Next Previous Contents

4. Advice on: sooner, faster, smaller, stingier

Please advise us of other ``helpful hints'' that should go here!

4.1 Sooner: producing a program more quickly

Don't use -O or (especially) -O2:

By using them, you are telling GHC that you are willing to suffer longer compilation times for better-quality code.

GHC is surprisingly zippy for normal compilations without -O!

Use more memory:

Within reason, more memory for heap space means less garbage collection for GHC, which means less compilation time. If you use the -Rgc-stats option, you'll get a garbage-collector report. (Again, you can use the cheap-and-nasty -optCrts-Sstderr option to send the GC stats straight to standard error.)

If it says you're using more than 20% of total time in garbage collecting, then more memory would help.

If the heap size is approaching the maximum (64M by default), and you have lots of memory, try increasing the maximum with the -M<size> option, e.g.: ghc -c -O -M16m Foo.hs.

Increasing the default allocation area size used by the compiler's RTS might also help: use the -A<size> option.

If GHC persists in being a bad memory citizen, please report it as a bug.

Don't use too much memory!

As soon as GHC plus its ``fellow citizens'' (other processes on your machine) start using more than the real memory on your machine, and the machine starts ``thrashing,'' the party is over. Compile times will be worse than terrible! Use something like the csh-builtin time command to get a report on how many page faults you're getting.

If you don't know what virtual memory, thrashing, and page faults are, or you don't know the memory configuration of your machine, don't try to be clever about memory use: you'll just make your life a misery (and for other people, too, probably).

Try to use local disks when linking:

Because Haskell objects and libraries tend to be large, it can take many real seconds to slurp the bits to/from a remote filesystem.

It would be quite sensible to compile on a fast machine using remotely-mounted disks; then link on a slow machine that had your disks directly mounted.

Don't derive/use Read unnecessarily:

It's ugly and slow.

GHC compiles some program constructs slowly:

Deeply-nested list comprehensions seem to be one such; in the past, very large constant tables were bad, too.

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

The parts of the compiler that seem most prone to wandering off for a long time are the abstract interpreters (strictness and update analysers). You can turn these off individually with -fno-strictness and -fno-update-analysis.

To figure out which part of the compiler is badly behaved, the -dshow-passes option is your friend.

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 -fvia-C (not that GCC will be that quick about it, either).

Avoid the consistency-check on linking:

Use -no-link-chk ; saves effort. This is probably safe in a I-only-compile-things-one-way setup.

Explicit import declarations:

Instead of saying import Foo, say import Foo (...stuff I want...).

Truthfully, the reduction on compilation time will be very small. However, judicious use of import declarations can make a program easier to understand, so it may be a good idea anyway.

4.2 Faster: producing a program that runs quicker

The key tool to use in making your Haskell program run faster are GHC's profiling facilities, described separately in Section Profiling. There is no substitute for finding where your program's time/space is really going, as opposed to where you imagine it is going.

Another point to bear in mind: By far the best way to improve a program's performance dramatically 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.

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

Please report any overly-slow GHC-compiled programs. The current definition of ``overly-slow'' is ``the HBC-compiled version ran faster''...

Optimise, using -O or -O2:

This is the most basic way to make your program go faster. Compilation time will be slower, especially with -O2.

At present, -O2 is nearly indistinguishable from -O.

Compile via C and crank up GCC:

Even with -O, GHC tries to use a native-code generator, if available. But 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.

So, when we want very fast code, we use: -O -fvia-C -O2-for-C.

Overloaded functions are not your friend:

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?

Give explicit type signatures:

Signatures are the basic trick; putting them on exported, top-level functions is good software-engineering practice, anyway. (Tip: using -fwarn-missing-signatures can help enforce good signature-practice).

The automatic specialisation of overloaded functions (with -O) should take care of overloaded local and/or unexported functions.

Use SPECIALIZE pragmas:

(UK spelling also accepted.) For key overloaded functions, you can create extra versions (NB: more code space) specialised to particular types. Thus, if you have an overloaded function:

hammeredLookup :: Ord key => [(key, value)] -> key -> value

If it is heavily used on lists with Widget keys, you could specialise it as follows:

{-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}

To get very fancy, you can also specify a named function to use for the specialised value, by adding = blah, as in:

{-# SPECIALIZE hammeredLookup :: before... = blah #-}
It's Your Responsibility to make sure that blah really behaves as a specialised version of hammeredLookup!!!

[NOTE: this feature isn't implemented in GHC 4.00... ]

An example in which the = blah form will Win Big:

toDouble :: Real a => a -> Double
toDouble = fromRational . toRational

{-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
The i2d function is virtually one machine instruction; the default conversion---via an intermediate Rational---is obscenely expensive by comparison.

By using the US spelling, your SPECIALIZE pragma will work with HBC, too. Note that HBC doesn't support the = blah form.

A SPECIALIZE pragma for a function can be put anywhere its type signature could be put.

Use SPECIALIZE instance pragmas:

Same idea, except for instance declarations. For example:

instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }

{-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
Compatible with HBC, by the way.

``But how do I know where overloading is creeping in?'':

A low-tech way: grep (search) your interface files for overloaded type signatures; e.g.,:

% egrep '^[a-z].*::.*=>' *.hi

Strict functions are your dear friends:

and, among other things, lazy pattern-matching is your enemy.

(If you don't know what a ``strict function'' is, please consult a functional-programming textbook. A sentence or two of explanation here probably would not do much good.)

Consider these two code fragments:

f (Wibble x y) =  ... # strict

f arg = let { (Wibble x y) = arg } in ... # lazy
The former will result in far better code.

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

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) ->
    case (unpackFoo y) of { (a2, b2, c2) ->

GHC loves single-constructor data-types:

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).

Newtypes are better than datatypes:

If your datatype has a single constructor with a single field, use a newtype declaration instead of a data declaration. The newtype will be optimised away in most cases.

``How do I find out a function's strictness?''

Don't guess---look it up.

Look for your function in the interface file, then for the third field in the pragma; it should say _S_ <string>. The <string> gives the strictness of the function's arguments. L is lazy (bad), S and E are strict (good), P is ``primitive'' (good), U(...) is strict and ``unpackable'' (very good), and A is absent (very good).

For an ``unpackable'' U(...) argument, the info inside tells the strictness of its components. So, if the argument is a pair, and it says U(AU(LSS)), that means ``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 \& third).''

If the function isn't exported, just compile with the extra flag -ddump-simpl; 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!)

Force key functions to be INLINEd (esp. monads):

GHC (with -O, as always) tries to inline (or ``unfold'') functions/values that are ``small enough,'' thus avoiding the call overhead and possibly exposing other more-wonderful optimisations.

You will probably see these unfoldings (in Core syntax) in your interface files.

Normally, if GHC decides a function is ``too expensive'' to inline, it will not do so, nor will it export that unfolding for other modules to use.

The sledgehammer you can bring to bear is the INLINE pragma, used thusly:

key_function :: Int -> String -> (Bool, Double) 

{-# INLINE key_function #-}
(You don't need to do the C pre-processor carry-on unless you're going to stick the code through HBC---it doesn't like INLINE pragmas.)

The major effect of an INLINE pragma is to declare a function's ``cost'' to be very low. The normal unfolding machinery will then be very keen to inline it.

An INLINE pragma for a function can be put anywhere its type signature could be put.

INLINE pragmas are a particularly good idea for the then/return (or bind/unit) functions in a monad. For example, in GHC's own UniqueSupply monad code, we have:

{-# INLINE thenUs #-}
{-# INLINE returnUs #-}

Incedentally, there's also a NOINLINE pragma which does the obvious thing.

Explicit export list:

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 unused (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.

GHC can be quite a bit more aggressive with pieces of code if it knows they are not exported.

Look at the Core syntax!

(The form in which GHC manipulates your code.) Just run your compilation with -ddump-simpl (don't forget the -O).

If profiling has pointed the finger at particular functions, look at their Core code. lets are bad, cases are good, dictionaries (d.<Class>.<Unique>) [or anything overloading-ish] are bad, nested lambdas are bad, explicit data constructors are good, primitive operations (e.g., eqInt#) are good, ...

Use unboxed types (a GHC extension):

When you are really desperate for speed, and you want to get right down to the ``raw bits.'' Please see Section Unboxed types for some information about using unboxed types.

Use _ccall_s (a GHC extension) to plug into fast libraries:

This may take real work, but... There exist piles of massively-tuned library code, and the best thing is not to compete with it, but link with it.

Section Calling~C directly from Haskell says a little about how to use C calls.

Don't use Floats:

We don't provide specialisations of Prelude functions for Float (but we do for Double). If you end up executing overloaded code, you will lose on performance, perhaps badly.

Floats (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---modern machines will use the same floating-point unit for both. With Doubles, you are much less likely to hang yourself with numerical errors.

One time when Float might be a good idea is if you have a lot of them, say a giant array of Floats. They take up half the space in the heap compared to Doubles. However, this isn't true on a 64-bit machine.

Use a bigger heap!

If your program's GC stats (-S RTS option) indicate that it's doing lots of garbage-collection (say, more than 20% of execution time), more memory might help---with the -M<size> or -A<size> RTS options (see Section RTS options to control the garbage-collector).

4.3 Smaller: producing a program that is smaller

Decrease the ``go-for-it'' threshold for unfolding smallish expressions. Give a -funfolding-use-threshold0 option for the extreme case. (``Only unfoldings with zero cost should proceed.'') Warning: except in certain specialiised cases (like Happy parsers) this is likely to actually increase the size of your program, because unfolding generally enables extra simplifying optimisations to be performed.

Avoid Read.

Use strip on your executables.

4.4 Stingier: producing a program that gobbles less heap space

``I think I have a space leak...'' Re-run your program with +RTS -Sstderr, and remove all doubt! (You'll see the heap usage get bigger and bigger...) [Hmmm... this might be even easier with the -F2s RTS option; so... ./a.out +RTS -Sstderr -F2s...]

Once again, the profiling facilities (Section Profiling) are the basic tool for demystifying the space behaviour of your program.

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).

Next Previous Contents