Difference between revisions of "GHC/Data Parallel Haskell"

From HaskellWiki
< GHC
Jump to: navigation, search
(Data Parallel Haskell)
m (Data Parallel Haskell)
 
(71 intermediate revisions by 7 users not shown)
Line 2: Line 2:
 
== Data Parallel Haskell ==
 
== Data Parallel Haskell ==
   
''Data Parallel Haskell'' is the codename for an extension to the Glasgow Haskell Compiler and its libraries to support [http://www.cs.cmu.edu/~scandal/cacm/cacm2.html nested data parallelism] with a focus to utilise multi-core CPUs. Nested data parallelism extends the programming model of flat data parallelism, as known from parallel Fortran dialects, to irregular parallel computations (such as divide-and-conquer algorithms) and irregular data structures (such as sparse matrices and tree structures). An introduction to nested data parallelism in Haskell, including some examples, can be found in the paper [http://www.cse.unsw.edu.au/~chak/papers/papers.html#ndp-haskell Nepal -- Nested Data-Parallelism in Haskell].
 
  +
:''Searching for Parallel Haskell? DPH is a fantastic effort, but it's not the only way to do parallelism in Haskell. Try the [[Parallel|Parallel Haskell portal]] for a more general view.''
  +
  +
''Data Parallel Haskell'' is the codename for an extension to the Glasgow Haskell Compiler and its libraries to support [http://www.cs.cmu.edu/~scandal/cacm/cacm2.html nested data parallelism] with a focus to utilise multicore CPUs. Nested data parallelism extends the programming model of flat data parallelism, as known from parallel Fortran dialects, to irregular parallel computations (such as divide-and-conquer algorithms) and irregular data structures (such as sparse matrices and tree structures). An introduction to nested data parallelism in Haskell, including some examples, can be found in the paper [http://www.cse.unsw.edu.au/~chak/papers/papers.html#ndp-haskell Nepal – Nested Data-Parallelism in Haskell].
  +
  +
<center>
  +
http://17.media.tumblr.com/VtG26AnzIklk0sh6YkZSLYNPo1_400.png
  +
</center>
  +
  +
''This is the performance of a dot product of two vectors of 10 million doubles each using Data Parallel Haskell. Both machines have 8 cores. Each core of the T2 has 8 hardware thread contexts. ''
  +
  +
__TOC__
  +
  +
  +
   
 
=== Project status ===
 
=== Project status ===
   
A first ''technology preview'' of Data Parallel Haskell (DPH) is included in the 6.10.1 release of GHC. All major components of DPH are implemented, including code vectorisation and parallel execution on multicore systems. However, the implementation has many limitations and probably also many bugs. Major limitations include the inability to mix vectorised and non-vectorised code in a single Haskell module, the need to use a feature-deprived, special-purpose Prelude in vectorised code, and a lack of optimisations (leading to poor performance).
+
Data Parallel Haskell (DPH) is available as an add-on for [http://haskell.org/ghc/download_ghc_7_4_1 GHC 7.4] in the form of a few separate cabal package. All major components of DPH are implemented, including code vectorisation and parallel execution on multicore systems. However, the implementation has many limitations and probably also many bugs. Major limitations include the inability to mix vectorised and non-vectorised code in a single Haskell module, the need to use a feature-deprived, special-purpose Prelude for vectorised code, and a lack of optimisations (leading to poor performance in some cases).
  +
  +
The current implementation should work well for code with nested parallelism, where the depth of nesting is statically fixed or no user-defined nested-parallel datatypes are used. Support for user-defined nested-parallel datatypes is still rather experimental and will likely result in inefficient code.
  +
  +
DPH focuses on irregular data parallelism. For regular data parallel code in Haskell, please consider using the companion library [http://repa.ouroborus.net/ Repa], which builds on the parallel array infrastructure of DPH.
  +
  +
'''Note:''' This page describes version 0.6.* of the DPH libraries. We only support this version of DPH as well as the current development version.
  +
  +
'''Disclaimer:''' Data Parallel Haskell is very much '''work in progress.''' Some components are already usable, and we explain here how to use them. However, please be aware that APIs are still in flux and functionality may change during development.
  +
  +
=== Where to get it ===
  +
  +
To get DPH, install [http://haskell.org/ghc/download_ghc_7_4_1 GHC 7.4] and then install the DPH libraries with <code>cabal install</code> as follows:
  +
<blockquote>
  +
<code>$ cabal update</code><br>
  +
<code>$ cabal install dph-examples</code>
  +
</blockquote>
  +
This will install all DPH packages, including a set of simple examples, see [http://hackage.haskell.org/package/dph-examples dph-examples]. (The package [http://hackage.haskell.org/package/dph-examples dph-examples] does depend on OpenGL and Gloss as both are used in a visualiser for the nobody example.)
  +
  +
'''WARNING:''' The vanilla GHC distribution does '''not''' include <code>cabal install</code>. This is in contrast to the Haskell Platform, which does include <code>cabal install</code>. If you want to avoid installing the <code>cabal-intstall</code> package and its dependencies explicitly, simply install GHC 7.4.1 in addition to your current Haskell Platform installation. (How to do that depends on your platform and personal preferences. One option is to install a bindist into your home directory with symbolic links to the binaries including the version number.) Then, install DPH with the following command:
  +
<blockquote>
  +
<code>cabal install --with-compiler=`which ghc-7.4.1` --with-hc-pkg=`which ghc-pkg-7.4.1` dph-examples</code>
  +
</blockquote>
   
 
=== Overview ===
 
=== Overview ===
   
The design, some implementation details, and first results of Data Parallel Haskell are described in the paper [http://www.cse.unsw.edu.au/~chak/papers/CLPKM06.html Data Parallel Haskell: a status report]. The same topic is covered in the slides for the two talks [http://research.microsoft.com/~simonpj/papers/ndp/NdpSlides.pdf Nested data parallelism in Haskell] and [http://dataparallel.googlegroups.com/web/UNSW%20CGO%20DP%202007.pdf Compiling nested data parallelism by program transformation].
 
  +
From a user's point of view, Data Parallel Haskell adds a new data type to Haskell –namely, ''parallel arrays''– as well as operations on parallel arrays. Syntactically, parallel arrays are like lists, only that instead of square brackets <hask>[</hask> and <hask>]</hask>, parallel arrays use square brackets with a colon <hask>[:</hask> and <hask>:]</hask>. In particular, <hask>[:e:]</hask> is the type of parallel arrays with elements of type <hask>e</hask>; the expression <hask>[:x, y, z:]</hask> denotes a three element parallel array with elements <hask>x</hask>, <hask>y</hask>, and <hask>z</hask>; and <hask>[:x + 1 | x <- xs:]</hask> represents a simple array comprehension. More sophisticated array comprehensions (including the equivalent of [http://www.haskell.org/ghc/docs/latest/html/users_guide/syntax-extns.html#parallel-list-comprehensions parallel list comprehensions]) as well as enumerations and pattern matching proceed in an analog manner. Moreover, the array library of DPH defines variants of most list operations from the Haskell Prelude and the standard <hask>List</hask> library (e.g., we have <hask>lengthP</hask>, <hask>sumP</hask>, <hask>mapP</hask>, and so on).
   
Data Parallel Haskell essentially extends GHC in two ways:
 
  +
The two main differences between lists and parallel arrays are that (1) parallel arrays are a strict data structure and (2) that they are not inductively defined. Parallel arrays are strict in that by using a single element, all elements of an array are demanded. Hence, all elements of a parallel array might be evaluated in parallel. To facilitate such parallel evaluation, operations on parallel arrays should treat arrays as aggregate structures that are manipulated in their entirety (instead of the inductive, element-wise processing that is the foundation of all Haskell list functions.)
# by a concurrent, high-performance library of strict, segmented arrays, which we call '''package ndp''' and
 
  +
# by syntactic sugar for nested parallel arrays and array comprehensions, including a program transformation, called '''vectorisation''', which maps nested array parallelism to the API provided by package ndp.
 
  +
As a consequence, parallel arrays are always finite, and standard functions that yield infinite lists, such as <hask>enumFrom</hask> and <hask>repeat</hask>, have no corresponding array operation. Moreover, parallel arrays only have an undirected fold function <hask>foldP</hask> that requires an associative function as an argument – such a fold function has a parallel step complexity of O(log ''n'') for arrays of length ''n''. Parallel arrays also come with some aggregate operations that are absent from the standard list library, such as <hask>permuteP</hask>.
Both package ndp and the syntactic sugar are already usable, but as we have yet to implement vectorisation, they currently cannot be used together. To get an idea of code using array comprehensions, consider the following definition of the dot product of two vectors:
 
  +
  +
=== A simple example ===
  +
  +
As a simple example of a DPH program, consider the following code that computes the dot product of two vectors given as parallel arrays:
 
<haskell>
 
<haskell>
 
dotp :: Num a => [:a:] -> [:a:] -> a
 
dotp :: Num a => [:a:] -> [:a:] -> a
 
dotp xs ys = sumP [:x * y | x <- xs | y <- ys:]
 
dotp xs ys = sumP [:x * y | x <- xs | y <- ys:]
 
</haskell>
 
</haskell>
This code uses an array variant of [http://www.haskell.org/ghc/docs/latest/html/users_guide/syntax-extns.html#parallel-list-comprehensions parallel list comprehensions], but should otherwise be self-explanatory to any Haskell programmer.
+
This code uses an array variant of [http://www.haskell.org/ghc/docs/latest/html/users_guide/syntax-extns.html#parallel-list-comprehensions parallel list comprehensions], which could alternatively be written as <hask>[:x * y | (x, y) <- zipP xs ys:]</hask>, but should otherwise be self-explanatory to any Haskell programmer.
   
In contrast, package ndp requires us to manually desugar the function definition to
 
  +
=== Running DPH programs ===
  +
  +
Unfortunately, we cannot use the above implementation of <hask>dotp</hask> directly in the current preliminary implementation of DPH. In the following, we will discuss how the code needs to be modified and how it needs to be compiled and run for parallel execution. GHC applies an elaborate transformation to DPH code, called ''vectorisation'', that turns nested into flat data parallelism. This transformation is only useful for code that is executed in parallel (i.e., code that manipulates parallel arrays), but for parallel code it dramatically simplifies load balancing.
  +
  +
==== No type classes ====
  +
  +
Unfortunately, vectorisation does not handle type classes at the moment. Hence, we currently need to avoid overloaded operations in parallel code. To account for that limitation, we specialise <hask>dotp</hask> on doubles.
 
<haskell>
 
<haskell>
dotp :: (Num a, UA a) => UArr a -> UArr a -> a
 
  +
dotp_double :: [:Double:] -> [:Double:] -> Double
dotp xs ys = sumU (zipWithU (*) xs ys)
+
dotp_double xs ys = sumP [:x * y | x <- xs | y <- ys:]
 
</haskell>
 
</haskell>
This is still pretty straight forward, combinator-based code. However, desugaring becomes more burdensome when algorithms use nested arrays.
 
   
Until we have added vectorisation to GHC, we recommend to prototype parallel algorithms using nested arrays and array comprehensions, and then, to manually desugar them to use package ndp for performance. For more examples of programs using nested arrays and array comprehensions, refer to [[GHC/Data Parallel Haskell/GHC.PArr|convenience without the speed]], and for more examples of programs using package ndp, refer to [[GHC/Data Parallel Haskell/Package NDP|speed without the convenience]].
 
  +
==== Special Prelude ====
   
=== Download ===
 
  +
As the current implementation of vectorisation cannot handle some language constructs, we cannot use it to vectorise those parts of the standard Prelude that might be used in parallel code (such as arithmetic operations). Instead, DPH comes with its own (rather limited) Prelude in [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude.hs Data.Array.Parallel.Prelude] plus three extra modules to support one numeric type each [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude/Float.hs Data.Array.Parallel.Prelude.Float], [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude/Double.hs Data.Array.Parallel.Prelude.Double], [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude/Int.hs Data.Array.Parallel.Prelude.Int], and [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude/Word8.hs Data.Array.Parallel.Prelude.Word8]. These four modules support the same functions (on different types) and if a program needs to use more than one, they need to be imported qualified (as we cannot use type classes in vectorised code in the current version). Moreover, we have [http://darcs.haskell.org/packages/dph/dph-common/Data/Array/Parallel/Prelude/Bool.hs Data.Array.Parallel.Prelude.Bool]. If your code needs any other numeric types or functions that are not implemented in these Prelude modules, you currently need to implement and vectorise that functionality yourself.
   
'''Important Note:''' Data Parallel Haskell is currently being completely restructured. Consequently, it may or may not properly build in the current development version, and it most certainly will not generate efficient code. We are working on getting it into a reasonable shape again for the GHC 6.10 release (at which point we will update the documentation to explain the current feature set).
 
  +
To compile <hask>dotp_double</hask>, we add the following import statements:
  +
<haskell>
  +
import qualified Prelude
  +
import Data.Array.Parallel
  +
import Data.Array.Parallel.Prelude
  +
import Data.Array.Parallel.Prelude.Double
  +
</haskell>
   
As we are still missing some parts in the implementation of Data Parallel Haskell (most importantly the vectorisation transformation), you need to choose between the following two options:
 
  +
==== Impedance matching ====
   
; '''Convenience without the speed''' : Nice surface syntax for parallel arrays and array comprehensions (very similar to that for Haskell lists) is available in all GHC releases of the 6.8 series. To use parallel arrays, specify the option <code>-XPArr</code> and import <code>GHC.PArr</code>. This option is nice to implement and test parallel algorithms, but the code will be executed purely sequentially with limited performance. Further details including example programs are at [[Data_Parallel_Haskell/GHC.PArr|convenience without the speed]].
 
  +
Special care is needed at the interface between vectorised and non-vectorised code. Currently, only simple types can be passed between these different kinds of code. In particular, parallel arrays (which might be nested) '''cannot''' be passed. Instead, we need to pass flat arrays of type <hask>PArray</hask>. This type is exported by our special-purpose Prelude together with a conversion function <hask>fromPArrayP</hask> (which is specific to the element type due to the lack of type classes in vectorised code).
; '''Speed without the convenience''' : Parallel high-performance arrays that use aggressive [[fusion]] and transparently utilise multicore and SMP architectures are available in the library package <code>ndp</code>. In contrast to the previous option, there is no special syntactic sugar for array types and array comprehensions, and only flat and segmented arrays of basic types and pairs are available. The library is only provided in source form and needs to be compiled in a GHC build tree. However, this is pretty easy to do. The details including example programs are at [[Data_Parallel_Haskell/PackageNDP|speed without the convenience]].
 
   
We are currently working to integrate these two components to finally provide convenience and speed in one.
 
  +
Using this conversion function, we define a wrapper function for <hask>dotp_double</hask> that we export and use from non-vectorised code.
  +
<haskell>
  +
dotp_wrapper :: PArray Double -> PArray Double -> Double
  +
{-# NOINLINE dotp_wrapper #-}
  +
dotp_wrapper v w = dotp_double (fromPArrayP v) (fromPArrayP w)
  +
</haskell>
  +
It is important to mark this function as <hask>NOINLINE</hask> as we don't want it to be inlined into non-vectorised code.
   
'''Disclaimer:''' Data Parallel Haskell is very much '''work in progress.''' Some components are already usable, and we explain here how to use them. However, please be aware that APIs are still in flux and functionality may change during development.
 
  +
==== Compiling vectorised code ====
  +
  +
The syntax for parallel arrays is an extension to Haskell 2010 that needs to be enabled with the language option <hask>ParallelArrays</hask>. Furthermore, we need to explicitly tell GHC if we want to vectorise a module by using the <hask>-fvectorise</hask> option.
  +
  +
Currently, GHC either vectorises all code in a module or none. This can be inconvenient as some parts of a program cannot be vectorised – for example, code in the <hask>IO</hask> monad (the radical re-ordering of computations performed by the vectorisation transformation is only valid for pure code). As a consequence, the programmer currently needs to partition vectorised and non-vectorised code carefully over different modules.
  +
  +
Overall, we get the following complete module definition for the dot-product code:
  +
<haskell>
  +
{-# LANGUAGE ParallelArrays #-}
  +
{-# OPTIONS_GHC -fvectorise #-}
  +
  +
module DotP (dotp_wrapper)
  +
where
  +
  +
import qualified Prelude
  +
import Data.Array.Parallel
  +
import Data.Array.Parallel.Prelude
  +
import Data.Array.Parallel.Prelude.Double
  +
  +
dotp_double :: [:Double:] -> [:Double:] -> Double
  +
dotp_double xs ys = sumP [:x * y | x <- xs | y <- ys:]
  +
  +
dotp_wrapper :: PArray Double -> PArray Double -> Double
  +
{-# NOINLINE dotp_wrapper #-}
  +
dotp_wrapper v w = dotp_double (fromPArrayP v) (fromPArrayP w)
  +
</haskell>
  +
Assuming the module is in a file <hask>DotP.hs</hask>, we compile it as follows:
  +
<blockquote>
  +
<code>ghc -c -Odph -fdph-par DotP.hs</code>
  +
</blockquote>
  +
The option <code>-Odph</code> enables a predefined set of GHC optimisation options that works best for DPH code and <code>-fdph-par</code> selects the standard parallel DPH backend library. (This is currently the only relevant backend, but there may be others in the future.)
  +
  +
==== Using vectorised code ====
  +
  +
Finally, we need a main module that calls the vectorised code, but is itself not vectorised, so that it may contain I/O. In this simple example, we convert two simple lists to parallel arrays, compute their dot product, and print the result:
  +
<haskell>
  +
import Data.Array.Parallel
  +
import Data.Array.Parallel.PArray (PArray, fromList)
  +
  +
import DotP (dotp_wrapper) -- import vectorised code
  +
  +
main :: IO ()
  +
main
  +
= let v = fromList [1..10] -- convert lists...
  +
w = fromList [1,2..20] -- ...to parallel arrays
  +
result = dotp_wrapper v w -- invoke vectorised code
  +
in
  +
print result -- print the result
  +
</haskell>
  +
We compile this module with
  +
<blockquote>
  +
<code>ghc -c -Odph -fdph-par Main.hs</code>
  +
</blockquote>
  +
and finally link the two modules into an executable <code>dotp</code> with
  +
<blockquote>
  +
<code>ghc -o dotp -threaded -fdph-par -rtsopts DotP.o Main.o</code>
  +
</blockquote>
  +
We need the <code>-threaded</code> option to link with GHC's multi-threaded runtime and <code>-fdph-par</code> to link with the standard parallel DPH backend. We include <code>-rtsopts</code> to be able to explicitly determine the number of OS threads used to execute our code.
  +
  +
==== Generating input data ====
  +
  +
To see any benefit from parallel execution, a data-parallel program needs to operate on a sufficiently large data set. Hence, instead of two small constant vectors, we might want to generate some larger input data:
  +
<haskell>
  +
import System.Random (newStdGen)
  +
import Data.Array.Parallel
  +
import Data.Array.Parallel.PArray (PArray, randomRs)
  +
  +
import DotP (dotp_wrapper) -- import vectorised code
  +
  +
main :: IO ()
  +
main
  +
= do
  +
gen1 <- newStdGen
  +
gen2 <- newStdGen
  +
let v = randomRs n range gen1
  +
w = randomRs n range gen2
  +
print $ dotp_wrapper v w -- invoke vectorised code and print the result
  +
where
  +
n = 10000 -- vector length
  +
range = (-100, 100) -- range of vector elements
  +
</haskell>
  +
We compile and link the program as described above.
  +
  +
'''NOTE:''' The code as presented is unsuitable for benchmarking as we wouldn't want to measure the purely sequential random number generation (that dominates this simple program). For benchmarking, we would want to guarantee that the generated vectors are fully evaluated before taking the time. The module [http://www.haskell.org/ghc/docs/latest/html/libraries/dph-par/Data-Array-Parallel-PArray.html Data.Array.Parallel.PArray] exports the function <hask>nf</hask> for this purpose. For a variant of the dot-product example code that determines the CPU time consumed by <hask>dotp_wrapper</hask>, see [[GHC/Data Parallel Haskell/MainTimed|timed dot product]].
  +
  +
==== Parallel execution ====
  +
  +
By default, a Haskell program uses only one OS thread, and hence, also only one CPU core for execution. To use multiple cores, we need to invoke the executable with an explicit RTS command line option — e.g., <code>./dotp +RTS -N2</code> uses two cores. (Strictly speaking, it uses two OS threads, which will be scheduled on two separate cores if available.) To determine the runtime of parallel code, measuring CPU time, as demonstrated in the [[GHC/Data Parallel Haskell/MainTimed|timed variant of the dot product example]], is not sufficient anymore. We need to measure wall clock times instead. For proper benchmarking, it is advisable to use a library, such as [http://hackage.haskell.org/package/criterion criterion].
  +
  +
A beautiful property of DPH is that the number of threads used to execute a program only affects its performance, but not the result. So, it is fine to do all debugging concerning correctness with just one core and to move to multiple cores only for performance debugging.
  +
  +
Data Parallel Haskell –and more generally, GHC's multi-threading support– currently only aims at multicore processors or uniform memory access (UMA) multi-processors. Performance on non-uniform memory access (NUMA) machines is often bad as GHC's runtime makes no effort at optimising placement.
  +
  +
=== Further examples and documentation ===
  +
  +
Further examples are available in the [http://darcs.haskell.org/packages/dph/dph-examples/ examples directory of the package dph source]. This code also includes reference implementations for some of the example that are useful for benchmarking.
  +
  +
The interfaces of the various components of the DPH library are in the [http://hackage.haskell.org/package/dph-par library documentation] on Hackage.
  +
  +
=== Designing parallel programs ===
  +
  +
Data Parallel Haskell is a high-level language to code parallel algorithms. Like plain Haskell, DPH frees the programmer from many low-level operational considerations (such as thread creation, thread synchronisation, critical sections, and deadlock avoidance). Nevertheless, the full responsibility for parallel algorithm design and many performance considerations (such as when does a computation have sufficient parallelism to make it worthwhile to exploit that parallelism) are still with the programmer.
  +
  +
DPH encourages a data-driven style of parallel programming and, in good Haskell tradition, puts the choice of data types first. Specifically, the choice between using lists or parallel arrays for a data structure determines whether operations on the structure will be executed sequentially or in parallel. In addition to suitably combining standard lists and parallel arrays, it is often also useful to embed parallel arrays in a user-defined inductive structure, such as the following definition of parallel rose trees:
  +
<haskell>
  +
data RTree a = RNode [:RTree a:]
  +
</haskell>
  +
The tree is inductively defined; hence, tree traversals will proceed sequentially, level by level. However, the children of each node are held in parallel arrays, and hence, may be traversed in parallel. This structure is, for example, useful in parallel adaptive algorithms based on a hierarchical decomposition, such as the Barnes-Hut algorithm for solving the ''N''-body problem as discussed in more detail in the paper [http://www.cse.unsw.edu.au/~chak/papers/PLKC08.html Harnessing the Multicores: Nested Data Parallelism in Haskell.]
  +
  +
For a general introduction to nested data parallelism and its cost model, see Blelloch's [http://www.cs.cmu.edu/~scandal/cacm/cacm2.html Programming Parallel Algorithms.]
  +
  +
=== Further reading and information on the implementation ===
  +
  +
DPH has two major components: (1) the ''vectorisation transformation'' and (2) the ''generic DPH library for flat parallel arrays''. The vectorisation transformation turns nested into flat data-parallelism and is described in detail in the paper [http://www.cse.unsw.edu.au/~chak/papers/PLKC08.html Harnessing the Multicores: Nested Data Parallelism in Haskell.] The generic array library maps flat data-parallelism to GHC's multi-threaded multicore support and is described in the paper [http://www.cse.unsw.edu.au/~chak/papers/CLPKM06.html Data Parallel Haskell: a status report]. The same topics are also covered in the slides for the two talks [http://research.microsoft.com/~simonpj/papers/ndp/NdpSlides.pdf Nested data parallelism in Haskell] and [http://dataparallel.googlegroups.com/web/UNSW%20CGO%20DP%202007.pdf Compiling nested data parallelism by program transformation].
  +
  +
For further reading, consult this [[GHC/Data Parallel Haskell/References|collection of background papers, and pointers to other people's work]]. If you are really curious and like to know implementation details and the internals of the Data Parallel Haskell project, much of it is described on the GHC developer wiki on the pages covering [http://hackage.haskell.org/trac/ghc/wiki/DataParallel data parallelism] and [http://hackage.haskell.org/trac/ghc/wiki/TypeFunctions type families].
   
=== Further information ===
+
=== Feedback ===
   
For further reading, refer to this [[GHC/Data Parallel Haskell/References|collection of background papers, and pointers to other people's work]]. If you are really curious and like to know implementation details and the internals of the Data Parallel Haskell project, it is all on the GHC developer wiki on the pages covering [http://hackage.haskell.org/trac/ghc/wiki/DataParallel data parallelism] and [http://hackage.haskell.org/trac/ghc/wiki/TypeFunctions associated types].
 
  +
Please file bug reports at [http://hackage.haskell.org/trac/ghc/ GHC's bug tracker]. Moreover, comments and suggestions are very welcome. Please post them to the [mailto:glasgow-haskell-users@haskell.org GHC user's mailing list], or contact the DPH developers directly:
  +
* [http://www.cse.unsw.edu.au/~chak/ Manuel Chakravarty]
  +
* [http://www.cse.unsw.edu.au/~keller/ Gabriele Keller]
  +
* [http://www.cse.unsw.edu.au/~rl/ Roman Leshchinskiy]
  +
* [http://www.cse.unsw.edu.au/~benl/ Ben Lippmeier]
  +
* [http://research.microsoft.com/~simonpj/ Simon Peyton Jones]

Latest revision as of 10:35, 12 July 2012

Data Parallel Haskell

Searching for Parallel Haskell? DPH is a fantastic effort, but it's not the only way to do parallelism in Haskell. Try the Parallel Haskell portal for a more general view.

Data Parallel Haskell is the codename for an extension to the Glasgow Haskell Compiler and its libraries to support nested data parallelism with a focus to utilise multicore CPUs. Nested data parallelism extends the programming model of flat data parallelism, as known from parallel Fortran dialects, to irregular parallel computations (such as divide-and-conquer algorithms) and irregular data structures (such as sparse matrices and tree structures). An introduction to nested data parallelism in Haskell, including some examples, can be found in the paper Nepal – Nested Data-Parallelism in Haskell.

VtG26AnzIklk0sh6YkZSLYNPo1_400.png

This is the performance of a dot product of two vectors of 10 million doubles each using Data Parallel Haskell. Both machines have 8 cores. Each core of the T2 has 8 hardware thread contexts.



Project status

Data Parallel Haskell (DPH) is available as an add-on for GHC 7.4 in the form of a few separate cabal package. All major components of DPH are implemented, including code vectorisation and parallel execution on multicore systems. However, the implementation has many limitations and probably also many bugs. Major limitations include the inability to mix vectorised and non-vectorised code in a single Haskell module, the need to use a feature-deprived, special-purpose Prelude for vectorised code, and a lack of optimisations (leading to poor performance in some cases).

The current implementation should work well for code with nested parallelism, where the depth of nesting is statically fixed or no user-defined nested-parallel datatypes are used. Support for user-defined nested-parallel datatypes is still rather experimental and will likely result in inefficient code.

DPH focuses on irregular data parallelism. For regular data parallel code in Haskell, please consider using the companion library Repa, which builds on the parallel array infrastructure of DPH.

Note: This page describes version 0.6.* of the DPH libraries. We only support this version of DPH as well as the current development version.

Disclaimer: Data Parallel Haskell is very much work in progress. Some components are already usable, and we explain here how to use them. However, please be aware that APIs are still in flux and functionality may change during development.

Where to get it

To get DPH, install GHC 7.4 and then install the DPH libraries with cabal install as follows:

$ cabal update
$ cabal install dph-examples

This will install all DPH packages, including a set of simple examples, see dph-examples. (The package dph-examples does depend on OpenGL and Gloss as both are used in a visualiser for the nobody example.)

WARNING: The vanilla GHC distribution does not include cabal install. This is in contrast to the Haskell Platform, which does include cabal install. If you want to avoid installing the cabal-intstall package and its dependencies explicitly, simply install GHC 7.4.1 in addition to your current Haskell Platform installation. (How to do that depends on your platform and personal preferences. One option is to install a bindist into your home directory with symbolic links to the binaries including the version number.) Then, install DPH with the following command:

cabal install --with-compiler=`which ghc-7.4.1` --with-hc-pkg=`which ghc-pkg-7.4.1` dph-examples

Overview

From a user's point of view, Data Parallel Haskell adds a new data type to Haskell –namely, parallel arrays– as well as operations on parallel arrays. Syntactically, parallel arrays are like lists, only that instead of square brackets [ and ], parallel arrays use square brackets with a colon [: and :]. In particular, [:e:] is the type of parallel arrays with elements of type e; the expression [:x, y, z:] denotes a three element parallel array with elements x, y, and z; and [:x + 1 | x <- xs:] represents a simple array comprehension. More sophisticated array comprehensions (including the equivalent of parallel list comprehensions) as well as enumerations and pattern matching proceed in an analog manner. Moreover, the array library of DPH defines variants of most list operations from the Haskell Prelude and the standard List library (e.g., we have lengthP, sumP, mapP, and so on).

The two main differences between lists and parallel arrays are that (1) parallel arrays are a strict data structure and (2) that they are not inductively defined. Parallel arrays are strict in that by using a single element, all elements of an array are demanded. Hence, all elements of a parallel array might be evaluated in parallel. To facilitate such parallel evaluation, operations on parallel arrays should treat arrays as aggregate structures that are manipulated in their entirety (instead of the inductive, element-wise processing that is the foundation of all Haskell list functions.)

As a consequence, parallel arrays are always finite, and standard functions that yield infinite lists, such as enumFrom and repeat, have no corresponding array operation. Moreover, parallel arrays only have an undirected fold function foldP that requires an associative function as an argument – such a fold function has a parallel step complexity of O(log n) for arrays of length n. Parallel arrays also come with some aggregate operations that are absent from the standard list library, such as permuteP.

A simple example

As a simple example of a DPH program, consider the following code that computes the dot product of two vectors given as parallel arrays:

dotp :: Num a => [:a:] -> [:a:] -> a
dotp xs ys = sumP [:x * y | x <- xs | y <- ys:]

This code uses an array variant of parallel list comprehensions, which could alternatively be written as [:x * y | (x, y) <- zipP xs ys:], but should otherwise be self-explanatory to any Haskell programmer.

Running DPH programs

Unfortunately, we cannot use the above implementation of dotp directly in the current preliminary implementation of DPH. In the following, we will discuss how the code needs to be modified and how it needs to be compiled and run for parallel execution. GHC applies an elaborate transformation to DPH code, called vectorisation, that turns nested into flat data parallelism. This transformation is only useful for code that is executed in parallel (i.e., code that manipulates parallel arrays), but for parallel code it dramatically simplifies load balancing.

No type classes

Unfortunately, vectorisation does not handle type classes at the moment. Hence, we currently need to avoid overloaded operations in parallel code. To account for that limitation, we specialise dotp on doubles.

dotp_double :: [:Double:] -> [:Double:] -> Double
dotp_double xs ys = sumP [:x * y | x <- xs | y <- ys:]

Special Prelude

As the current implementation of vectorisation cannot handle some language constructs, we cannot use it to vectorise those parts of the standard Prelude that might be used in parallel code (such as arithmetic operations). Instead, DPH comes with its own (rather limited) Prelude in Data.Array.Parallel.Prelude plus three extra modules to support one numeric type each Data.Array.Parallel.Prelude.Float, Data.Array.Parallel.Prelude.Double, Data.Array.Parallel.Prelude.Int, and Data.Array.Parallel.Prelude.Word8. These four modules support the same functions (on different types) and if a program needs to use more than one, they need to be imported qualified (as we cannot use type classes in vectorised code in the current version). Moreover, we have Data.Array.Parallel.Prelude.Bool. If your code needs any other numeric types or functions that are not implemented in these Prelude modules, you currently need to implement and vectorise that functionality yourself.

To compile dotp_double, we add the following import statements:

import qualified Prelude
import Data.Array.Parallel
import Data.Array.Parallel.Prelude
import Data.Array.Parallel.Prelude.Double

Impedance matching

Special care is needed at the interface between vectorised and non-vectorised code. Currently, only simple types can be passed between these different kinds of code. In particular, parallel arrays (which might be nested) cannot be passed. Instead, we need to pass flat arrays of type PArray. This type is exported by our special-purpose Prelude together with a conversion function fromPArrayP (which is specific to the element type due to the lack of type classes in vectorised code).

Using this conversion function, we define a wrapper function for dotp_double that we export and use from non-vectorised code.

dotp_wrapper :: PArray Double -> PArray Double -> Double
{-# NOINLINE dotp_wrapper #-}
dotp_wrapper v w = dotp_double (fromPArrayP v) (fromPArrayP w)

It is important to mark this function as NOINLINE as we don't want it to be inlined into non-vectorised code.

Compiling vectorised code

The syntax for parallel arrays is an extension to Haskell 2010 that needs to be enabled with the language option ParallelArrays. Furthermore, we need to explicitly tell GHC if we want to vectorise a module by using the -fvectorise option.

Currently, GHC either vectorises all code in a module or none. This can be inconvenient as some parts of a program cannot be vectorised – for example, code in the IO monad (the radical re-ordering of computations performed by the vectorisation transformation is only valid for pure code). As a consequence, the programmer currently needs to partition vectorised and non-vectorised code carefully over different modules.

Overall, we get the following complete module definition for the dot-product code:

{-# LANGUAGE ParallelArrays #-}
{-# OPTIONS_GHC -fvectorise #-}

module DotP (dotp_wrapper)
where

import qualified Prelude
import Data.Array.Parallel
import Data.Array.Parallel.Prelude
import Data.Array.Parallel.Prelude.Double

dotp_double :: [:Double:] -> [:Double:] -> Double
dotp_double xs ys = sumP [:x * y | x <- xs | y <- ys:]

dotp_wrapper :: PArray Double -> PArray Double -> Double
{-# NOINLINE dotp_wrapper #-}
dotp_wrapper v w = dotp_double (fromPArrayP v) (fromPArrayP w)

Assuming the module is in a file DotP.hs, we compile it as follows:

ghc -c -Odph -fdph-par DotP.hs

The option -Odph enables a predefined set of GHC optimisation options that works best for DPH code and -fdph-par selects the standard parallel DPH backend library. (This is currently the only relevant backend, but there may be others in the future.)

Using vectorised code

Finally, we need a main module that calls the vectorised code, but is itself not vectorised, so that it may contain I/O. In this simple example, we convert two simple lists to parallel arrays, compute their dot product, and print the result:

import Data.Array.Parallel
import Data.Array.Parallel.PArray (PArray, fromList)

import DotP (dotp_wrapper)  -- import vectorised code

main :: IO ()
main
  = let v      = fromList [1..10]    -- convert lists...
        w      = fromList [1,2..20]  -- ...to parallel arrays
        result = dotp_wrapper v w    -- invoke vectorised code
    in
    print result                     -- print the result

We compile this module with

ghc -c -Odph -fdph-par Main.hs

and finally link the two modules into an executable dotp with

ghc -o dotp -threaded -fdph-par -rtsopts DotP.o Main.o

We need the -threaded option to link with GHC's multi-threaded runtime and -fdph-par to link with the standard parallel DPH backend. We include -rtsopts to be able to explicitly determine the number of OS threads used to execute our code.

Generating input data

To see any benefit from parallel execution, a data-parallel program needs to operate on a sufficiently large data set. Hence, instead of two small constant vectors, we might want to generate some larger input data:

import System.Random (newStdGen)
import Data.Array.Parallel
import Data.Array.Parallel.PArray (PArray, randomRs)

import DotP (dotp_wrapper)  -- import vectorised code

main :: IO ()
main
  = do 
      gen1 <- newStdGen
      gen2 <- newStdGen
      let v = randomRs n range gen1
          w = randomRs n range gen2
      print $ dotp_wrapper v w   -- invoke vectorised code and print the result
  where
    n     = 10000        -- vector length
    range = (-100, 100)  -- range of vector elements

We compile and link the program as described above.

NOTE: The code as presented is unsuitable for benchmarking as we wouldn't want to measure the purely sequential random number generation (that dominates this simple program). For benchmarking, we would want to guarantee that the generated vectors are fully evaluated before taking the time. The module Data.Array.Parallel.PArray exports the function nf for this purpose. For a variant of the dot-product example code that determines the CPU time consumed by dotp_wrapper, see timed dot product.

Parallel execution

By default, a Haskell program uses only one OS thread, and hence, also only one CPU core for execution. To use multiple cores, we need to invoke the executable with an explicit RTS command line option — e.g., ./dotp +RTS -N2 uses two cores. (Strictly speaking, it uses two OS threads, which will be scheduled on two separate cores if available.) To determine the runtime of parallel code, measuring CPU time, as demonstrated in the timed variant of the dot product example, is not sufficient anymore. We need to measure wall clock times instead. For proper benchmarking, it is advisable to use a library, such as criterion.

A beautiful property of DPH is that the number of threads used to execute a program only affects its performance, but not the result. So, it is fine to do all debugging concerning correctness with just one core and to move to multiple cores only for performance debugging.

Data Parallel Haskell –and more generally, GHC's multi-threading support– currently only aims at multicore processors or uniform memory access (UMA) multi-processors. Performance on non-uniform memory access (NUMA) machines is often bad as GHC's runtime makes no effort at optimising placement.

Further examples and documentation

Further examples are available in the examples directory of the package dph source. This code also includes reference implementations for some of the example that are useful for benchmarking.

The interfaces of the various components of the DPH library are in the library documentation on Hackage.

Designing parallel programs

Data Parallel Haskell is a high-level language to code parallel algorithms. Like plain Haskell, DPH frees the programmer from many low-level operational considerations (such as thread creation, thread synchronisation, critical sections, and deadlock avoidance). Nevertheless, the full responsibility for parallel algorithm design and many performance considerations (such as when does a computation have sufficient parallelism to make it worthwhile to exploit that parallelism) are still with the programmer.

DPH encourages a data-driven style of parallel programming and, in good Haskell tradition, puts the choice of data types first. Specifically, the choice between using lists or parallel arrays for a data structure determines whether operations on the structure will be executed sequentially or in parallel. In addition to suitably combining standard lists and parallel arrays, it is often also useful to embed parallel arrays in a user-defined inductive structure, such as the following definition of parallel rose trees:

data RTree a = RNode [:RTree a:]

The tree is inductively defined; hence, tree traversals will proceed sequentially, level by level. However, the children of each node are held in parallel arrays, and hence, may be traversed in parallel. This structure is, for example, useful in parallel adaptive algorithms based on a hierarchical decomposition, such as the Barnes-Hut algorithm for solving the N-body problem as discussed in more detail in the paper Harnessing the Multicores: Nested Data Parallelism in Haskell.

For a general introduction to nested data parallelism and its cost model, see Blelloch's Programming Parallel Algorithms.

Further reading and information on the implementation

DPH has two major components: (1) the vectorisation transformation and (2) the generic DPH library for flat parallel arrays. The vectorisation transformation turns nested into flat data-parallelism and is described in detail in the paper Harnessing the Multicores: Nested Data Parallelism in Haskell. The generic array library maps flat data-parallelism to GHC's multi-threaded multicore support and is described in the paper Data Parallel Haskell: a status report. The same topics are also covered in the slides for the two talks Nested data parallelism in Haskell and Compiling nested data parallelism by program transformation.

For further reading, consult this collection of background papers, and pointers to other people's work. If you are really curious and like to know implementation details and the internals of the Data Parallel Haskell project, much of it is described on the GHC developer wiki on the pages covering data parallelism and type families.

Feedback

Please file bug reports at GHC's bug tracker. Moreover, comments and suggestions are very welcome. Please post them to the GHC user's mailing list, or contact the DPH developers directly: