# Concurrency demos/Zeta

### From HaskellWiki

< Concurrency demos(Difference between revisions)

m |
|||

Line 16: | Line 16: | ||

-- by all the other threads so as to avoid accumulating rounding imprecisions. |
-- by all the other threads so as to avoid accumulating rounding imprecisions. |
||

zetaRange :: (Floating a, Integral b) => a -> (b, b) -> [a] |
zetaRange :: (Floating a, Integral b) => a -> (b, b) -> [a] |
||

− | zetaRange s (x,y) = [ (fromIntegral n) ** (-s) | n <- [x..y] ] |
+ | zetaRange s (x,y) = [ fromIntegral n ** (-s) | n <- [x..y] ] |

cut :: (Integral a) => (a, a) -> a -> [(a, a)] |
cut :: (Integral a) => (a, a) -> a -> [(a, a)] |

## Revision as of 15:10, 29 November 2006

## Contents |

## 1 A simple example of parallelism in Haskell

This little piece of code computes an approximation of Riemann's zeta function, balancing the work to be done between N threads.

import Control.Concurrent import Control.Concurrent.MVar import Control.Monad import Data.Complex import System.Environment -- Return the list of the terms of the zeta function for the given range. -- We don't sum the terms here but let the main thread sum the lists returned -- by all the other threads so as to avoid accumulating rounding imprecisions. zetaRange :: (Floating a, Integral b) => a -> (b, b) -> [a] zetaRange s (x,y) = [ fromIntegral n ** (-s) | n <- [x..y] ] cut :: (Integral a) => (a, a) -> a -> [(a, a)] cut (x,y) n = (x, x + mine - 1) : cut' (x + mine) size (y - mine) where (size, modulo) = y `divMod` n mine = size + modulo cut' _ _ 0 = [] cut' x' size' n' = (x', x' + size' - 1) : cut' (x' + size') size' (n' - size') getParams :: IO (Int, Int, Complex Double) getParams = do argv <- getArgs case argv of (t:n:s:[]) -> return (read t, read n, read s) _ -> error "usage: zeta <nthreads> <boundary> <s>" main :: IO () main = do (t, n, s) <- getParams childs <- mapM (thread s) (cut (1, n) t) results <- mapM takeMVar childs print (sum (concat results)) where thread s range = do putStrLn ("Starting thread for range " ++ show range) mvar <- newEmptyMVar forkIO (putMVar mvar $! zetaRange s range) return mvar

### 1.1 Or using Strategies

Replace theControl.Concurrent...

import Control.Parallel.Strategies

and replace main by

main :: IO () main = do (t, n, s) <- getParams let ranges = cut (1, n) t results = map (zetaRange s) ranges `using` parList rnf putStr $ unlines [ "Starting thread for range " ++ show r | r <- ranges ] print (sum (concat results))

### 1.2 Using a Chan instead of MVars

Replace the main function with:

main :: IO () main = do (t, n, s) <- getParams chan <- newChan terms <- getChanContents chan forM_ (cut (1,n) t) $ thread chan s let wait xs i result | i >= t = print result -- Done. | otherwise = case xs of Nothing : rest -> wait rest (i + 1) result Just x : rest -> wait rest i (result + x) wait terms 0 0 where thread chan s range = do putStrLn ("Starting thread for range " ++ show range) forkIO $ do mapM_ (writeChan chan . Just) (zetaRange s range) writeChan chan Nothing

## 2 Benchmarks

Insert benchmarks here! :-)