1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <sect1 id="runtime-control">
3 <title>Running a compiled program</title>
5 <indexterm><primary>runtime control of Haskell programs</primary></indexterm>
6 <indexterm><primary>running, compiled program</primary></indexterm>
7 <indexterm><primary>RTS options</primary></indexterm>
9 <para>To make an executable program, the GHC system compiles your
10 code and then links it with a non-trivial runtime system (RTS),
11 which handles storage management, profiling, etc.</para>
13 <para>If you use the <literal>-rtsopts</literal> flag when linking,
14 you have some control over the behaviour of the RTS, by giving
15 special command-line arguments to your program.</para>
17 <para>When your Haskell program starts up, its RTS extracts
18 command-line arguments bracketed between
19 <option>+RTS</option><indexterm><primary><option>+RTS</option></primary></indexterm>
21 <option>-RTS</option><indexterm><primary><option>-RTS</option></primary></indexterm>
22 as its own. For example:</para>
25 % ./a.out -f +RTS -p -S -RTS -h foo bar
28 <para>The RTS will snaffle <option>-p</option> <option>-S</option>
29 for itself, and the remaining arguments <literal>-f -h foo bar</literal>
30 will be handed to your program if/when it calls
31 <function>System.getArgs</function>.</para>
33 <para>No <option>-RTS</option> option is required if the
34 runtime-system options extend to the end of the command line, as in
38 % hls -ltr /usr/etc +RTS -A5m
41 <para>If you absolutely positively want all the rest of the options
42 in a command line to go to the program (and not the RTS), use a
43 <option>––RTS</option><indexterm><primary><option>--RTS</option></primary></indexterm>.</para>
45 <para>As always, for RTS options that take
46 <replaceable>size</replaceable>s: If the last character of
47 <replaceable>size</replaceable> is a K or k, multiply by 1000; if an
48 M or m, by 1,000,000; if a G or G, by 1,000,000,000. (And any
49 wraparound in the counters is <emphasis>your</emphasis>
52 <para>Giving a <literal>+RTS -?</literal>
53 <indexterm><primary><option>-?</option></primary><secondary>RTS option</secondary></indexterm> option
54 will print out the RTS options actually available in your program
55 (which vary, depending on how you compiled).</para>
57 <para>NOTE: since GHC is itself compiled by GHC, you can change RTS
58 options in the compiler using the normal
59 <literal>+RTS ... -RTS</literal>
60 combination. eg. to increase the maximum heap
61 size for a compilation to 128M, you would add
62 <literal>+RTS -M128m -RTS</literal>
63 to the command line.</para>
65 <sect2 id="rts-options-environment">
66 <title>Setting global RTS options</title>
68 <indexterm><primary>RTS options</primary><secondary>from the environment</secondary></indexterm>
69 <indexterm><primary>environment variable</primary><secondary>for
70 setting RTS options</secondary></indexterm>
72 <para>When the <literal>-rtsopts</literal> flag is used when linking,
73 RTS options are also taken from the environment variable
74 <envar>GHCRTS</envar><indexterm><primary><envar>GHCRTS</envar></primary>
75 </indexterm>. For example, to set the maximum heap size
76 to 128M for all GHC-compiled programs (using an
77 <literal>sh</literal>-like shell):</para>
84 <para>RTS options taken from the <envar>GHCRTS</envar> environment
85 variable can be overridden by options given on the command
90 <sect2 id="rts-options-misc">
91 <title>Miscellaneous RTS options</title>
95 <term><option>-V<replaceable>secs</replaceable></option>
96 <indexterm><primary><option>-V</option></primary><secondary>RTS
97 option</secondary></indexterm></term>
99 <para>Sets the interval that the RTS clock ticks at. The
100 runtime uses a single timer signal to count ticks; this timer
101 signal is used to control the context switch timer (<xref
102 linkend="using-concurrent" />) and the heap profiling
103 timer <xref linkend="rts-options-heap-prof" />. Also, the
104 time profiler uses the RTS timer signal directly to record
105 time profiling samples.</para>
107 <para>Normally, setting the <option>-V</option> option
108 directly is not necessary: the resolution of the RTS timer is
109 adjusted automatically if a short interval is requested with
110 the <option>-C</option> or <option>-i</option> options.
111 However, setting <option>-V</option> is required in order to
112 increase the resolution of the time profiler.</para>
114 <para>Using a value of zero disables the RTS clock
115 completely, and has the effect of disabling timers that
116 depend on it: the context switch timer and the heap profiling
117 timer. Context switches will still happen, but
118 deterministically and at a rate much faster than normal.
119 Disabling the interval timer is useful for debugging, because
120 it eliminates a source of non-determinism at runtime.</para>
125 <term><option>--install-signal-handlers=<replaceable>yes|no</replaceable></option>
126 <indexterm><primary><option>--install-signal-handlers</option></primary><secondary>RTS
127 option</secondary></indexterm></term>
129 <para>If yes (the default), the RTS installs signal handlers to catch
130 things like ctrl-C. This option is primarily useful for when
131 you are using the Haskell code as a DLL, and want to set your
132 own signal handlers.</para>
135 with <option>--install-signal-handlers=no</option>, the RTS
136 interval timer signal is still enabled. The timer signal
137 is either SIGVTALRM or SIGALRM, depending on the RTS
138 configuration and OS capabilities. To disable the timer
139 signal, use the <literal>-V0</literal> RTS option (see
146 <term><option>-xm<replaceable>address</replaceable></option>
147 <indexterm><primary><option>-xm</option></primary><secondary>RTS
148 option</secondary></indexterm></term>
151 WARNING: this option is for working around memory
152 allocation problems only. Do not use unless GHCi fails
153 with a message like “<literal>failed to mmap() memory below 2Gb</literal>”. If you need to use this option to get GHCi working
154 on your machine, please file a bug.
158 On 64-bit machines, the RTS needs to allocate memory in the
159 low 2Gb of the address space. Support for this across
160 different operating systems is patchy, and sometimes fails.
161 This option is there to give the RTS a hint about where it
162 should be able to allocate memory in the low 2Gb of the
163 address space. For example, <literal>+RTS -xm20000000
164 -RTS</literal> would hint that the RTS should allocate
165 starting at the 0.5Gb mark. The default is to use the OS's
166 built-in support for allocating memory in the low 2Gb if
167 available (e.g. <literal>mmap</literal>
168 with <literal>MAP_32BIT</literal> on Linux), or
169 otherwise <literal>-xm40000000</literal>.
176 <sect2 id="rts-options-gc">
177 <title>RTS options to control the garbage collector</title>
179 <indexterm><primary>garbage collector</primary><secondary>options</secondary></indexterm>
180 <indexterm><primary>RTS options</primary><secondary>garbage collection</secondary></indexterm>
182 <para>There are several options to give you precise control over
183 garbage collection. Hopefully, you won't need any of these in
184 normal operation, but there are several things that can be tweaked
185 for maximum performance.</para>
191 <option>-A</option><replaceable>size</replaceable>
192 <indexterm><primary><option>-A</option></primary><secondary>RTS option</secondary></indexterm>
193 <indexterm><primary>allocation area, size</primary></indexterm>
196 <para>[Default: 512k] Set the allocation area size
197 used by the garbage collector. The allocation area
198 (actually generation 0 step 0) is fixed and is never resized
199 (unless you use <option>-H</option>, below).</para>
201 <para>Increasing the allocation area size may or may not
202 give better performance (a bigger allocation area means
203 worse cache behaviour but fewer garbage collections and less
206 <para>With only 1 generation (<option>-G1</option>) the
207 <option>-A</option> option specifies the minimum allocation
208 area, since the actual size of the allocation area will be
209 resized according to the amount of data in the heap (see
210 <option>-F</option>, below).</para>
217 <indexterm><primary><option>-c</option></primary><secondary>RTS option</secondary></indexterm>
218 <indexterm><primary>garbage collection</primary><secondary>compacting</secondary></indexterm>
219 <indexterm><primary>compacting garbage collection</primary></indexterm>
222 <para>Use a compacting algorithm for collecting the oldest
223 generation. By default, the oldest generation is collected
224 using a copying algorithm; this option causes it to be
225 compacted in-place instead. The compaction algorithm is
226 slower than the copying algorithm, but the savings in memory
227 use can be considerable.</para>
229 <para>For a given heap size (using the <option>-H</option>
230 option), compaction can in fact reduce the GC cost by
231 allowing fewer GCs to be performed. This is more likely
232 when the ratio of live data to heap size is high, say
233 >30%.</para>
235 <para>NOTE: compaction doesn't currently work when a single
236 generation is requested using the <option>-G1</option>
242 <term><option>-c</option><replaceable>n</replaceable></term>
245 <para>[Default: 30] Automatically enable
246 compacting collection when the live data exceeds
247 <replaceable>n</replaceable>% of the maximum heap size
248 (see the <option>-M</option> option). Note that the maximum
249 heap size is unlimited by default, so this option has no
250 effect unless the maximum heap size is set with
251 <option>-M</option><replaceable>size</replaceable>. </para>
257 <option>-F</option><replaceable>factor</replaceable>
258 <indexterm><primary><option>-F</option></primary><secondary>RTS option</secondary></indexterm>
259 <indexterm><primary>heap size, factor</primary></indexterm>
263 <para>[Default: 2] This option controls the amount
264 of memory reserved for the older generations (and in the
265 case of a two space collector the size of the allocation
266 area) as a factor of the amount of live data. For example,
267 if there was 2M of live data in the oldest generation when
268 we last collected it, then by default we'll wait until it
269 grows to 4M before collecting it again.</para>
271 <para>The default seems to work well here. If you have
272 plenty of memory, it is usually better to use
273 <option>-H</option><replaceable>size</replaceable> than to
275 <option>-F</option><replaceable>factor</replaceable>.</para>
277 <para>The <option>-F</option> setting will be automatically
278 reduced by the garbage collector when the maximum heap size
279 (the <option>-M</option><replaceable>size</replaceable>
280 setting) is approaching.</para>
286 <option>-G</option><replaceable>generations</replaceable>
287 <indexterm><primary><option>-G</option></primary><secondary>RTS option</secondary></indexterm>
288 <indexterm><primary>generations, number of</primary></indexterm>
291 <para>[Default: 2] Set the number of generations
292 used by the garbage collector. The default of 2 seems to be
293 good, but the garbage collector can support any number of
294 generations. Anything larger than about 4 is probably not a
295 good idea unless your program runs for a
296 <emphasis>long</emphasis> time, because the oldest
297 generation will hardly ever get collected.</para>
299 <para>Specifying 1 generation with <option>+RTS -G1</option>
300 gives you a simple 2-space collector, as you would expect.
301 In a 2-space collector, the <option>-A</option> option (see
302 above) specifies the <emphasis>minimum</emphasis> allocation
303 area size, since the allocation area will grow with the
304 amount of live data in the heap. In a multi-generational
305 collector the allocation area is a fixed size (unless you
306 use the <option>-H</option> option, see below).</para>
312 <option>-qg<optional><replaceable>gen</replaceable></optional></option>
313 <indexterm><primary><option>-qg</option><secondary>RTS
314 option</secondary></primary></indexterm>
317 <para>[New in GHC 6.12.1] [Default: 0]
319 generation <replaceable>gen</replaceable> and higher.
320 Omitting <replaceable>gen</replaceable> turns off the
321 parallel GC completely, reverting to sequential GC.</para>
323 <para>The default parallel GC settings are usually suitable
324 for parallel programs (i.e. those
325 using <literal>par</literal>, Strategies, or with multiple
326 threads). However, it is sometimes beneficial to enable
327 the parallel GC for a single-threaded sequential program
328 too, especially if the program has a large amount of heap
329 data and GC is a significant fraction of runtime. To use
330 the parallel GC in a sequential program, enable the
331 parallel runtime with a suitable <literal>-N</literal>
332 option, and additionally it might be beneficial to
333 restrict parallel GC to the old generation
334 with <literal>-qg1</literal>.</para>
340 <option>-qb<optional><replaceable>gen</replaceable></optional></option>
341 <indexterm><primary><option>-qb</option><secondary>RTS
342 option</secondary></primary></indexterm>
346 [New in GHC 6.12.1] [Default: 1] Use
347 load-balancing in the parallel GC in
348 generation <replaceable>gen</replaceable> and higher.
349 Omitting <replaceable>gen</replaceable> disables
350 load-balancing entirely.</para>
353 Load-balancing shares out the work of GC between the
354 available cores. This is a good idea when the heap is
355 large and we need to parallelise the GC work, however it
356 is also pessimal for the short young-generation
357 collections in a parallel program, because it can harm
358 locality by moving data from the cache of the CPU where is
359 it being used to the cache of another CPU. Hence the
360 default is to do load-balancing only in the
361 old-generation. In fact, for a parallel program it is
362 sometimes beneficial to disable load-balancing entirely
363 with <literal>-qb</literal>.
370 <option>-H</option><replaceable>size</replaceable>
371 <indexterm><primary><option>-H</option></primary><secondary>RTS option</secondary></indexterm>
372 <indexterm><primary>heap size, suggested</primary></indexterm>
375 <para>[Default: 0] This option provides a
376 “suggested heap size” for the garbage collector. The
377 garbage collector will use about this much memory until the
378 program residency grows and the heap size needs to be
379 expanded to retain reasonable performance.</para>
381 <para>By default, the heap will start small, and grow and
382 shrink as necessary. This can be bad for performance, so if
383 you have plenty of memory it's worthwhile supplying a big
384 <option>-H</option><replaceable>size</replaceable>. For
385 improving GC performance, using
386 <option>-H</option><replaceable>size</replaceable> is
387 usually a better bet than
388 <option>-A</option><replaceable>size</replaceable>.</para>
394 <option>-I</option><replaceable>seconds</replaceable>
395 <indexterm><primary><option>-I</option></primary>
396 <secondary>RTS option</secondary>
398 <indexterm><primary>idle GC</primary>
402 <para>(default: 0.3) In the threaded and SMP versions of the RTS (see
403 <option>-threaded</option>, <xref linkend="options-linker" />), a
404 major GC is automatically performed if the runtime has been idle
405 (no Haskell computation has been running) for a period of time.
406 The amount of idle time which must pass before a GC is performed is
407 set by the <option>-I</option><replaceable>seconds</replaceable>
408 option. Specifying <option>-I0</option> disables the idle GC.</para>
410 <para>For an interactive application, it is probably a good idea to
411 use the idle GC, because this will allow finalizers to run and
412 deadlocked threads to be detected in the idle time when no Haskell
413 computation is happening. Also, it will mean that a GC is less
414 likely to happen when the application is busy, and so
415 responsiveness may be improved. However, if the amount of live data in
416 the heap is particularly large, then the idle GC can cause a
417 significant delay, and too small an interval could adversely affect
418 interactive responsiveness.</para>
420 <para>This is an experimental feature, please let us know if it
421 causes problems and/or could benefit from further tuning.</para>
427 <option>-ki</option><replaceable>size</replaceable>
428 <indexterm><primary><option>-k</option></primary><secondary>RTS option</secondary></indexterm>
429 <indexterm><primary>stack, initial size</primary></indexterm>
433 [Default: 1k] Set the initial stack size for new
434 threads. (Note: this flag used to be
435 simply <option>-k</option>, but was renamed
436 to <option>-ki</option> in GHC 7.2.1. The old name is
437 still accepted for backwards compatibility, but that may
438 be removed in a future version).
442 Thread stacks (including the main thread's stack) live on
443 the heap. As the stack grows, new stack chunks are added
444 as required; if the stack shrinks again, these extra stack
445 chunks are reclaimed by the garbage collector. The
446 default initial stack size is deliberately small, in order
447 to keep the time and space overhead for thread creation to
448 a minimum, and to make it practical to spawn threads for
449 even tiny pieces of work.
456 <option>-kc</option><replaceable>size</replaceable>
457 <indexterm><primary><option>-kc</option></primary><secondary>RTS
458 option</secondary></indexterm>
459 <indexterm><primary>stack</primary><secondary>chunk size</secondary></indexterm>
463 [Default: 32k] Set the size of “stack
464 chunks”. When a thread's current stack overflows, a
465 new stack chunk is created and added to the thread's
466 stack, until the limit set by <option>-K</option> is
471 The advantage of smaller stack chunks is that the garbage
472 collector can avoid traversing stack chunks if they are
473 known to be unmodified since the last collection, so
474 reducing the chunk size means that the garbage collector
475 can identify more stack as unmodified, and the GC overhead
476 might be reduced. On the other hand, making stack chunks
477 too small adds some overhead as there will be more
478 overflow/underflow between chunks. The default setting of
479 32k appears to be a reasonable compromise in most cases.
486 <option>-kb</option><replaceable>size</replaceable>
487 <indexterm><primary><option>-kc</option></primary><secondary>RTS
488 option</secondary></indexterm>
489 <indexterm><primary>stack</primary><secondary>chunk buffer size</secondary></indexterm>
493 [Default: 1k] Sets the stack chunk buffer size.
494 When a stack chunk overflows and a new stack chunk is
495 created, some of the data from the previous stack chunk is
496 moved into the new chunk, to avoid an immediate underflow
497 and repeated overflow/underflow at the boundary. The
498 amount of stack moved is set by the <option>-kb</option>
502 Note that to avoid wasting space, this value should
503 typically be less than 10% of the size of a stack
504 chunk (<option>-kc</option>), because in a chain of stack
505 chunks, each chunk will have a gap of unused space of this
513 <option>-K</option><replaceable>size</replaceable>
514 <indexterm><primary><option>-K</option></primary><secondary>RTS option</secondary></indexterm>
515 <indexterm><primary>stack, maximum size</primary></indexterm>
518 <para>[Default: 8M] Set the maximum stack size for
519 an individual thread to <replaceable>size</replaceable>
520 bytes. If the thread attempts to exceed this limit, it will
521 be send the <literal>StackOverflow</literal> exception.
524 This option is there mainly to stop the program eating up
525 all the available memory in the machine if it gets into an
533 <option>-m</option><replaceable>n</replaceable>
534 <indexterm><primary><option>-m</option></primary><secondary>RTS option</secondary></indexterm>
535 <indexterm><primary>heap, minimum free</primary></indexterm>
538 <para>Minimum % <replaceable>n</replaceable> of heap
539 which must be available for allocation. The default is
546 <option>-M</option><replaceable>size</replaceable>
547 <indexterm><primary><option>-M</option></primary><secondary>RTS option</secondary></indexterm>
548 <indexterm><primary>heap size, maximum</primary></indexterm>
551 <para>[Default: unlimited] Set the maximum heap size to
552 <replaceable>size</replaceable> bytes. The heap normally
553 grows and shrinks according to the memory requirements of
554 the program. The only reason for having this option is to
555 stop the heap growing without bound and filling up all the
556 available swap space, which at the least will result in the
557 program being summarily killed by the operating
560 <para>The maximum heap size also affects other garbage
561 collection parameters: when the amount of live data in the
562 heap exceeds a certain fraction of the maximum heap size,
563 compacting collection will be automatically enabled for the
564 oldest generation, and the <option>-F</option> parameter
565 will be reduced in order to avoid exceeding the maximum heap
572 <option>-t</option><optional><replaceable>file</replaceable></optional>
573 <indexterm><primary><option>-t</option></primary><secondary>RTS option</secondary></indexterm>
576 <option>-s</option><optional><replaceable>file</replaceable></optional>
577 <indexterm><primary><option>-s</option></primary><secondary>RTS option</secondary></indexterm>
580 <option>-S</option><optional><replaceable>file</replaceable></optional>
581 <indexterm><primary><option>-S</option></primary><secondary>RTS option</secondary></indexterm>
584 <option>--machine-readable</option>
585 <indexterm><primary><option>--machine-readable</option></primary><secondary>RTS option</secondary></indexterm>
588 <para>These options produce runtime-system statistics, such
589 as the amount of time spent executing the program and in the
590 garbage collector, the amount of memory allocated, the
591 maximum size of the heap, and so on. The three
592 variants give different levels of detail:
593 <option>-t</option> produces a single line of output in the
594 same format as GHC's <option>-Rghc-timing</option> option,
595 <option>-s</option> produces a more detailed summary at the
596 end of the program, and <option>-S</option> additionally
597 produces information about each and every garbage
600 <para>The output is placed in
601 <replaceable>file</replaceable>. If
602 <replaceable>file</replaceable> is omitted, then the output
603 is sent to <constant>stderr</constant>.</para>
606 If you use the <literal>-t</literal> flag then, when your
607 program finishes, you will see something like this:
611 <<ghc: 36169392 bytes, 69 GCs, 603392/1065272 avg/max bytes residency (2 samples), 3M in use, 0.00 INIT (0.00 elapsed), 0.02 MUT (0.02 elapsed), 0.07 GC (0.07 elapsed) :ghc>>
621 The total number of bytes allocated by the program over the
627 The total number of garbage collections performed.
632 The average and maximum "residency", which is the amount of
633 live data in bytes. The runtime can only determine the
634 amount of live data during a major GC, which is why the
635 number of samples corresponds to the number of major GCs
636 (and is usually relatively small). To get a better picture
637 of the heap profile of your program, use
638 the <option>-hT</option> RTS option
639 (<xref linkend="rts-profiling" />).
644 The peak memory the RTS has allocated from the OS.
649 The amount of CPU time and elapsed wall clock time while
650 initialising the runtime system (INIT), running the program
651 itself (MUT, the mutator), and garbage collecting (GC).
657 You can also get this in a more future-proof, machine readable
658 format, with <literal>-t --machine-readable</literal>:
662 [("bytes allocated", "36169392")
664 ,("average_bytes_used", "603392")
665 ,("max_bytes_used", "1065272")
666 ,("num_byte_usage_samples", "2")
667 ,("peak_megabytes_allocated", "3")
668 ,("init_cpu_seconds", "0.00")
669 ,("init_wall_seconds", "0.00")
670 ,("mutator_cpu_seconds", "0.02")
671 ,("mutator_wall_seconds", "0.02")
672 ,("GC_cpu_seconds", "0.07")
673 ,("GC_wall_seconds", "0.07")
678 If you use the <literal>-s</literal> flag then, when your
679 program finishes, you will see something like this (the exact
680 details will vary depending on what sort of RTS you have, e.g.
681 you will only see profiling data if your RTS is compiled for
686 36,169,392 bytes allocated in the heap
687 4,057,632 bytes copied during GC
688 1,065,272 bytes maximum residency (2 sample(s))
689 54,312 bytes maximum slop
690 3 MB total memory in use (0 MB lost due to fragmentation)
692 Generation 0: 67 collections, 0 parallel, 0.04s, 0.03s elapsed
693 Generation 1: 2 collections, 0 parallel, 0.03s, 0.04s elapsed
695 SPARKS: 359207 (557 converted, 149591 pruned)
697 INIT time 0.00s ( 0.00s elapsed)
698 MUT time 0.01s ( 0.02s elapsed)
699 GC time 0.07s ( 0.07s elapsed)
700 EXIT time 0.00s ( 0.00s elapsed)
701 Total time 0.08s ( 0.09s elapsed)
703 %GC time 89.5% (75.3% elapsed)
705 Alloc rate 4,520,608,923 bytes per MUT second
707 Productivity 10.5% of total user, 9.1% of total elapsed
713 The "bytes allocated in the heap" is the total bytes allocated
714 by the program over the whole run.
719 GHC uses a copying garbage collector by default. "bytes copied
720 during GC" tells you how many bytes it had to copy during
726 The maximum space actually used by your program is the
727 "bytes maximum residency" figure. This is only checked during
728 major garbage collections, so it is only an approximation;
729 the number of samples tells you how many times it is checked.
734 The "bytes maximum slop" tells you the most space that is ever
735 wasted due to the way GHC allocates memory in blocks. Slop is
736 memory at the end of a block that was wasted. There's no way
737 to control this; we just like to see how much memory is being
743 The "total memory in use" tells you the peak memory the RTS has
744 allocated from the OS.
749 Next there is information about the garbage collections done.
750 For each generation it says how many garbage collections were
751 done, how many of those collections were done in parallel,
752 the total CPU time used for garbage collecting that generation,
753 and the total wall clock time elapsed while garbage collecting
758 <para>The <literal>SPARKS</literal> statistic refers to the
759 use of <literal>Control.Parallel.par</literal> and related
760 functionality in the program. Each spark represents a call
761 to <literal>par</literal>; a spark is "converted" when it is
762 executed in parallel; and a spark is "pruned" when it is
763 found to be already evaluated and is discarded from the pool
764 by the garbage collector. Any remaining sparks are
765 discarded at the end of execution, so "converted" plus
766 "pruned" does not necessarily add up to the total.</para>
770 Next there is the CPU time and wall clock time elapsed broken
771 down by what the runtime system was doing at the time.
772 INIT is the runtime system initialisation.
773 MUT is the mutator time, i.e. the time spent actually running
775 GC is the time spent doing garbage collection.
776 RP is the time spent doing retainer profiling.
777 PROF is the time spent doing other profiling.
778 EXIT is the runtime system shutdown time.
779 And finally, Total is, of course, the total.
782 %GC time tells you what percentage GC is of Total.
783 "Alloc rate" tells you the "bytes allocated in the heap" divided
785 "Productivity" tells you what percentage of the Total CPU and wall
786 clock elapsed times are spent in the mutator (MUT).
792 The <literal>-S</literal> flag, as well as giving the same
793 output as the <literal>-s</literal> flag, prints information
794 about each GC as it happens:
798 Alloc Copied Live GC GC TOT TOT Page Flts
799 bytes bytes bytes user elap user elap
800 528496 47728 141512 0.01 0.02 0.02 0.02 0 0 (Gen: 1)
802 524944 175944 1726384 0.00 0.00 0.08 0.11 0 0 (Gen: 0)
806 For each garbage collection, we print:
812 How many bytes we allocated this garbage collection.
817 How many bytes we copied this garbage collection.
822 How many bytes are currently live.
827 How long this garbage collection took (CPU time and elapsed
833 How long the program has been running (CPU time and elapsed
839 How many page faults occured this garbage collection.
844 How many page faults occured since the end of the last garbage
850 Which generation is being garbage collected.
862 <title>RTS options for concurrency and parallelism</title>
864 <para>The RTS options related to concurrency are described in
865 <xref linkend="using-concurrent" />, and those for parallelism in
866 <xref linkend="parallel-options"/>.</para>
869 <sect2 id="rts-profiling">
870 <title>RTS options for profiling</title>
872 <para>Most profiling runtime options are only available when you
873 compile your program for profiling (see
874 <xref linkend="prof-compiler-options" />, and
875 <xref linkend="rts-options-heap-prof" /> for the runtime options).
876 However, there is one profiling option that is available
877 for ordinary non-profiled executables:</para>
883 <indexterm><primary><option>-hT</option></primary><secondary>RTS
884 option</secondary></indexterm>
887 <para>Generates a basic heap profile, in the
888 file <literal><replaceable>prog</replaceable>.hp</literal>.
889 To produce the heap profile graph,
890 use <command>hp2ps</command> (see <xref linkend="hp2ps"
891 />). The basic heap profile is broken down by data
892 constructor, with other types of closures (functions, thunks,
893 etc.) grouped into broad categories
894 (e.g. <literal>FUN</literal>, <literal>THUNK</literal>). To
895 get a more detailed profile, use the full profiling
896 support (<xref linkend="profiling" />).</para>
902 <sect2 id="rts-eventlog">
903 <title>Tracing</title>
905 <indexterm><primary>tracing</primary></indexterm>
906 <indexterm><primary>events</primary></indexterm>
907 <indexterm><primary>eventlog files</primary></indexterm>
910 When the program is linked with the <option>-eventlog</option>
911 option (<xref linkend="options-linker" />), runtime events can
912 be logged in two ways:
918 In binary format to a file for later analysis by a
919 variety of tools. One such tool
920 is <ulink url="http://hackage.haskell.org/package/ThreadScope">ThreadScope</ulink><indexterm><primary>ThreadScope</primary></indexterm>,
921 which interprets the event log to produce a visual parallel
922 execution profile of the program.
927 As text to standard output, for debugging purposes.
935 <option>-l<optional><replaceable>flags</replaceable></optional></option>
936 <indexterm><primary><option>-l</option></primary><secondary>RTS option</secondary></indexterm>
940 Log events in binary format to the
941 file <filename><replaceable>program</replaceable>.eventlog</filename>,
942 where <replaceable>flags</replaceable> is a sequence of
943 zero or more characters indicating which kinds of events
944 to log. Currently there is only one type
945 supported: <literal>-ls</literal>, for scheduler events.
949 The format of the log file is described by the header
950 <filename>EventLogFormat.h</filename> that comes with
951 GHC, and it can be parsed in Haskell using
952 the <ulink url="http://hackage.haskell.org/package/ghc-events">ghc-events</ulink>
953 library. To dump the contents of
954 a <literal>.eventlog</literal> file as text, use the
955 tool <literal>show-ghc-events</literal> that comes with
956 the <ulink url="http://hackage.haskell.org/package/ghc-events">ghc-events</ulink>
964 <option>-v</option><optional><replaceable>flags</replaceable></optional>
965 <indexterm><primary><option>-v</option></primary><secondary>RTS option</secondary></indexterm>
969 Log events as text to standard output, instead of to
970 the <literal>.eventlog</literal> file.
971 The <replaceable>flags</replaceable> are the same as
972 for <option>-l</option>, with the additional
973 option <literal>t</literal> which indicates that the
974 each event printed should be preceded by a timestamp value
975 (in the binary <literal>.eventlog</literal> file, all
976 events are automatically associated with a timestamp).
985 options <option>-D<replaceable>x</replaceable></option> also
986 generate events which are logged using the tracing framework.
987 By default those events are dumped as text to stdout
988 (<option>-D<replaceable>x</replaceable></option>
989 implies <option>-v</option>), but they may instead be stored in
990 the binary eventlog file by using the <option>-l</option>
995 <sect2 id="rts-options-debugging">
996 <title>RTS options for hackers, debuggers, and over-interested
999 <indexterm><primary>RTS options, hacking/debugging</primary></indexterm>
1001 <para>These RTS options might be used (a) to avoid a GHC bug,
1002 (b) to see “what's really happening”, or
1003 (c) because you feel like it. Not recommended for everyday
1011 <indexterm><primary><option>-B</option></primary><secondary>RTS option</secondary></indexterm>
1014 <para>Sound the bell at the start of each (major) garbage
1017 <para>Oddly enough, people really do use this option! Our
1018 pal in Durham (England), Paul Callaghan, writes: “Some
1019 people here use it for a variety of
1020 purposes—honestly!—e.g., confirmation that the
1021 code/machine is doing something, infinite loop detection,
1022 gauging cost of recently added code. Certain people can even
1023 tell what stage [the program] is in by the beep
1024 pattern. But the major use is for annoying others in the
1025 same office…”</para>
1031 <option>-D</option><replaceable>x</replaceable>
1032 <indexterm><primary>-D</primary><secondary>RTS option</secondary></indexterm>
1036 An RTS debugging flag; only availble if the program was
1037 linked with the <option>-debug</option> option. Various
1038 values of <replaceable>x</replaceable> are provided to
1039 enable debug messages and additional runtime sanity checks
1040 in different subsystems in the RTS, for
1041 example <literal>+RTS -Ds -RTS</literal> enables debug
1042 messages from the scheduler.
1043 Use <literal>+RTS -?</literal> to find out which
1044 debug flags are supported.
1048 Debug messages will be sent to the binary event log file
1049 instead of stdout if the <option>-l</option> option is
1050 added. This might be useful for reducing the overhead of
1058 <option>-r</option><replaceable>file</replaceable>
1059 <indexterm><primary><option>-r</option></primary><secondary>RTS option</secondary></indexterm>
1060 <indexterm><primary>ticky ticky profiling</primary></indexterm>
1061 <indexterm><primary>profiling</primary><secondary>ticky ticky</secondary></indexterm>
1064 <para>Produce “ticky-ticky” statistics at the
1065 end of the program run (only available if the program was
1066 linked with <option>-debug</option>).
1067 The <replaceable>file</replaceable> business works just like
1068 on the <option>-S</option> RTS option, above.</para>
1070 <para>For more information on ticky-ticky profiling, see
1071 <xref linkend="ticky-ticky"/>.</para>
1077 <option>-xc</option>
1078 <indexterm><primary><option>-xc</option></primary><secondary>RTS option</secondary></indexterm>
1081 <para>(Only available when the program is compiled for
1082 profiling.) When an exception is raised in the program,
1083 this option causes the current cost-centre-stack to be
1084 dumped to <literal>stderr</literal>.</para>
1086 <para>This can be particularly useful for debugging: if your
1087 program is complaining about a <literal>head []</literal>
1088 error and you haven't got a clue which bit of code is
1089 causing it, compiling with <literal>-prof
1090 -auto-all</literal> and running with <literal>+RTS -xc
1091 -RTS</literal> will tell you exactly the call stack at the
1092 point the error was raised.</para>
1094 <para>The output contains one line for each exception raised
1095 in the program (the program might raise and catch several
1096 exceptions during its execution), where each line is of the
1100 < cc<subscript>1</subscript>, ..., cc<subscript>n</subscript> >
1102 <para>each <literal>cc</literal><subscript>i</subscript> is
1103 a cost centre in the program (see <xref
1104 linkend="cost-centres"/>), and the sequence represents the
1105 “call stack” at the point the exception was
1106 raised. The leftmost item is the innermost function in the
1107 call stack, and the rightmost item is the outermost
1116 <indexterm><primary><option>-Z</option></primary><secondary>RTS option</secondary></indexterm>
1119 <para>Turn <emphasis>off</emphasis> “update-frame
1120 squeezing” at garbage-collection time. (There's no
1121 particularly good reason to turn it off, except to ensure
1122 the accuracy of certain data collected regarding thunk entry
1131 <title>Linker flags to change RTS behaviour</title>
1133 <indexterm><primary>RTS behaviour, changing</primary></indexterm>
1136 GHC lets you exercise rudimentary control over the RTS settings
1137 for any given program, by using the <literal>-with-rtsopts</literal>
1138 linker flag. For example, to set <literal>-H128m -K1m</literal>,
1139 link with <literal>-with-rtsopts="-H128m -K1m"</literal>.
1144 <sect2 id="rts-hooks">
1145 <title>“Hooks” to change RTS behaviour</title>
1147 <indexterm><primary>hooks</primary><secondary>RTS</secondary></indexterm>
1148 <indexterm><primary>RTS hooks</primary></indexterm>
1149 <indexterm><primary>RTS behaviour, changing</primary></indexterm>
1151 <para>GHC lets you exercise rudimentary control over the RTS
1152 settings for any given program, by compiling in a
1153 “hook” that is called by the run-time system. The RTS
1154 contains stub definitions for all these hooks, but by writing your
1155 own version and linking it on the GHC command line, you can
1156 override the defaults.</para>
1158 <para>Owing to the vagaries of DLL linking, these hooks don't work
1159 under Windows when the program is built dynamically.</para>
1161 <para>The hook <literal>ghc_rts_opts</literal><indexterm><primary><literal>ghc_rts_opts</literal></primary>
1162 </indexterm>lets you set RTS
1163 options permanently for a given program. A common use for this is
1164 to give your program a default heap and/or stack size that is
1165 greater than the default. For example, to set <literal>-H128m
1166 -K1m</literal>, place the following definition in a C source
1170 char *ghc_rts_opts = "-H128m -K1m";
1173 <para>Compile the C file, and include the object file on the
1174 command line when you link your Haskell program.</para>
1176 <para>These flags are interpreted first, before any RTS flags from
1177 the <literal>GHCRTS</literal> environment variable and any flags
1178 on the command line.</para>
1180 <para>You can also change the messages printed when the runtime
1181 system “blows up,” e.g., on stack overflow. The hooks
1182 for these are as follows:</para>
1188 <function>void OutOfHeapHook (unsigned long, unsigned long)</function>
1189 <indexterm><primary><function>OutOfHeapHook</function></primary></indexterm>
1192 <para>The heap-overflow message.</para>
1198 <function>void StackOverflowHook (long int)</function>
1199 <indexterm><primary><function>StackOverflowHook</function></primary></indexterm>
1202 <para>The stack-overflow message.</para>
1208 <function>void MallocFailHook (long int)</function>
1209 <indexterm><primary><function>MallocFailHook</function></primary></indexterm>
1212 <para>The message printed if <function>malloc</function>
1218 <para>For examples of the use of these hooks, see GHC's own
1219 versions in the file
1220 <filename>ghc/compiler/parser/hschooks.c</filename> in a GHC
1225 <title>Getting information about the RTS</title>
1227 <indexterm><primary>RTS</primary></indexterm>
1229 <para>It is possible to ask the RTS to give some information about
1230 itself. To do this, use the <option>--info</option> flag, e.g.</para>
1232 $ ./a.out +RTS --info
1234 ,("GHC version", "6.7")
1235 ,("RTS way", "rts_p")
1236 ,("Host platform", "x86_64-unknown-linux")
1237 ,("Host architecture", "x86_64")
1238 ,("Host OS", "linux")
1239 ,("Host vendor", "unknown")
1240 ,("Build platform", "x86_64-unknown-linux")
1241 ,("Build architecture", "x86_64")
1242 ,("Build OS", "linux")
1243 ,("Build vendor", "unknown")
1244 ,("Target platform", "x86_64-unknown-linux")
1245 ,("Target architecture", "x86_64")
1246 ,("Target OS", "linux")
1247 ,("Target vendor", "unknown")
1248 ,("Word size", "64")
1249 ,("Compiler unregisterised", "NO")
1250 ,("Tables next to code", "YES")
1253 <para>The information is formatted such that it can be read as a
1254 of type <literal>[(String, String)]</literal>. Currently the following
1255 fields are present:</para>
1260 <term><literal>GHC RTS</literal></term>
1262 <para>Is this program linked against the GHC RTS? (always
1268 <term><literal>GHC version</literal></term>
1270 <para>The version of GHC used to compile this program.</para>
1275 <term><literal>RTS way</literal></term>
1277 <para>The variant (“way”) of the runtime. The
1278 most common values are <literal>rts</literal> (vanilla),
1279 <literal>rts_thr</literal> (threaded runtime, i.e. linked using the
1280 <literal>-threaded</literal> option) and <literal>rts_p</literal>
1281 (profiling runtime, i.e. linked using the <literal>-prof</literal>
1282 option). Other variants include <literal>debug</literal>
1283 (linked using <literal>-debug</literal>),
1284 <literal>t</literal> (ticky-ticky profiling) and
1285 <literal>dyn</literal> (the RTS is
1286 linked in dynamically, i.e. a shared library, rather than statically
1287 linked into the executable itself). These can be combined,
1288 e.g. you might have <literal>rts_thr_debug_p</literal>.</para>
1294 <literal>Target platform</literal>,
1295 <literal>Target architecture</literal>,
1296 <literal>Target OS</literal>,
1297 <literal>Target vendor</literal>
1300 <para>These are the platform the program is compiled to run on.</para>
1306 <literal>Build platform</literal>,
1307 <literal>Build architecture</literal>,
1308 <literal>Build OS</literal>,
1309 <literal>Build vendor</literal>
1312 <para>These are the platform where the program was built
1313 on. (That is, the target platform of GHC itself.) Ordinarily
1314 this is identical to the target platform. (It could potentially
1315 be different if cross-compiling.)</para>
1321 <literal>Host platform</literal>,
1322 <literal>Host architecture</literal>
1323 <literal>Host OS</literal>
1324 <literal>Host vendor</literal>
1327 <para>These are the platform where GHC itself was compiled.
1328 Again, this would normally be identical to the build and
1329 target platforms.</para>
1334 <term><literal>Word size</literal></term>
1336 <para>Either <literal>"32"</literal> or <literal>"64"</literal>,
1337 reflecting the word size of the target platform.</para>
1342 <term><literal>Compiler unregistered</literal></term>
1344 <para>Was this program compiled with an “unregistered”
1345 version of GHC? (I.e., a version of GHC that has no platform-specific
1346 optimisations compiled in, usually because this is a currently
1347 unsupported platform.) This value will usually be no, unless you're
1348 using an experimental build of GHC.</para>
1353 <term><literal>Tables next to code</literal></term>
1355 <para>Putting info tables directly next to entry code is a useful
1356 performance optimisation that is not available on all platforms.
1357 This field tells you whether the program has been compiled with
1358 this optimisation. (Usually yes, except on unusual platforms.)</para>
1368 ;;; Local Variables: ***
1369 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter" "sect1") ***