1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <chapter id="sooner-faster-quicker">
3 <title>Advice on: sooner, faster, smaller, thriftier</title>
5 <para>Please advise us of other “helpful hints” that
9 <title>Sooner: producing a program more quickly
12 <indexterm><primary>compiling faster</primary></indexterm>
13 <indexterm><primary>faster compiling</primary></indexterm>
17 <term>Don't use <option>-O</option> or (especially) <option>-O2</option>:</term>
19 <para>By using them, you are telling GHC that you are
20 willing to suffer longer compilation times for
21 better-quality code.</para>
23 <para>GHC is surprisingly zippy for normal compilations
24 without <option>-O</option>!</para>
29 <term>Use more memory:</term>
31 <para>Within reason, more memory for heap space means less
32 garbage collection for GHC, which means less compilation
33 time. If you use the <option>-Rghc-timing</option> option,
34 you'll get a garbage-collector report. (Again, you can use
35 the cheap-and-nasty <option>+RTS -S -RTS</option>
36 option to send the GC stats straight to standard
39 <para>If it says you're using more than 20% of total
40 time in garbage collecting, then more memory might
42 <option>-H<size></option><indexterm><primary><option>-H</option></primary></indexterm>
43 option. Increasing the default allocation area size used by
44 the compiler's RTS might also help: use the
45 <option>+RTS -A<size> -RTS</option><indexterm><primary>-A<size>
46 RTS option</primary></indexterm> option.</para>
48 <para>If GHC persists in being a bad memory citizen, please
49 report it as a bug.</para>
54 <term>Don't use too much memory!</term>
56 <para>As soon as GHC plus its “fellow citizens”
57 (other processes on your machine) start using more than the
58 <emphasis>real memory</emphasis> on your machine, and the
59 machine starts “thrashing,” <emphasis>the party
60 is over</emphasis>. Compile times will be worse than
61 terrible! Use something like the csh-builtin
62 <command>time</command> command to get a report on how many
63 page faults you're getting.</para>
65 <para>If you don't know what virtual memory, thrashing, and
66 page faults are, or you don't know the memory configuration
67 of your machine, <emphasis>don't</emphasis> try to be clever
68 about memory use: you'll just make your life a misery (and
69 for other people, too, probably).</para>
74 <term>Try to use local disks when linking:</term>
76 <para>Because Haskell objects and libraries tend to be
77 large, it can take many real seconds to slurp the bits
78 to/from a remote filesystem.</para>
80 <para>It would be quite sensible to
81 <emphasis>compile</emphasis> on a fast machine using
82 remotely-mounted disks; then <emphasis>link</emphasis> on a
83 slow machine that had your disks directly mounted.</para>
88 <term>Don't derive/use <function>Read</function> unnecessarily:</term>
90 <para>It's ugly and slow.</para>
95 <term>GHC compiles some program constructs slowly:</term>
97 <para>We'd rather you reported such behaviour as a bug, so
98 that we can try to correct it.</para>
100 <para>To figure out which part of the compiler is badly
102 <option>-v2</option><indexterm><primary><option>-v</option></primary>
103 </indexterm> option is your friend.</para>
110 <title>Faster: producing a program that runs quicker</title>
112 <indexterm><primary>faster programs, how to produce</primary></indexterm>
114 <para>The key tool to use in making your Haskell program run
115 faster are GHC's profiling facilities, described separately in
116 <xref linkend="profiling"/>. There is <emphasis>no
117 substitute</emphasis> for finding where your program's time/space
118 is <emphasis>really</emphasis> going, as opposed to where you
119 imagine it is going.</para>
121 <para>Another point to bear in mind: By far the best way to
122 improve a program's performance <emphasis>dramatically</emphasis>
123 is to use better algorithms. Once profiling has thrown the
124 spotlight on the guilty time-consumer(s), it may be better to
125 re-think your program than to try all the tweaks listed below.</para>
127 <para>Another extremely efficient way to make your program snappy
128 is to use library code that has been Seriously Tuned By Someone
129 Else. You <emphasis>might</emphasis> be able to write a better
130 quicksort than the one in <literal>Data.List</literal>, but it
131 will take you much longer than typing <literal>import
132 Data.List</literal>.</para>
134 <para>Please report any overly-slow GHC-compiled programs. Since
135 GHC doesn't have any credible competition in the performance
136 department these days it's hard to say what overly-slow means, so
137 just use your judgement! Of course, if a GHC compiled program
138 runs slower than the same program compiled with NHC or Hugs, then
139 it's definitely a bug.</para>
143 <term>Optimise, using <option>-O</option> or <option>-O2</option>:</term>
145 <para>This is the most basic way to make your program go
146 faster. Compilation time will be slower, especially with
147 <option>-O2</option>.</para>
149 <para>At present, <option>-O2</option> is nearly
150 indistinguishable from <option>-O</option>.</para>
155 <term>Compile via C and crank up GCC:</term>
157 <para>The native code-generator is designed to be quick, not
158 mind-bogglingly clever. Better to let GCC have a go, as it
159 tries much harder on register allocation, etc.</para>
161 <para>So, when we want very fast code, we use: <option>-O
162 -fvia-C</option>.</para>
167 <term>Overloaded functions are not your friend:</term>
169 <para>Haskell's overloading (using type classes) is elegant,
170 neat, etc., etc., but it is death to performance if left to
171 linger in an inner loop. How can you squash it?</para>
175 <term>Give explicit type signatures:</term>
177 <para>Signatures are the basic trick; putting them on
178 exported, top-level functions is good
179 software-engineering practice, anyway. (Tip: using
180 <option>-fwarn-missing-signatures</option><indexterm><primary>-fwarn-missing-signatures
181 option</primary></indexterm> can help enforce good
182 signature-practice).</para>
184 <para>The automatic specialisation of overloaded
185 functions (with <option>-O</option>) should take care
186 of overloaded local and/or unexported functions.</para>
191 <term>Use <literal>SPECIALIZE</literal> pragmas:</term>
193 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
194 <indexterm><primary>overloading, death to</primary></indexterm>
196 <para>Specialize the overloading on key functions in
197 your program. See <xref linkend="specialize-pragma"/>
198 and <xref linkend="specialize-instance-pragma"/>.</para>
203 <term>“But how do I know where overloading is creeping in?”:</term>
205 <para>A low-tech way: grep (search) your interface
206 files for overloaded type signatures. You can view
207 interface files using the
208 <option>--show-iface</option> option (see <xref
209 linkend="hi-options"/>).
212 % ghc --show-iface Foo.hi | egrep '^[a-z].*::.*=>'
222 <term>Strict functions are your dear friends:</term>
224 <para>and, among other things, lazy pattern-matching is your
227 <para>(If you don't know what a “strict
228 function” is, please consult a functional-programming
229 textbook. A sentence or two of explanation here probably
230 would not do much good.)</para>
232 <para>Consider these two code fragments:
235 f (Wibble x y) = ... # strict
237 f arg = let { (Wibble x y) = arg } in ... # lazy
240 The former will result in far better code.</para>
242 <para>A less contrived example shows the use of
243 <literal>cases</literal> instead of <literal>lets</literal>
244 to get stricter code (a good thing):
247 f (Wibble x y) # beautiful but slow
249 (a1, b1, c1) = unpackFoo x
250 (a2, b2, c2) = unpackFoo y
253 f (Wibble x y) # ugly, and proud of it
254 = case (unpackFoo x) of { (a1, b1, c1) ->
255 case (unpackFoo y) of { (a2, b2, c2) ->
265 <term>GHC loves single-constructor data-types:</term>
267 <para>It's all the better if a function is strict in a
268 single-constructor type (a type with only one
269 data-constructor; for example, tuples are single-constructor
275 <term>Newtypes are better than datatypes:</term>
277 <para>If your datatype has a single constructor with a
278 single field, use a <literal>newtype</literal> declaration
279 instead of a <literal>data</literal> declaration. The
280 <literal>newtype</literal> will be optimised away in most
286 <term>“How do I find out a function's strictness?”</term>
288 <para>Don't guess—look it up.</para>
290 <para>Look for your function in the interface file, then for
291 the third field in the pragma; it should say
292 <literal>__S <string></literal>. The
293 <literal><string></literal> gives the strictness of
294 the function's arguments. <function>L</function> is lazy
295 (bad), <function>S</function> and <function>E</function> are
296 strict (good), <function>P</function> is
297 “primitive” (good), <function>U(...)</function>
298 is strict and “unpackable” (very good), and
299 <function>A</function> is absent (very good).</para>
301 <para>For an “unpackable”
302 <function>U(...)</function> argument, the info inside tells
303 the strictness of its components. So, if the argument is a
304 pair, and it says <function>U(AU(LSS))</function>, that
305 means “the first component of the pair isn't used; the
306 second component is itself unpackable, with three components
307 (lazy in the first, strict in the second \&
308 third).”</para>
310 <para>If the function isn't exported, just compile with the
311 extra flag <option>-ddump-simpl</option>; next to the
312 signature for any binder, it will print the self-same
313 pragmatic information as would be put in an interface file.
314 (Besides, Core syntax is fun to look at!)</para>
319 <term>Force key functions to be <literal>INLINE</literal>d (esp. monads):</term>
321 <para>Placing <literal>INLINE</literal> pragmas on certain
322 functions that are used a lot can have a dramatic effect.
323 See <xref linkend="inline-pragma"/>.</para>
328 <term>Explicit <literal>export</literal> list:</term>
330 <para>If you do not have an explicit export list in a
331 module, GHC must assume that everything in that module will
332 be exported. This has various pessimising effects. For
333 example, if a bit of code is actually
334 <emphasis>unused</emphasis> (perhaps because of unfolding
335 effects), GHC will not be able to throw it away, because it
336 is exported and some other module may be relying on its
339 <para>GHC can be quite a bit more aggressive with pieces of
340 code if it knows they are not exported.</para>
345 <term>Look at the Core syntax!</term>
347 <para>(The form in which GHC manipulates your code.) Just
348 run your compilation with <option>-ddump-simpl</option>
349 (don't forget the <option>-O</option>).</para>
351 <para>If profiling has pointed the finger at particular
352 functions, look at their Core code. <literal>lets</literal>
353 are bad, <literal>cases</literal> are good, dictionaries
354 (<literal>d.<Class>.<Unique></literal>) [or
355 anything overloading-ish] are bad, nested lambdas are
356 bad, explicit data constructors are good, primitive
357 operations (e.g., <literal>eqInt#</literal>) are
363 <term>Use strictness annotations:</term>
365 <para>Putting a strictness annotation ('!') on a constructor
366 field helps in two ways: it adds strictness to the program,
367 which gives the strictness analyser more to work with, and
368 it might help to reduce space leaks.</para>
370 <para>It can also help in a third way: when used with
371 <option>-funbox-strict-fields</option> (see <xref
372 linkend="options-f"/>), a strict field can be unpacked or
373 unboxed in the constructor, and one or more levels of
374 indirection may be removed. Unpacking only happens for
375 single-constructor datatypes (<literal>Int</literal> is a
376 good candidate, for example).</para>
378 <para>Using <option>-funbox-strict-fields</option> is only
379 really a good idea in conjunction with <option>-O</option>,
380 because otherwise the extra packing and unpacking won't be
381 optimised away. In fact, it is possible that
382 <option>-funbox-strict-fields</option> may worsen
383 performance even <emphasis>with</emphasis>
384 <option>-O</option>, but this is unlikely (let us know if it
385 happens to you).</para>
390 <term>Use unboxed types (a GHC extension):</term>
392 <para>When you are <emphasis>really</emphasis> desperate for
393 speed, and you want to get right down to the “raw
394 bits.” Please see <xref linkend="glasgow-unboxed"/> for
395 some information about using unboxed types.</para>
397 <para>Before resorting to explicit unboxed types, try using
398 strict constructor fields and
399 <option>-funbox-strict-fields</option> first (see above).
400 That way, your code stays portable.</para>
405 <term>Use <literal>foreign import</literal> (a GHC extension) to plug into fast libraries:</term>
407 <para>This may take real work, but… There exist piles
408 of massively-tuned library code, and the best thing is not
409 to compete with it, but link with it.</para>
411 <para><xref linkend="ffi"/> describes the foreign function
417 <term>Don't use <literal>Float</literal>s:</term>
419 <para>If you're using <literal>Complex</literal>, definitely
420 use <literal>Complex Double</literal> rather than
421 <literal>Complex Float</literal> (the former is specialised
422 heavily, but the latter isn't).</para>
424 <para><literal>Floats</literal> (probably 32-bits) are
425 almost always a bad idea, anyway, unless you Really Know
426 What You Are Doing. Use <literal>Double</literal>s.
427 There's rarely a speed disadvantage—modern machines
428 will use the same floating-point unit for both. With
429 <literal>Double</literal>s, you are much less likely to hang
430 yourself with numerical errors.</para>
432 <para>One time when <literal>Float</literal> might be a good
433 idea is if you have a <emphasis>lot</emphasis> of them, say
434 a giant array of <literal>Float</literal>s. They take up
435 half the space in the heap compared to
436 <literal>Doubles</literal>. However, this isn't true on a
437 64-bit machine.</para>
442 <term>Use unboxed arrays (<literal>UArray</literal>)</term>
444 <para>GHC supports arrays of unboxed elements, for several
445 basic arithmetic element types including
446 <literal>Int</literal> and <literal>Char</literal>: see the
447 <literal>Data.Array.Unboxed</literal> library for details.
448 These arrays are likely to be much faster than using
449 standard Haskell 98 arrays from the
450 <literal>Data.Array</literal> library.</para>
455 <term>Use a bigger heap!</term>
457 <para>If your program's GC stats
458 (<option>-S</option><indexterm><primary>-S RTS
459 option</primary></indexterm> RTS option) indicate that it's
460 doing lots of garbage-collection (say, more than 20%
461 of execution time), more memory might help—with the
462 <option>-M<size></option><indexterm><primary>-M<size>
463 RTS option</primary></indexterm> or
464 <option>-A<size></option><indexterm><primary>-A<size>
465 RTS option</primary></indexterm> RTS options (see <xref
466 linkend="rts-options-gc"/>).</para>
474 <title>Smaller: producing a program that is smaller
478 <indexterm><primary>smaller programs, how to produce</primary></indexterm>
482 Decrease the “go-for-it” threshold for unfolding smallish
484 <option>-funfolding-use-threshold0</option><indexterm><primary>-funfolding-use-threshold0
485 option</primary></indexterm> option for the extreme case. (“Only unfoldings with
486 zero cost should proceed.”) Warning: except in certain specialised
487 cases (like Happy parsers) this is likely to actually
488 <emphasis>increase</emphasis> the size of your program, because unfolding
489 generally enables extra simplifying optimisations to be performed.
493 Avoid <function>Read</function>.
497 Use <literal>strip</literal> on your executables.
502 <sect1 id="thriftier">
503 <title>Thriftier: producing a program that gobbles less heap space
507 <indexterm><primary>memory, using less heap</primary></indexterm>
508 <indexterm><primary>space-leaks, avoiding</primary></indexterm>
509 <indexterm><primary>heap space, using less</primary></indexterm>
513 “I think I have a space leak…” Re-run your program
514 with <option>+RTS -S</option>, and remove all doubt! (You'll
515 see the heap usage get bigger and bigger…)
516 [Hmmm…this might be even easier with the
517 <option>-G1</option> RTS option; so… <command>./a.out +RTS
519 <indexterm><primary>-G RTS option</primary></indexterm>
520 <indexterm><primary>-S RTS option</primary></indexterm>
524 Once again, the profiling facilities (<xref linkend="profiling"/>) are
525 the basic tool for demystifying the space behaviour of your program.
529 Strict functions are good for space usage, as they are for time, as
530 discussed in the previous section. Strict functions get right down to
531 business, rather than filling up the heap with closures (the system's
532 notes to itself about how to evaluate something, should it eventually
541 ;;; Local Variables: ***
543 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***