>type</firstterm>. For closures which have
586 function type or unknown/polymorphic type, the string will
587 represent an approximation to the actual type.</para>
594 <indexterm><primary><option>-hr</option></primary><secondary>RTS option</secondary></indexterm>
597 <para>Break down the graph by <firstterm>retainer
598 set</firstterm>. Retainer profiling is described in more
599 detail below (<xref linkend="retainer-prof"/>).</para>
606 <indexterm><primary><option>-hb</option></primary><secondary>RTS option</secondary></indexterm>
609 <para>Break down the graph by
610 <firstterm>biography</firstterm>. Biographical profiling
611 is described in more detail below (<xref
612 linkend="biography-prof"/>).</para>
617 <para>In addition, the profile can be restricted to heap data
618 which satisfies certain criteria - for example, you might want
619 to display a profile by type but only for data produced by a
620 certain module, or a profile by retainer for a certain type of
621 data. Restrictions are specified as follows:</para>
626 <option>-hc</option><replaceable>name</replaceable>,...
627 <indexterm><primary><option>-hc</option></primary><secondary>RTS option</secondary></indexterm>
630 <para>Restrict the profile to closures produced by
631 cost-centre stacks with one of the specified cost centres
638 <option>-hC</option><replaceable>name</replaceable>,...
639 <indexterm><primary><option>-hC</option></primary><secondary>RTS option</secondary></indexterm>
642 <para>Restrict the profile to closures produced by
643 cost-centre stacks with one of the specified cost centres
644 anywhere in the stack.</para>
650 <option>-hm</option><replaceable>module</replaceable>,...
651 <indexterm><primary><option>-hm</option></primary><secondary>RTS option</secondary></indexterm>
654 <para>Restrict the profile to closures produced by the
655 specified modules.</para>
661 <option>-hd</option><replaceable>desc</replaceable>,...
662 <indexterm><primary><option>-hd</option></primary><secondary>RTS option</secondary></indexterm>
665 <para>Restrict the profile to closures with the specified
666 description strings.</para>
672 <option>-hy</option><replaceable>type</replaceable>,...
673 <indexterm><primary><option>-hy</option></primary><secondary>RTS option</secondary></indexterm>
676 <para>Restrict the profile to closures with the specified
683 <option>-hr</option><replaceable>cc</replaceable>,...
684 <indexterm><primary><option>-hr</option></primary><secondary>RTS option</secondary></indexterm>
687 <para>Restrict the profile to closures with retainer sets
688 containing cost-centre stacks with one of the specified
689 cost centres at the top.</para>
695 <option>-hb</option><replaceable>bio</replaceable>,...
696 <indexterm><primary><option>-hb</option></primary><secondary>RTS option</secondary></indexterm>
699 <para>Restrict the profile to closures with one of the
700 specified biographies, where
701 <replaceable>bio</replaceable> is one of
702 <literal>lag</literal>, <literal>drag</literal>,
703 <literal>void</literal>, or <literal>use</literal>.</para>
708 <para>For example, the following options will generate a
709 retainer profile restricted to <literal>Branch</literal> and
710 <literal>Leaf</literal> constructors:</para>
713 <replaceable>prog</replaceable> +RTS -hr -hdBranch,Leaf
716 <para>There can only be one "break-down" option
717 (eg. <option>-hr</option> in the example above), but there is no
718 limit on the number of further restrictions that may be applied.
719 All the options may be combined, with one exception: GHC doesn't
720 currently support mixing the <option>-hr</option> and
721 <option>-hb</option> options.</para>
723 <para>There are three more options which relate to heap
729 <option>-i<replaceable>secs</replaceable></option>:
730 <indexterm><primary><option>-i</option></primary></indexterm>
733 <para>Set the profiling (sampling) interval to
734 <replaceable>secs</replaceable> seconds (the default is
735 0.1 second). Fractions are allowed: for example
736 <option>-i0.2</option> will get 5 samples per second.
737 This only affects heap profiling; time profiles are always
738 sampled on a 1/50 second frequency.</para>
745 <indexterm><primary><option>-xt</option></primary><secondary>RTS option</secondary></indexterm>
748 <para>Include the memory occupied by threads in a heap
749 profile. Each thread takes up a small area for its thread
750 state in addition to the space allocated for its stack
751 (stacks normally start small and then grow as
754 <para>This includes the main thread, so using
755 <option>-xt</option> is a good way to see how much stack
756 space the program is using.</para>
758 <para>Memory occupied by threads and their stacks is
759 labelled as “TSO” when displaying the profile
760 by closure description or type description.</para>
766 <option>-L<replaceable>num</replaceable></option>
767 <indexterm><primary><option>-L</option></primary><secondary>RTS option</secondary></indexterm>
771 Sets the maximum length of a cost-centre stack name in a
772 heap profile. Defaults to 25.
780 <sect2 id="retainer-prof">
781 <title>Retainer Profiling</title>
783 <para>Retainer profiling is designed to help answer questions
784 like <quote>why is this data being retained?</quote>. We start
785 by defining what we mean by a retainer:</para>
788 <para>A retainer is either the system stack, or an unevaluated
789 closure (thunk).</para>
792 <para>In particular, constructors are <emphasis>not</emphasis>
795 <para>An object B retains object A if (i) B is a retainer object and
796 (ii) object A can be reached by recursively following pointers
797 starting from object B, but not meeting any other retainer
798 objects on the way. Each live object is retained by one or more
799 retainer objects, collectively called its retainer set, or its
800 <firstterm>retainer set</firstterm>, or its
801 <firstterm>retainers</firstterm>.</para>
803 <para>When retainer profiling is requested by giving the program
804 the <option>-hr</option> option, a graph is generated which is
805 broken down by retainer set. A retainer set is displayed as a
806 set of cost-centre stacks; because this is usually too large to
807 fit on the profile graph, each retainer set is numbered and
808 shown abbreviated on the graph along with its number, and the
809 full list of retainer sets is dumped into the file
810 <filename><replaceable>prog</replaceable>.prof</filename>.</para>
812 <para>Retainer profiling requires multiple passes over the live
813 heap in order to discover the full retainer set for each
814 object, which can be quite slow. So we set a limit on the
815 maximum size of a retainer set, where all retainer sets larger
816 than the maximum retainer set size are replaced by the special
817 set <literal>MANY</literal>. The maximum set size defaults to 8
818 and can be altered with the <option>-R</option> RTS
823 <term><option>-R</option><replaceable>size</replaceable></term>
825 <para>Restrict the number of elements in a retainer set to
826 <replaceable>size</replaceable> (default 8).</para>
832 <title>Hints for using retainer profiling</title>
834 <para>The definition of retainers is designed to reflect a
835 common cause of space leaks: a large structure is retained by
836 an unevaluated computation, and will be released once the
837 computation is forced. A good example is looking up a value in
838 a finite map, where unless the lookup is forced in a timely
839 manner the unevaluated lookup will cause the whole mapping to
840 be retained. These kind of space leaks can often be
841 eliminated by forcing the relevant computations to be
842 performed eagerly, using <literal>seq</literal> or strictness
843 annotations on data constructor fields.</para>
845 <para>Often a particular data structure is being retained by a
846 chain of unevaluated closures, only the nearest of which will
847 be reported by retainer profiling - for example A retains B, B
848 retains C, and C retains a large structure. There might be a
849 large number of Bs but only a single A, so A is really the one
850 we're interested in eliminating. However, retainer profiling
851 will in this case report B as the retainer of the large
852 structure. To move further up the chain of retainers, we can
853 ask for another retainer profile but this time restrict the
854 profile to B objects, so we get a profile of the retainers of
858 <replaceable>prog</replaceable> +RTS -hr -hcB
861 <para>This trick isn't foolproof, because there might be other
862 B closures in the heap which aren't the retainers we are
863 interested in, but we've found this to be a useful technique
864 in most cases.</para>
868 <sect2 id="biography-prof">
869 <title>Biographical Profiling</title>
871 <para>A typical heap object may be in one of the following four
872 states at each point in its lifetime:</para>
876 <para>The <firstterm>lag</firstterm> stage, which is the
877 time between creation and the first use of the
881 <para>the <firstterm>use</firstterm> stage, which lasts from
882 the first use until the last use of the object, and</para>
885 <para>The <firstterm>drag</firstterm> stage, which lasts
886 from the final use until the last reference to the object
890 <para>An object which is never used is said to be in the
891 <firstterm>void</firstterm> state for its whole
896 <para>A biographical heap profile displays the portion of the
897 live heap in each of the four states listed above. Usually the
898 most interesting states are the void and drag states: live heap
899 in these states is more likely to be wasted space than heap in
900 the lag or use states.</para>
902 <para>It is also possible to break down the heap in one or more
903 of these states by a different criteria, by restricting a
904 profile by biography. For example, to show the portion of the
905 heap in the drag or void state by producer: </para>
908 <replaceable>prog</replaceable> +RTS -hc -hbdrag,void
911 <para>Once you know the producer or the type of the heap in the
912 drag or void states, the next step is usually to find the
916 <replaceable>prog</replaceable> +RTS -hr -hc<replaceable>cc</replaceable>...
919 <para>NOTE: this two stage process is required because GHC
920 cannot currently profile using both biographical and retainer
921 information simultaneously.</para>
924 <sect2 id="mem-residency">
925 <title>Actual memory residency</title>
927 <para>How does the heap residency reported by the heap profiler relate to
928 the actual memory residency of your program when you run it? You might
929 see a large discrepancy between the residency reported by the heap
930 profiler, and the residency reported by tools on your system
931 (eg. <literal>ps</literal> or <literal>top</literal> on Unix, or the
932 Task Manager on Windows). There are several reasons for this:</para>
936 <para>There is an overhead of profiling itself, which is subtracted
937 from the residency figures by the profiler. This overhead goes
938 away when compiling without profiling support, of course. The
939 space overhead is currently 2 extra
940 words per heap object, which probably results in
941 about a 30% overhead.</para>
945 <para>Garbage collection requires more memory than the actual
946 residency. The factor depends on the kind of garbage collection
947 algorithm in use: a major GC in the standard
948 generation copying collector will usually require 3L bytes of
949 memory, where L is the amount of live data. This is because by
950 default (see the <option>+RTS -F</option> option) we allow the old
951 generation to grow to twice its size (2L) before collecting it, and
952 we require additionally L bytes to copy the live data into. When
953 using compacting collection (see the <option>+RTS -c</option>
954 option), this is reduced to 2L, and can further be reduced by
955 tweaking the <option>-F</option> option. Also add the size of the
956 allocation area (currently a fixed 512Kb).</para>
960 <para>The stack isn't counted in the heap profile by default. See the
961 <option>+RTS -xt</option> option.</para>
965 <para>The program text itself, the C stack, any non-heap data (eg. data
966 allocated by foreign libraries, and data allocated by the RTS), and
967 <literal>mmap()</literal>'d memory are not counted in the heap profile.</para>
975 <title><command>hp2ps</command>––heap profile to PostScript</title>
977 <indexterm><primary><command>hp2ps</command></primary></indexterm>
978 <indexterm><primary>heap profiles</primary></indexterm>
979 <indexterm><primary>postscript, from heap profiles</primary></indexterm>
980 <indexterm><primary><option>-h<break-down></option></primary></indexterm>
985 hp2ps [flags] [<file>[.hp]]
989 <command>hp2ps</command><indexterm><primary>hp2ps
990 program</primary></indexterm> converts a heap profile as produced
991 by the <option>-h<break-down></option> runtime option into a
992 PostScript graph of the heap profile. By convention, the file to
993 be processed by <command>hp2ps</command> has a
994 <filename>.hp</filename> extension. The PostScript output is
995 written to <filename><file>@.ps</filename>. If
996 <filename><file></filename> is omitted entirely, then the
997 program behaves as a filter.</para>
999 <para><command>hp2ps</command> is distributed in
1000 <filename>ghc/utils/hp2ps</filename> in a GHC source
1001 distribution. It was originally developed by Dave Wakeling as part
1002 of the HBC/LML heap profiler.</para>
1004 <para>The flags are:</para>
1009 <term><option>-d</option></term>
1011 <para>In order to make graphs more readable,
1012 <command>hp2ps</command> sorts the shaded bands for each
1013 identifier. The default sort ordering is for the bands with
1014 the largest area to be stacked on top of the smaller ones.
1015 The <option>-d</option> option causes rougher bands (those
1016 representing series of values with the largest standard
1017 deviations) to be stacked on top of smoother ones.</para>
1022 <term><option>-b</option></term>
1024 <para>Normally, <command>hp2ps</command> puts the title of
1025 the graph in a small box at the top of the page. However, if
1026 the JOB string is too long to fit in a small box (more than
1027 35 characters), then <command>hp2ps</command> will choose to
1028 use a big box instead. The <option>-b</option> option
1029 forces <command>hp2ps</command> to use a big box.</para>
1034 <term><option>-e<float>[in|mm|pt]</option></term>
1036 <para>Generate encapsulated PostScript suitable for
1037 inclusion in LaTeX documents. Usually, the PostScript graph
1038 is drawn in landscape mode in an area 9 inches wide by 6
1039 inches high, and <command>hp2ps</command> arranges for this
1040 area to be approximately centred on a sheet of a4 paper.
1041 This format is convenient of studying the graph in detail,
1042 but it is unsuitable for inclusion in LaTeX documents. The
1043 <option>-e</option> option causes the graph to be drawn in
1044 portrait mode, with float specifying the width in inches,
1045 millimetres or points (the default). The resulting
1046 PostScript file conforms to the Encapsulated PostScript
1047 (EPS) convention, and it can be included in a LaTeX document
1048 using Rokicki's dvi-to-PostScript converter
1049 <command>dvips</command>.</para>
1054 <term><option>-g</option></term>
1056 <para>Create output suitable for the <command>gs</command>
1057 PostScript previewer (or similar). In this case the graph is
1058 printed in portrait mode without scaling. The output is
1059 unsuitable for a laser printer.</para>
1064 <term><option>-l</option></term>
1066 <para>Normally a profile is limited to 20 bands with
1067 additional identifiers being grouped into an
1068 <literal>OTHER</literal> band. The <option>-l</option> flag
1069 removes this 20 band and limit, producing as many bands as
1070 necessary. No key is produced as it won't fit!. It is useful
1071 for creation time profiles with many bands.</para>
1076 <term><option>-m<int></option></term>
1078 <para>Normally a profile is limited to 20 bands with
1079 additional identifiers being grouped into an
1080 <literal>OTHER</literal> band. The <option>-m</option> flag
1081 specifies an alternative band limit (the maximum is
1084 <para><option>-m0</option> requests the band limit to be
1085 removed. As many bands as necessary are produced. However no
1086 key is produced as it won't fit! It is useful for displaying
1087 creation time profiles with many bands.</para>
1092 <term><option>-p</option></term>
1094 <para>Use previous parameters. By default, the PostScript
1095 graph is automatically scaled both horizontally and
1096 vertically so that it fills the page. However, when
1097 preparing a series of graphs for use in a presentation, it
1098 is often useful to draw a new graph using the same scale,
1099 shading and ordering as a previous one. The
1100 <option>-p</option> flag causes the graph to be drawn using
1101 the parameters determined by a previous run of
1102 <command>hp2ps</command> on <filename>file</filename>. These
1103 are extracted from <filename>file@.aux</filename>.</para>
1108 <term><option>-s</option></term>
1110 <para>Use a small box for the title.</para>
1115 <term><option>-t<float></option></term>
1117 <para>Normally trace elements which sum to a total of less
1118 than 1% of the profile are removed from the
1119 profile. The <option>-t</option> option allows this
1120 percentage to be modified (maximum 5%).</para>
1122 <para><option>-t0</option> requests no trace elements to be
1123 removed from the profile, ensuring that all the data will be
1129 <term><option>-c</option></term>
1131 <para>Generate colour output.</para>
1136 <term><option>-y</option></term>
1138 <para>Ignore marks.</para>
1143 <term><option>-?</option></term>
1145 <para>Print out usage information.</para>
1151 <sect2 id="manipulating-hp">
1152 <title>Manipulating the hp file</title>
1154 <para>(Notes kindly offered by Jan-Willhem Maessen.)</para>
1157 The <filename>FOO.hp</filename> file produced when you ask for the
1158 heap profile of a program <filename>FOO</filename> is a text file with a particularly
1159 simple structure. Here's a representative example, with much of the
1160 actual data omitted:
1163 DATE "Thu Dec 26 18:17 2002"
1164 SAMPLE_UNIT "seconds"
1175 BEGIN_SAMPLE 11695.47
1178 The first four lines (<literal>JOB</literal>, <literal>DATE</literal>, <literal>SAMPLE_UNIT</literal>, <literal>VALUE_UNIT</literal>) form a
1179 header. Each block of lines starting with <literal>BEGIN_SAMPLE</literal> and ending
1180 with <literal>END_SAMPLE</literal> forms a single sample (you can think of this as a
1181 vertical slice of your heap profile). The hp2ps utility should accept
1182 any input with a properly-formatted header followed by a series of
1188 <title>Zooming in on regions of your profile</title>
1191 You can look at particular regions of your profile simply by loading a
1192 copy of the <filename>.hp</filename> file into a text editor and deleting the unwanted
1193 samples. The resulting <filename>.hp</filename> file can be run through <command>hp2ps</command> and viewed
1199 <title>Viewing the heap profile of a running program</title>
1202 The <filename>.hp</filename> file is generated incrementally as your
1203 program runs. In principle, running <command>hp2ps</command> on the incomplete file
1204 should produce a snapshot of your program's heap usage. However, the
1205 last sample in the file may be incomplete, causing <command>hp2ps</command> to fail. If
1206 you are using a machine with UNIX utilities installed, it's not too
1207 hard to work around this problem (though the resulting command line
1208 looks rather Byzantine):
1210 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1214 The command <command>fgrep -n END_SAMPLE FOO.hp</command> finds the
1215 end of every complete sample in <filename>FOO.hp</filename>, and labels each sample with
1216 its ending line number. We then select the line number of the last
1217 complete sample using <command>tail</command> and <command>cut</command>. This is used as a
1218 parameter to <command>head</command>; the result is as if we deleted the final
1219 incomplete sample from <filename>FOO.hp</filename>. This results in a properly-formatted
1220 .hp file which we feed directly to <command>hp2ps</command>.
1224 <title>Viewing a heap profile in real time</title>
1227 The <command>gv</command> and <command>ghostview</command> programs
1228 have a "watch file" option can be used to view an up-to-date heap
1229 profile of your program as it runs. Simply generate an incremental
1230 heap profile as described in the previous section. Run <command>gv</command> on your
1233 gv -watch -seascape FOO.ps
1235 If you forget the <literal>-watch</literal> flag you can still select
1236 "Watch file" from the "State" menu. Now each time you generate a new
1237 profile <filename>FOO.ps</filename> the view will update automatically.
1241 This can all be encapsulated in a little script:
1244 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1246 gv -watch -seascape FOO.ps &
1248 sleep 10 # We generate a new profile every 10 seconds.
1249 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1253 Occasionally <command>gv</command> will choke as it tries to read an incomplete copy of
1254 <filename>FOO.ps</filename> (because <command>hp2ps</command> is still running as an update
1255 occurs). A slightly more complicated script works around this
1256 problem, by using the fact that sending a SIGHUP to gv will cause it
1257 to re-read its input file:
1260 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1266 head -`fgrep -n END_SAMPLE FOO.hp | tail -1 | cut -d : -f 1` FOO.hp \
1276 <title>Observing Code Coverage</title>
1277 <indexterm><primary>code coverage</primary></indexterm>
1278 <indexterm><primary>Haskell Program Coverage</primary></indexterm>
1279 <indexterm><primary>hpc</primary></indexterm>
1282 Code coverage tools allow a programmer to determine what parts of
1283 their code have been actually executed, and which parts have
1284 never actually been invoked. GHC has an option for generating
1285 instrumented code that records code coverage as part of the
1286 <ulink url="http://www.haskell.org/hpc">Haskell Program Coverage
1287 </ulink>(HPC) toolkit, which is included with GHC. HPC tools can
1288 be used to render the generated code coverage information into
1289 human understandable format. </para>
1292 Correctly instrumented code provides coverage information of two
1293 kinds: source coverage and boolean-control coverage. Source
1294 coverage is the extent to which every part of the program was
1295 used, measured at three different levels: declarations (both
1296 top-level and local), alternatives (among several equations or
1297 case branches) and expressions (at every level). Boolean
1298 coverage is the extent to which each of the values True and
1299 False is obtained in every syntactic boolean context (ie. guard,
1300 condition, qualifier). </para>
1303 HPC displays both kinds of information in two primary ways:
1304 textual reports with summary statistics (hpc report) and sources
1305 with color mark-up (hpc markup). For boolean coverage, there
1306 are four possible outcomes for each guard, condition or
1307 qualifier: both True and False values occur; only True; only
1308 False; never evaluated. In hpc-markup output, highlighting with
1309 a yellow background indicates a part of the program that was
1310 never evaluated; a green background indicates an always-True
1311 expression and a red background indicates an always-False one.
1314 <sect2><title>A small example: Reciprocation</title>
1317 For an example we have a program, called Recip.hs, which computes exact decimal
1318 representations of reciprocals, with recurring parts indicated in
1322 reciprocal :: Int -> (String, Int)
1323 reciprocal n | n > 1 = ('0' : '.' : digits, recur)
1325 "attempting to compute reciprocal of number <= 1"
1327 (digits, recur) = divide n 1 []
1328 divide :: Int -> Int -> [Int] -> (String, Int)
1329 divide n c cs | c `elem` cs = ([], position c cs)
1330 | r == 0 = (show q, 0)
1331 | r /= 0 = (show q ++ digits, recur)
1333 (q, r) = (c*10) `quotRem` n
1334 (digits, recur) = divide n r (c:cs)
1336 position :: Int -> [Int] -> Int
1337 position n (x:xs) | n==x = 1
1338 | otherwise = 1 + position n xs
1340 showRecip :: Int -> String
1342 "1/" ++ show n ++ " = " ++
1343 if r==0 then d else take p d ++ "(" ++ drop p d ++ ")"
1346 (d, r) = reciprocal n
1350 putStrLn (showRecip number)
1354 <para>The HPC instrumentation is enabled using the -fhpc flag.
1358 $ ghc -fhpc Recip.hs --make
1360 <para>HPC index (.mix) files are placed placed in .hpc subdirectory. These can be considered like
1361 the .hi files for HPC.
1368 <para>We can generate a textual summary of coverage:</para>
1371 80% expressions used (81/101)
1372 12% boolean coverage (1/8)
1373 14% guards (1/7), 3 always True,
1376 0% 'if' conditions (0/1), 1 always False
1377 100% qualifiers (0/0)
1378 55% alternatives used (5/9)
1379 100% local declarations used (9/9)
1380 100% top-level declarations used (5/5)
1382 <para>We can also generate a marked-up version of the source.</para>
1385 writing Recip.hs.html
1388 This generates one file per Haskell module, and 4 index files,
1389 hpc_index.html, hpc_index_alt.html, hpc_index_exp.html,
1394 <sect2><title>Options for instrumenting code for coverage</title>
1396 Turning on code coverage is easy, use the -fhpc flag.
1397 Instrumented and non-instrumented can be freely mixed.
1398 When compiling the Main module GHC automatically detects when there
1399 is an hpc compiled file, and adds the correct initialization code.
1404 <sect2><title>The hpc toolkit</title>
1407 The hpc toolkit uses a cvs/svn/darcs-like interface, where a
1408 single binary contains many function units.</para>
1411 Usage: hpc COMMAND ...
1414 help Display help for hpc or a single command
1416 report Output textual report about program coverage
1417 markup Markup Haskell source with program coverage
1418 Processing Coverage files:
1419 sum Sum multiple .tix files in a single .tix file
1420 combine Combine two .tix files in a single .tix file
1421 map Map a function over a single .tix file
1423 overlay Generate a .tix file from an overlay file
1424 draft Generate draft overlay that provides 100% coverage
1426 show Show .tix file in readable, verbose format
1427 version Display version for hpc
1430 <para>In general, these options act on .tix file after an
1431 instrumented binary has generated it, which hpc acting as a
1432 conduit between the raw .tix file, and the more detailed reports
1437 The hpc tool assumes you are in the top-level directory of
1438 the location where you built your application, and the .tix
1439 file is in the same top-level directory. You can use the
1440 flag --srcdir to use hpc for any other directory, and use
1441 --srcdir multiple times to analyse programs compiled from
1442 difference locations, as is typical for packages.
1446 We now explain in more details the major modes of hpc.
1449 <sect3><title>hpc report</title>
1450 <para>hpc report gives a textual report of coverage. By default,
1451 all modules and packages are considered in generating report,
1452 unless include or exclude are used. The report is a summary
1453 unless the --per-module flag is used. The --xml-output option
1454 allows for tools to use hpc to glean coverage.
1458 Usage: hpc report [OPTION] .. <TIX_FILE> [<MODULE> [<MODULE> ..]]
1462 --per-module show module level detail
1463 --decl-list show unused decls
1464 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1465 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1466 --srcdir=DIR path to source directory of .hs files
1467 multi-use of srcdir possible
1468 --hpcdir=DIR sub-directory that contains .mix files
1469 default .hpc [rarely used]
1470 --xml-output show output in XML
1473 <sect3><title>hpc markup</title>
1474 <para>hpc markup marks up source files into colored html.
1478 Usage: hpc markup [OPTION] .. <TIX_FILE> [<MODULE> [<MODULE> ..]]
1482 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1483 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1484 --srcdir=DIR path to source directory of .hs files
1485 multi-use of srcdir possible
1486 --hpcdir=DIR sub-directory that contains .mix files
1487 default .hpc [rarely used]
1488 --fun-entry-count show top-level function entry counts
1489 --highlight-covered highlight covered code, rather that code gaps
1490 --destdir=DIR path to write output to
1494 <sect3><title>hpc sum</title>
1495 <para>hpc sum adds together any number of .tix files into a single
1496 .tix file. hpc sum does not change the original .tix file; it generates a new .tix file.
1500 Usage: hpc sum [OPTION] .. <TIX_FILE> [<TIX_FILE> [<TIX_FILE> ..]]
1501 Sum multiple .tix files in a single .tix file
1505 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1506 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1507 --output=FILE output FILE
1508 --union use the union of the module namespace (default is intersection)
1511 <sect3><title>hpc combine</title>
1512 <para>hpc combine is the swiss army knife of hpc. It can be
1513 used to take the difference between .tix files, to subtract one
1514 .tix file from another, or to add two .tix files. hpc combine does not
1515 change the original .tix file; it generates a new .tix file.
1519 Usage: hpc combine [OPTION] .. <TIX_FILE> <TIX_FILE>
1520 Combine two .tix files in a single .tix file
1524 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1525 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1526 --output=FILE output FILE
1527 --function=FUNCTION combine .tix files with join function, default = ADD
1528 FUNCTION = ADD | DIFF | SUB
1529 --union use the union of the module namespace (default is intersection)
1532 <sect3><title>hpc map</title>
1533 <para>hpc map inverts or zeros a .tix file. hpc map does not
1534 change the original .tix file; it generates a new .tix file.
1538 Usage: hpc map [OPTION] .. <TIX_FILE>
1539 Map a function over a single .tix file
1543 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1544 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1545 --output=FILE output FILE
1546 --function=FUNCTION apply function to .tix files, default = ID
1547 FUNCTION = ID | INV | ZERO
1548 --union use the union of the module namespace (default is intersection)
1551 <sect3><title>hpc overlay and hpc draft</title>
1553 Overlays are an experimental feature of HPC, a textual description
1554 of coverage. hpc draft is used to generate a draft overlay from a .tix file,
1555 and hpc overlay generates a .tix files from an overlay.
1559 Usage: hpc overlay [OPTION] .. <OVERLAY_FILE> [<OVERLAY_FILE> [...]]
1563 --srcdir=DIR path to source directory of .hs files
1564 multi-use of srcdir possible
1565 --hpcdir=DIR sub-directory that contains .mix files
1566 default .hpc [rarely used]
1567 --output=FILE output FILE
1569 Usage: hpc draft [OPTION] .. <TIX_FILE>
1573 --exclude=[PACKAGE:][MODULE] exclude MODULE and/or PACKAGE
1574 --include=[PACKAGE:][MODULE] include MODULE and/or PACKAGE
1575 --srcdir=DIR path to source directory of .hs files
1576 multi-use of srcdir possible
1577 --hpcdir=DIR sub-directory that contains .mix files
1578 default .hpc [rarely used]
1579 --output=FILE output FILE
1583 <sect2><title>Caveats and Shortcomings of Haskell Program Coverage</title>
1585 HPC does not attempt to lock the .tix file, so multiple concurrently running
1586 binaries in the same directory will exhibit a race condition. There is no way
1587 to change the name of the .tix file generated, apart from renaming the binary.
1588 HPC does not work with GHCi.
1593 <sect1 id="ticky-ticky">
1594 <title>Using “ticky-ticky” profiling (for implementors)</title>
1595 <indexterm><primary>ticky-ticky profiling</primary></indexterm>
1597 <para>(ToDo: document properly.)</para>
1599 <para>It is possible to compile Glasgow Haskell programs so that
1600 they will count lots and lots of interesting things, e.g., number
1601 of updates, number of data constructors entered, etc., etc. We
1602 call this “ticky-ticky”
1603 profiling,<indexterm><primary>ticky-ticky
1604 profiling</primary></indexterm> <indexterm><primary>profiling,
1605 ticky-ticky</primary></indexterm> because that's the sound a Sun4
1606 makes when it is running up all those counters
1607 (<emphasis>slowly</emphasis>).</para>
1609 <para>Ticky-ticky profiling is mainly intended for implementors;
1610 it is quite separate from the main “cost-centre”
1611 profiling system, intended for all users everywhere.</para>
1613 <para>To be able to use ticky-ticky profiling, you will need to
1614 have built the ticky RTS. (This should be described in
1615 the building guide, but amounts to building the RTS with way
1616 "t" enabled.)</para>
1618 <para>To get your compiled program to spit out the ticky-ticky
1619 numbers, use a <option>-r</option> RTS
1620 option<indexterm><primary>-r RTS option</primary></indexterm>.
1621 See <xref linkend="runtime-control"/>.</para>
1623 <para>Compiling your program with the <option>-ticky</option>
1624 switch yields an executable that performs these counts. Here is a
1625 sample ticky-ticky statistics file, generated by the invocation
1626 <command>foo +RTS -rfoo.ticky</command>.</para>
1629 foo +RTS -rfoo.ticky
1632 ALLOCATIONS: 3964631 (11330900 words total: 3999476 admin, 6098829 goods, 1232595 slop)
1633 total words: 2 3 4 5 6+
1634 69647 ( 1.8%) function values 50.0 50.0 0.0 0.0 0.0
1635 2382937 ( 60.1%) thunks 0.0 83.9 16.1 0.0 0.0
1636 1477218 ( 37.3%) data values 66.8 33.2 0.0 0.0 0.0
1637 0 ( 0.0%) big tuples
1638 2 ( 0.0%) black holes 0.0 100.0 0.0 0.0 0.0
1639 0 ( 0.0%) prim things
1640 34825 ( 0.9%) partial applications 0.0 0.0 0.0 100.0 0.0
1641 2 ( 0.0%) thread state objects 0.0 0.0 0.0 0.0 100.0
1643 Total storage-manager allocations: 3647137 (11882004 words)
1644 [551104 words lost to speculative heap-checks]
1648 ENTERS: 9400092 of which 2005772 (21.3%) direct to the entry code
1649 [the rest indirected via Node's info ptr]
1650 1860318 ( 19.8%) thunks
1651 3733184 ( 39.7%) data values
1652 3149544 ( 33.5%) function values
1653 [of which 1999880 (63.5%) bypassed arg-satisfaction chk]
1654 348140 ( 3.7%) partial applications
1655 308906 ( 3.3%) normal indirections
1656 0 ( 0.0%) permanent indirections
1659 2137257 ( 36.4%) from entering a new constructor
1660 [the rest from entering an existing constructor]
1661 2349219 ( 40.0%) vectored [the rest unvectored]
1663 RET_NEW: 2137257: 32.5% 46.2% 21.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1664 RET_OLD: 3733184: 2.8% 67.9% 29.3% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1665 RET_UNBOXED_TUP: 2: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1667 RET_VEC_RETURN : 2349219: 0.0% 0.0%100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
1669 UPDATE FRAMES: 2241725 (0 omitted from thunks)
1673 0 ( 0.0%) data values
1674 34827 ( 1.6%) partial applications
1675 [2 in place, 34825 allocated new space]
1676 2206898 ( 98.4%) updates to existing heap objects (46 by squeezing)
1677 UPD_CON_IN_NEW: 0: 0 0 0 0 0 0 0 0 0
1678 UPD_PAP_IN_NEW: 34825: 0 0 0 34825 0 0 0 0 0
1680 NEW GEN UPDATES: 2274700 ( 99.9%)
1682 OLD GEN UPDATES: 1852 ( 0.1%)
1684 Total bytes copied during GC: 190096
1686 **************************************************
1687 3647137 ALLOC_HEAP_ctr
1688 11882004 ALLOC_HEAP_tot
1693 34831 ALLOC_FUN_hst_0
1694 34816 ALLOC_FUN_hst_1
1698 2382937 ALLOC_UP_THK_ctr
1701 0 E!NT_PERM_IND_ctr requires +RTS -Z
1702 [... lots more info omitted ...]
1703 0 GC_SEL_ABANDONED_ctr
1706 0 GC_FAILED_PROMOTION_ctr
1707 47524 GC_WORDS_COPIED_ctr
1710 <para>The formatting of the information above the row of asterisks
1711 is subject to change, but hopefully provides a useful
1712 human-readable summary. Below the asterisks <emphasis>all
1713 counters</emphasis> maintained by the ticky-ticky system are
1714 dumped, in a format intended to be machine-readable: zero or more
1715 spaces, an integer, a space, the counter name, and a newline.</para>
1717 <para>In fact, not <emphasis>all</emphasis> counters are
1718 necessarily dumped; compile- or run-time flags can render certain
1719 counters invalid. In this case, either the counter will simply
1720 not appear, or it will appear with a modified counter name,
1721 possibly along with an explanation for the omission (notice
1722 <literal>ENT_PERM_IND_ctr</literal> appears
1723 with an inserted <literal>!</literal> above). Software analysing
1724 this output should always check that it has the counters it
1725 expects. Also, beware: some of the counters can have
1726 <emphasis>large</emphasis> values!</para>
1733 ;;; Local Variables: ***
1735 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***