boxed.hs

import qualified Data.ByteString.Lazy as L
import Data.Word (Word8)

data Stats = Stats
    { total :: Int
    , count :: Int
    }

initStats :: Stats
initStats = Stats 0 0

addStat :: Int -> Stats -> Stats
addStat x (Stats t c) = Stats (t + x) (c + 1)

average :: Stats -> Double
average (Stats t c) = fromIntegral t / fromIntegral c

fileName :: FilePath
fileName = "sample-data.bin"

population :: Int
population = 9

addStats :: [Stats] -> [Word8] -> [Stats]
addStats stats0 bytes0 =
    loop [] stats0 bytes0
  where
    loop front stats [] = reverse front ++ stats
    loop front [] bytes = loop [] (reverse front) bytes
    loop front (x:stats) (y:bytes) = loop
        (addStat (fromIntegral y) x:front)
        stats
        bytes

main :: IO ()
main = do
    lbs <- L.readFile "sample-data.bin"
    let initial = replicate population initStats
        final = addStats initial (L.unpack lbs)
    print $ map average final

{-

[127.436561,127.449141,127.64308,127.551385,127.479474,127.457795,127.577525,127.697076,127.463543]
   2,948,968,072 bytes allocated in the heap
   4,447,810,144 bytes copied during GC
     397,149,080 bytes maximum residency (24 sample(s))
      48,124,104 bytes maximum slop
            1065 MB total memory in use (0 MB lost due to fragmentation)

                                     Tot time (elapsed)  Avg pause  Max pause
  Gen  0      5574 colls,     0 par    0.340s   2.030s     0.0004s    0.0014s
  Gen  1        24 colls,     0 par    1.440s   1.706s     0.0711s    0.2660s

  INIT    time    0.000s  (  0.000s elapsed)
  MUT     time    0.252s  (  1.178s elapsed)
  GC      time    1.780s  (  3.737s elapsed)
  EXIT    time    0.000s  (  0.004s elapsed)
  Total   time    2.032s  (  4.919s elapsed)

  %GC     time      87.6%  (76.0% elapsed)

  Alloc rate    11,702,254,253 bytes per MUT second

  Productivity  12.4% of total user, 5.1% of total elapsed


real    0m4.923s
user    0m2.032s
sys     0m2.884s

-}

make-data.sh

#!/usr/bin/env bash

set -exuo pipefail

dd if=/dev/urandom of=sample-data.bin bs=9000 count=1000

strict.hs

import qualified Data.ByteString.Lazy as L
import Data.Word (Word8)

data Stats = Stats
    { total :: !Int
    , count :: !Int
    }

initStats :: Stats
initStats = Stats 0 0

addStat :: Int -> Stats -> Stats
addStat x (Stats t c) = Stats (t + x) (c + 1)

average :: Stats -> Double
average (Stats t c) = fromIntegral t / fromIntegral c

fileName :: FilePath
fileName = "sample-data.bin"

population :: Int
population = 9

addStats :: [Stats] -> [Word8] -> [Stats]
addStats stats0 bytes0 =
    loop [] stats0 bytes0
  where
    loop front stats [] = reverse front ++ stats
    loop front [] bytes = loop [] (reverse front) bytes
    loop front (x:stats) (y:bytes) = y' `seq` loop
        (y':front)
        stats
        bytes
      where
        y' = addStat (fromIntegral y) x

main :: IO ()
main = do
    lbs <- L.readFile "sample-data.bin"
    let initial = replicate population initStats
        final = addStats initial (L.unpack lbs)
    print $ map average final

{-

[127.436561,127.449141,127.64308,127.551385,127.479474,127.457795,127.577525,127.697076,127.463543]
   1,046,425,584 bytes allocated in the heap
     247,482,784 bytes copied during GC
          83,160 bytes maximum residency (275 sample(s))
         193,920 bytes maximum slop
               2 MB total memory in use (0 MB lost due to fragmentation)

                                     Tot time (elapsed)  Avg pause  Max pause
  Gen  0      1713 colls,     0 par    0.008s   0.120s     0.0001s    0.0009s
  Gen  1       275 colls,     0 par    0.004s   0.013s     0.0000s    0.0002s

  INIT    time    0.000s  (  0.000s elapsed)
  MUT     time    0.008s  (  0.139s elapsed)
  GC      time    0.012s  (  0.133s elapsed)
  EXIT    time    0.000s  (  0.000s elapsed)
  Total   time    0.020s  (  0.272s elapsed)

  %GC     time      60.0%  (48.7% elapsed)

  Alloc rate    130,803,198,000 bytes per MUT second

  Productivity  40.0% of total user, 2.9% of total elapsed


real    0m0.274s
user    0m0.020s
sys     0m0.252s

-}

unboxed.hs

{-# LANGUAGE TemplateHaskell, TypeFamilies, MultiParamTypeClasses #-}
import qualified Data.ByteString.Lazy as L
import Data.Word (Word8)
import Data.Vector.Unboxed.Deriving
import qualified Data.Vector.Unboxed as V
import qualified Data.Vector.Unboxed.Mutable as M
import Control.Monad.Primitive (PrimState)

data Stats = Stats
    { total :: !Int
    , count :: !Int
    }

-- auto-generate the marshaling code
derivingUnbox "Stats"
    [t|Stats -> (Int, Int)|]
    [|\(Stats t c) -> (t, c)|]
    [|\(t, c) -> Stats t c|]

initStats :: Stats
initStats = Stats 0 0

addStat :: Int -> Stats -> Stats
addStat x (Stats t c) = Stats (t + x) (c + 1)

average :: Stats -> Double
average (Stats t c) = fromIntegral t / fromIntegral c

fileName :: FilePath
fileName = "sample-data.bin"

population :: Int
population = 9

addStats :: M.MVector (PrimState IO) Stats -> [Word8] -> IO ()
addStats m =
    loop 0
  where
    loop _ [] = return ()
    loop index bytes | index >= M.length m = loop 0 bytes
    loop index (x:bytes) = do
        stats <- M.unsafeRead m index
        M.unsafeWrite m index $ addStat (fromIntegral x) stats
        loop (index + 1) bytes

main :: IO ()
main = do
    lbs <- L.readFile "sample-data.bin"
    m <- M.replicate population initStats
    addStats m (L.unpack lbs)
    final <- V.unsafeFreeze m
    print $ V.map average final

{-

fromList [127.436561,127.449141,127.64308,127.551385,127.479474,127.457795,127.577525,127.697076,127.463543]
     374,426,560 bytes allocated in the heap
     148,927,200 bytes copied during GC
          57,208 bytes maximum residency (138 sample(s))
         244,032 bytes maximum slop
               3 MB total memory in use (0 MB lost due to fragmentation)

                                     Tot time (elapsed)  Avg pause  Max pause
  Gen  0       565 colls,     0 par    0.004s   0.075s     0.0001s    0.0010s
  Gen  1       138 colls,     0 par    0.000s   0.006s     0.0000s    0.0002s

  INIT    time    0.000s  (  0.000s elapsed)
  MUT     time    0.004s  (  0.069s elapsed)
  GC      time    0.004s  (  0.081s elapsed)
  EXIT    time    0.000s  (  0.000s elapsed)
  Total   time    0.008s  (  0.150s elapsed)

  %GC     time      50.0%  (54.0% elapsed)

  Alloc rate    93,606,640,000 bytes per MUT second

  Productivity  50.0% of total user, 2.7% of total elapsed


real    0m0.152s
user    0m0.008s
sys     0m0.140s

-}