[Haskell-cafe] Efficient signal processing

[Henning Thielemann]

I always thought that signal processing must be a good benchmark for
compiler optimizations and optimizer rules of libraries like 'binary' and
'fps'. Many signal processing functions process a signal from the
beginning to the end with a little of state, thus in many cases I expect
that they can be translated to simple loops. However, I still cannot get
the necessary speed for real-time processing out of 'binary' and 'fps'.

In my setup a signal is of type [Double]. After a series of sound
transformations a signal is finally converted to [Int16] and then to
ByteString. Can I expect that interim signals represented as lists are
optimized away?

I have simple example which writes zeros to disk. The content of the first
file is created by ByteString.replicate for the purpose of comparison with
the second file. With the second file I want to check the speed of the
conversion from [Int16] to ByteString.



module Main (main) where

import System.Time (getClockTime, diffClockTimes, tdSec, tdPicosec)

import qualified Data.ByteString.Lazy as B
import qualified Data.Binary.Put as Bin

import Foreign (Int16)


signalToBinaryPut :: [Int16] -> B.ByteString
signalToBinaryPut =
   Bin.runPut . mapM_ (Bin.putWord16host . fromIntegral)

writeSignalBinaryPut ::
   FilePath -> [Int16] -> IO ()
writeSignalBinaryPut fileName =
   B.writeFile fileName . signalToBinaryPut


measureTime :: String -> IO () -> IO ()
measureTime name act =
   do putStr (name++": ")
      timeA <- getClockTime
      act
      timeB <- getClockTime
      let td = diffClockTimes timeB timeA
      print (fromIntegral (tdSec td) +
             fromInteger (tdPicosec td) * 1e-12 :: Double)

numSamples :: Int
numSamples = 1000000

zeroSignal16 :: [Int16]
zeroSignal16 = replicate numSamples 0

zeroByteString :: B.ByteString
zeroByteString = B.replicate (fromIntegral (2 * numSamples)) 0

main :: IO ()
main =
   do measureTime "write zero bytestring"
         (B.writeFile "zero-bytestring.sw" zeroByteString)
      measureTime "put zero int16"
         (writeSignalBinaryPut "zero-int16string.sw" zeroSignal16)



The program is compiled with GHC-6.4 and option -O2, CPU clock 1.7 GHz.
This yields:

$ speedtest
write zero bytestring: 2.5674e-2
put zero int16: 1.080541


That is, not using ByteString.replicate and converting from [Int16] to
ByteString slows down computation by a factor of 40. What am I doing
wrong?