Haskell Parallelism Task

Question

I'm struggling to understand some aspects of parallelism on haskell. I have to apply parallelism to a piece of code, but the attempts I tried do not work properly.

The function is as follows:

fft :: [Complex Float] -> [Complex Float]
fft [a] = [a]
fft as = interleave ls rs
  where
    (cs,ds) = bflyS as
    ls = fft cs
    rs = fft ds

interleave [] bs = bs
interleave (a:as) bs = a : interleave bs as

halve as = splitAt n' as
   where
    n' = div (length as + 1) 2

-- twiddle factors
tw :: Int -> Int -> Complex Float
tw n k = cis (-2 * pi * fromIntegral k / fromIntegral n)

bflyS :: [Complex Float] -> ([Complex Float], [Complex Float])
bflyS as = (los,rts)
  where
    (ls,rs) = halve as
    los = zipWith (+) ls rs
    ros = zipWith (-) ls rs
    rts = zipWith (*) ros [tw n i | i <- [0..n - 1]]
    n = length a

My attempts to perform this parallel function task are as follows:

bflySTask :: [Complex Float] -> ([Complex Float], [Complex Float])
bflySTask as = (los,rts) `using` if n>1000 then  parTuple2 (parListChunk 500 rseq)    (parListChunk 500 rseq) else r0
  where
    (ls,rs) = halve as
    los = zipWith (+) ls rs 
    ros = zipWith (-) ls rs
    n = length as
    fx =  (map (tw (n)) [0..n-1])
    rts =  zipWith (*) ros fx

And with Par Monad

bflySTask :: [Complex Float] -> ([Complex Float], [Complex Float])
bflySTask as = f as
  where
    (ls,rs) = halve as
    los = zipWith (+) ls rs
    ros = zipWith (-) ls rs
    n = length as
    fx = (map (tw (n)) [0..n-1]) 

    f as = if n>10000 then
                runPar $ do
                v1<-new
            v2<-new
            fork $ put v1 (zipWith (*) ros fx)
            fork $ put v2 (zipWith (+) ls rs)
            a <- get v1
            b <- get v2
            return (a, b)
        else
            (zipWith (+) ls rs, zipWith (*) ros fx)

Both approaches slightly reduce parallelism, like the balance of Parallel GC 1.30, and some sparks GC'd and fizzled. Does anyone know what I can do for others to apply parallelism to these functions without changing the data structure?