The Haskell benchmark doesn't use the threaded runtime

[atemerev]

The Haskell benchmark doesn't use the threaded runtime

[bartavelle]

It's 15% faster with MVars on my computer.

[osa1]

Makes sense, as channels are more general than what's needed here.

Results on my laptop: (first one is the MVar version)

➜  haskell git:(master) ✗ time (./Skynet2 +RTS -N4)                                  
Result: 499999500000 in 1.90832s
( ./Skynet2 +RTS -N4; )  6.39s user 1.64s system 329% cpu 2.433 total

➜  haskell git:(master) ✗ time (./Skynet +RTS -N4) 
Result: 499999500000 in 2.027485s
( ./Skynet +RTS -N4; )  6.92s user 1.38s system 356% cpu 2.327 total

[bartavelle]

Also, I would have done this (even though we are not in the actor paradigm now) :

{-# LANGUAGE BangPatterns #-}
module Main (main) where

import Data.Time.Clock (getCurrentTime, diffUTCTime)

import Control.Parallel.Strategies

skynet :: Int -> Int -> Int -> Int
skynet num size dv
    | size == 1 = num
    | otherwise = sum (parMap rpar subcompute [0 .. dv-1])
    where
        subcompute i = let subNum  = num + i * sizeDiv
                           sizeDiv = size `quot` dv
                       in  skynet subNum sizeDiv dv
main :: IO ()
main = do
    start  <- getCurrentTime
    let !result = skynet 0 1000000 10
    end    <- getCurrentTime
    putStrLn $ concat ["Result: ", show result, " in ", show (diffUTCTime end start)]

And it is much faster (0.027s on my computer).

[bitemyapp]

@bartavelle thank you for submitting these, I am going to PR this and other fixes!