[ghc-hetmet.git] / ghc / docs / users_guide / sooner.sgml

<Chapter id="sooner-faster-quicker">
<Title>Advice on: sooner, faster, smaller, thriftier
</Title>

<Para>
Please advise us of other &ldquo;helpful hints&rdquo; that should go here!
</Para>

<Sect1 id="sooner">
<Title>Sooner: producing a program more quickly
</Title>

<Para>
<IndexTerm><Primary>compiling faster</Primary></IndexTerm>
<IndexTerm><Primary>faster compiling</Primary></IndexTerm>
<VariableList>

<VarListEntry>
<Term>Don't use <Option>-O</Option> or (especially) <Option>-O2</Option>:</Term>
<ListItem>
<Para>
By using them, you are telling GHC that you are willing to suffer
longer compilation times for better-quality code.
</Para>

<Para>
GHC is surprisingly zippy for normal compilations without <Option>-O</Option>!
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Use more memory:</Term>
<ListItem>
<Para>
Within reason, more memory for heap space means less garbage
collection for GHC, which means less compilation time.  If you use the
<Option>-Rghc-timing</Option> option, you'll get a garbage-collector
report.  (Again, you can use the cheap-and-nasty <Option>+RTS -Sstderr
-RTS</Option> option to send the GC stats straight to standard error.)
</Para>

<Para>
If it says you're using more than 20&percnt; of total time in garbage
collecting, then more memory would help.
</Para>

<Para>
If the heap size is approaching the maximum (64M by default), and you
have lots of memory, try increasing the maximum with the
<Option>-M&lt;size&gt;</Option><IndexTerm><Primary>-M&lt;size&gt; option</Primary></IndexTerm> option, e.g.: <Command>ghc -c -O
-M1024m Foo.hs</Command>.
</Para>

<Para>
Increasing the default allocation area size used by the compiler's RTS
might also help: use the <Option>-A&lt;size&gt;</Option><IndexTerm><Primary>-A&lt;size&gt; option</Primary></IndexTerm>
option.
</Para>

<Para>
If GHC persists in being a bad memory citizen, please report it as a
bug.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Don't use too much memory!</Term>
<ListItem>
<Para>
As soon as GHC plus its &ldquo;fellow citizens&rdquo; (other processes on your
machine) start using more than the <Emphasis>real memory</Emphasis> on your
machine, and the machine starts &ldquo;thrashing,&rdquo; <Emphasis>the party is
over</Emphasis>.  Compile times will be worse than terrible!  Use something
like the csh-builtin <Command>time</Command> command to get a report on how many page
faults you're getting.
</Para>

<Para>
If you don't know what virtual memory, thrashing, and page faults are,
or you don't know the memory configuration of your machine,
<Emphasis>don't</Emphasis> try to be clever about memory use: you'll just make
your life a misery (and for other people, too, probably).
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Try to use local disks when linking:</Term>
<ListItem>
<Para>
Because Haskell objects and libraries tend to be large, it can take
many real seconds to slurp the bits to/from a remote filesystem.
</Para>

<Para>
It would be quite sensible to <Emphasis>compile</Emphasis> on a fast machine using
remotely-mounted disks; then <Emphasis>link</Emphasis> on a slow machine that had
your disks directly mounted.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Don't derive/use <Function>Read</Function> unnecessarily:</Term>
<ListItem>
<Para>
It's ugly and slow.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>GHC compiles some program constructs slowly:</Term>
<ListItem>
<Para>
Deeply-nested list comprehensions seem to be one such; in the past,
very large constant tables were bad, too.
</Para>

<Para>
We'd rather you reported such behaviour as a bug, so that we can try
to correct it.
</Para>

<Para>
The part of the compiler that is occasionally prone to wandering off
for a long time is the strictness analyser.  You can turn this off
individually with <Option>-fno-strictness</Option>.
<IndexTerm><Primary>-fno-strictness anti-option</Primary></IndexTerm>
</Para>

<Para>
To figure out which part of the compiler is badly behaved, the
<option>-v2</option><indexterm><primary><option>-v</option></primary>
</indexterm> option is your friend.
</Para>

<Para>
If your module has big wads of constant data, GHC may produce a huge
basic block that will cause the native-code generator's register
allocator to founder.  Bring on <Option>-fvia-C</Option><IndexTerm><Primary>-fvia-C option</Primary></IndexTerm>
(not that GCC will be that quick about it, either).
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Explicit <Literal>import</Literal> declarations:</Term>
<ListItem>
<Para>
Instead of saying <Literal>import Foo</Literal>, say <Literal>import
Foo (...stuff I want...)</Literal> You can get GHC to tell you the
minimal set of required imports by using the
<option>-ddump-minimal-imports</option> option (see <xref
linkend="hi-options">).
</Para>

<Para>
Truthfully, the reduction on compilation time will be very small.
However, judicious use of <Literal>import</Literal> declarations can make a
program easier to understand, so it may be a good idea anyway.
</Para>
</ListItem>
</VarListEntry>
</VariableList>
</Para>

</Sect1>

<Sect1 id="faster">
<Title>Faster: producing a program that runs quicker
</Title>

<Para>
<IndexTerm><Primary>faster programs, how to produce</Primary></IndexTerm>
</Para>

<Para>
The key tool to use in making your Haskell program run faster are
GHC's profiling facilities, described separately in <XRef LinkEnd="profiling">.  There is <Emphasis>no substitute</Emphasis> for
finding where your program's time/space is <Emphasis>really</Emphasis> going, as
opposed to where you imagine it is going.
</Para>

<Para>
Another point to bear in mind: By far the best way to improve a
program's performance <Emphasis>dramatically</Emphasis> is to use better
algorithms.  Once profiling has thrown the spotlight on the guilty
time-consumer(s), it may be better to re-think your program than to
try all the tweaks listed below.
</Para>

<Para>
Another extremely efficient way to make your program snappy is to use
library code that has been Seriously Tuned By Someone Else.  You
<Emphasis>might</Emphasis> be able to write a better quicksort than the one in the
HBC library, but it will take you much longer than typing <Literal>import
QSort</Literal>.  (Incidentally, it doesn't hurt if the Someone Else is Lennart
Augustsson.)
</Para>

<Para>
Please report any overly-slow GHC-compiled programs.  The current
definition of &ldquo;overly-slow&rdquo; is &ldquo;the HBC-compiled version ran
faster&rdquo;&hellip;
</Para>

<Para>
<VariableList>

<VarListEntry>
<Term>Optimise, using <Option>-O</Option> or <Option>-O2</Option>:</Term>
<ListItem>
<Para>
This is the most basic way
to make your program go faster.  Compilation time will be slower,
especially with <Option>-O2</Option>.
</Para>

<Para>
At present, <Option>-O2</Option> is nearly indistinguishable from <Option>-O</Option>.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Compile via C and crank up GCC:</Term>
<ListItem>
<Para>
The native code-generator is designed to be quick, not mind-bogglingly
clever.  Better to let GCC have a go, as it tries much harder on
register allocation, etc.</para>

<para>At the moment, if you turn on <option>-O</option> you get GCC
instead.  This may change in the future.</para>

<Para>
So, when we want very fast code, we use: <Option>-O -fvia-C</Option>.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Overloaded functions are not your friend:</Term>
<ListItem>
<Para>
Haskell's overloading (using type classes) is elegant, neat, etc.,
etc., but it is death to performance if left to linger in an inner
loop.  How can you squash it?
</Para>

<Para>
<VariableList>

<VarListEntry>
<Term>Give explicit type signatures:</Term>
<ListItem>
<Para>
Signatures are the basic trick; putting them on exported, top-level
functions is good software-engineering practice, anyway.  (Tip: using
<Option>-fwarn-missing-signatures</Option><IndexTerm><Primary>-fwarn-missing-signatures
option</Primary></IndexTerm> can help enforce good signature-practice).
</Para>

<Para>
The automatic specialisation of overloaded functions (with <Option>-O</Option>)
should take care of overloaded local and/or unexported functions.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Use <Literal>SPECIALIZE</Literal> pragmas:</Term>
<ListItem>
<Para>
<IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
<IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
</Para>

<Para>
Specialize the overloading on key functions in your program.  See
<XRef LinkEnd="specialize-pragma"> and
<XRef LinkEnd="specialize-instance-pragma">.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>&ldquo;But how do I know where overloading is creeping in?&rdquo;:</Term>
<ListItem>
<Para>
A low-tech way: grep (search) your interface files for overloaded
type signatures; e.g.,:

<ProgramListing>
% egrep '^[a-z].*::.*=&#62;' *.hi
</ProgramListing>

</Para>
</ListItem>
</VarListEntry>
</VariableList>
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Strict functions are your dear friends:</Term>
<ListItem>
<Para>
and, among other things, lazy pattern-matching is your enemy.
</Para>

<Para>
(If you don't know what a &ldquo;strict function&rdquo; is, please consult a
functional-programming textbook.  A sentence or two of
explanation here probably would not do much good.)
</Para>

<Para>
Consider these two code fragments:

<ProgramListing>
f (Wibble x y) =  ... # strict

f arg = let { (Wibble x y) = arg } in ... # lazy
</ProgramListing>

The former will result in far better code.
</Para>

<Para>
A less contrived example shows the use of <Literal>cases</Literal> instead
of <Literal>lets</Literal> to get stricter code (a good thing):

<ProgramListing>
f (Wibble x y)  # beautiful but slow
  = let
        (a1, b1, c1) = unpackFoo x
        (a2, b2, c2) = unpackFoo y
    in ...

f (Wibble x y)  # ugly, and proud of it
  = case (unpackFoo x) of { (a1, b1, c1) -&#62;
    case (unpackFoo y) of { (a2, b2, c2) -&#62;
    ...
    }}
</ProgramListing>

</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>GHC loves single-constructor data-types:</Term>
<ListItem>
<Para>
It's all the better if a function is strict in a single-constructor
type (a type with only one data-constructor; for example, tuples are
single-constructor types).
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Newtypes are better than datatypes:</Term>
<ListItem>
<Para>
If your datatype has a single constructor with a single field, use a
<Literal>newtype</Literal> declaration instead of a <Literal>data</Literal> declaration.  The <Literal>newtype</Literal>
will be optimised away in most cases.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>&ldquo;How do I find out a function's strictness?&rdquo;</Term>
<ListItem>
<Para>
Don't guess&mdash;look it up.
</Para>

<Para>
Look for your function in the interface file, then for the third field
in the pragma; it should say <Literal>&lowbar;&lowbar;S
&lt;string&gt;</Literal>.  The <Literal>&lt;string&gt;</Literal> gives
the strictness of the function's arguments.  <Function>L</Function> is
lazy (bad), <Function>S</Function> and <Function>E</Function> are
strict (good), <Function>P</Function> is &ldquo;primitive&rdquo;
(good), <Function>U(...)</Function> is strict and
&ldquo;unpackable&rdquo; (very good), and <Function>A</Function> is
absent (very good).
</Para>

<Para>
For an &ldquo;unpackable&rdquo; <Function>U(...)</Function> argument, the info inside
tells the strictness of its components.  So, if the argument is a
pair, and it says <Function>U(AU(LSS))</Function>, that means &ldquo;the first component of the
pair isn't used; the second component is itself unpackable, with three
components (lazy in the first, strict in the second \&#38; third).&rdquo;
</Para>

<Para>
If the function isn't exported, just compile with the extra flag <Option>-ddump-simpl</Option>;
next to the signature for any binder, it will print the self-same
pragmatic information as would be put in an interface file.
(Besides, Core syntax is fun to look at!)
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Force key functions to be <Literal>INLINE</Literal>d (esp. monads):</Term>
<ListItem>
<Para>
Placing <Literal>INLINE</Literal> pragmas on certain functions that are used a lot can
have a dramatic effect.  See <XRef LinkEnd="inline-pragma">.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Explicit <Literal>export</Literal> list:</Term>
<ListItem>
<Para>
If you do not have an explicit export list in a module, GHC must
assume that everything in that module will be exported.  This has
various pessimising effects.  For example, if a bit of code is actually
<Emphasis>unused</Emphasis> (perhaps because of unfolding effects), GHC will not be
able to throw it away, because it is exported and some other module
may be relying on its existence.
</Para>

<Para>
GHC can be quite a bit more aggressive with pieces of code if it knows
they are not exported.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Look at the Core syntax!</Term>
<ListItem>
<Para>
(The form in which GHC manipulates your code.)  Just run your
compilation with <Option>-ddump-simpl</Option> (don't forget the <Option>-O</Option>).
</Para>

<Para>
If profiling has pointed the finger at particular functions, look at
their Core code.  <Literal>lets</Literal> are bad, <Literal>cases</Literal> are good, dictionaries
(<Literal>d.&lt;Class&gt;.&lt;Unique&gt;</Literal>) &lsqb;or anything overloading-ish&rsqb; are bad,
nested lambdas are bad, explicit data constructors are good, primitive
operations (e.g., <Literal>eqInt&num;</Literal>) are good,&hellip;
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Use unboxed types (a GHC extension):</Term>
<ListItem>
<Para>
When you are <Emphasis>really</Emphasis> desperate for speed, and you want to get
right down to the &ldquo;raw bits.&rdquo;  Please see <XRef LinkEnd="glasgow-unboxed"> for some information about using unboxed
types.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Use <Literal>foreign import</Literal> (a GHC extension) to plug into fast libraries:</Term>
<ListItem>
<Para>
This may take real work, but&hellip; There exist piles of
massively-tuned library code, and the best thing is not
to compete with it, but link with it.
</Para>

<Para>
<XRef LinkEnd="ffi"> describes the foreign function interface.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Don't use <Literal>Float</Literal>s:</Term>
<ListItem>
<Para>
We don't provide specialisations of Prelude functions for <Literal>Float</Literal>
(but we do for <Literal>Double</Literal>).  If you end up executing overloaded
code, you will lose on performance, perhaps badly.
</Para>

<Para>
<Literal>Floats</Literal> (probably 32-bits) are almost always a bad idea, anyway,
unless you Really Know What You Are Doing.  Use Doubles.  There's
rarely a speed disadvantage&mdash;modern machines will use the same
floating-point unit for both.  With <Literal>Doubles</Literal>, you are much less
likely to hang yourself with numerical errors.
</Para>

<Para>
One time when <Literal>Float</Literal> might be a good idea is if you have a
<Emphasis>lot</Emphasis> of them, say a giant array of <Literal>Float</Literal>s.  They take up
half the space in the heap compared to <Literal>Doubles</Literal>.  However, this isn't
true on a 64-bit machine.
</Para>
</ListItem>
</VarListEntry>
<VarListEntry>
<Term>Use a bigger heap!</Term>
<ListItem>
<Para>
If your program's GC stats (<Option>-S</Option><IndexTerm><Primary>-S RTS option</Primary></IndexTerm> RTS option)
indicate that it's doing lots of garbage-collection (say, more than
20&percnt; of execution time), more memory might help&mdash;with the
<Option>-M&lt;size&gt;</Option><IndexTerm><Primary>-M&lt;size&gt; RTS option</Primary></IndexTerm> or
<Option>-A&lt;size&gt;</Option><IndexTerm><Primary>-A&lt;size&gt; RTS option</Primary></IndexTerm> RTS options (see
<XRef LinkEnd="rts-options-gc">).
</Para>
</ListItem>
</VarListEntry>
</VariableList>
</Para>

</Sect1>

<Sect1 id="smaller">
<Title>Smaller: producing a program that is smaller
</Title>

<Para>
<IndexTerm><Primary>smaller programs, how to produce</Primary></IndexTerm>
</Para>

<Para>
Decrease the &ldquo;go-for-it&rdquo; threshold for unfolding smallish
expressions.  Give a
<Option>-funfolding-use-threshold0</Option><IndexTerm><Primary>-funfolding-use-threshold0
option</Primary></IndexTerm> option for the extreme case. (&ldquo;Only unfoldings with
zero cost should proceed.&rdquo;)  Warning: except in certain specialised
cases (like Happy parsers) this is likely to actually
<Emphasis>increase</Emphasis> the size of your program, because unfolding
generally enables extra simplifying optimisations to be performed.
</Para>

<Para>
Avoid <Function>Read</Function>.
</Para>

<Para>
Use <Literal>strip</Literal> on your executables.
</Para>

</Sect1>

<Sect1 id="thriftier">
<Title>Thriftier: producing a program that gobbles less heap space
</Title>

<Para>
<IndexTerm><Primary>memory, using less heap</Primary></IndexTerm>
<IndexTerm><Primary>space-leaks, avoiding</Primary></IndexTerm>
<IndexTerm><Primary>heap space, using less</Primary></IndexTerm>
</Para>

<Para>
&ldquo;I think I have a space leak&hellip;&rdquo; Re-run your program
with <Option>+RTS -Sstderr</Option>, and remove all doubt!  (You'll
see the heap usage get bigger and bigger&hellip;)
&lsqb;Hmmm&hellip;this might be even easier with the
<Option>-G1</Option> RTS option; so&hellip; <Command>./a.out +RTS
-Sstderr -G1</Command>...]
<IndexTerm><Primary>-G RTS option</Primary></IndexTerm>
<IndexTerm><Primary>-Sstderr RTS option</Primary></IndexTerm>
</Para>

<Para>
Once again, the profiling facilities (<XRef LinkEnd="profiling">) are
the basic tool for demystifying the space behaviour of your program.
</Para>

<Para>
Strict functions are good for space usage, as they are for time, as
discussed in the previous section.  Strict functions get right down to
business, rather than filling up the heap with closures (the system's
notes to itself about how to evaluate something, should it eventually
be required).
</Para>

</Sect1>

</Chapter>

<!-- Emacs stuff:
     ;;; Local Variables: ***
     ;;; mode: sgml ***
     ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter") ***
     ;;; End: ***
 -->