6

(此程序的依赖项:vector --anyJuicyPixels >= 2。代码可用作Gist。)

{-# LANGUAGE Haskell2010 #-}
{-# LANGUAGE BangPatterns #-}

import Control.Arrow
import Data.Bits
import Data.Vector.Unboxed ((!))
import Data.Word
import System.Environment (getArgs)

import qualified Codec.Picture as P
import qualified Data.ByteString as B
import qualified Data.Vector.Unboxed as V

我尝试将Ken Perlin 改进的噪音移植 到 Haskell,但我不完全确定我的方法是否正确。主要部分应该很好地推广到更高和更低的维度,但这是以后的事情:

perlin3 :: (Ord a, Num a, RealFrac a, V.Unbox a) => Permutation -> (a, a, a) -> a
perlin3 p (!x', !y', !z')
  = let (!xX, !x) = actuallyProperFraction x'
        (!yY, !y) = actuallyProperFraction y'
        (!zZ, !z) = actuallyProperFraction z'

        !u = fade x
        !v = fade y
        !w = fade z

        !h = xX
        !a = next p h + yY
        !b = next p (h+1) + yY
        !aa = next p a + zZ
        !ab = next p (a+1) + zZ
        !ba = next p b + zZ
        !bb = next p (b+1) + zZ
        !aaa = next p aa
        !aab = next p (aa+1)
        !aba = next p ab
        !abb = next p (ab+1)
        !baa = next p ba
        !bab = next p (ba+1)
        !bba = next p bb
        !bbb = next p (bb+1)

    in
        lerp w
            (lerp v
                (lerp u
                    (grad aaa (x, y, z))
                    (grad baa (x-1, y, z)))
                (lerp u
                    (grad aba (x, y-1, z))
                    (grad bba (x-1, y-1, z))))
            (lerp v
                (lerp u
                    (grad aab (x, y, z-1))
                    (grad bab (x-1, y, z-1)))
                (lerp u
                    (grad abb (x, y-1, z-1))
                    (grad bbb (x-1, y-1, z-1))))

这当然伴随着函数中提到的几个函数perlin3 ,我希望它们尽可能高效:

fade :: (Ord a, Num a) => a -> a
fade !t | 0 <= t, t <= 1 = t * t * t * (t * (t * 6 - 15) + 10)

lerp :: (Ord a, Num a) => a -> a -> a -> a
lerp !t !a !b | 0 <= t, t <= 1 = a + t * (b - a)

grad :: (Bits hash, Integral hash, Num a, V.Unbox a) => hash -> (a, a, a) -> a
grad !hash (!x, !y, !z) = dot3 (vks `V.unsafeIndex` fromIntegral (hash .&. 15)) (x, y, z)
  where
    vks = V.fromList
        [ (1,1,0), (-1,1,0), (1,-1,0), (-1,-1,0)
        , (1,0,1), (-1,0,1), (1,0,-1), (-1,0,-1)
        , (0,1,1), (0,-1,1), (0,1,-1), (0,-1,-1)
        , (1,1,0), (-1,1,0), (0,-1,1), (0,-1,-1)
        ]

dot3 :: Num a => (a, a, a) -> (a, a, a) -> a
dot3 (!x0, !y0, !z0) (!x1, !y1, !z1) = x0 * x1 + y0 * y1 + z0 * z1

-- Unlike `properFraction`, `actuallyProperFraction` rounds as intended.
actuallyProperFraction :: (RealFrac a, Integral b) => a -> (b, a)
actuallyProperFraction x
  = let (ipart, fpart) = properFraction x
        r = if x >= 0 then (ipart, fpart)
                      else (ipart-1, 1+fpart)
    in r

对于排列组,我只是复制了 Perlin 在他的网站上使用的一个:

newtype Permutation = Permutation (V.Vector Word8)

mkPermutation :: [Word8] -> Permutation
mkPermutation xs
    | length xs >= 256
    = Permutation . V.fromList $ xs

permutation :: Permutation
permutation = mkPermutation
    [151,160,137,91,90,15,
   131,13,201,95,96,53,194,233,7,225,140,36,103,30,69,142,8,99,37,240,21,10,23,
   190, 6,148,247,120,234,75,0,26,197,62,94,252,219,203,117,35,11,32,57,177,33,
   88,237,149,56,87,174,20,125,136,171,168, 68,175,74,165,71,134,139,48,27,166,
   77,146,158,231,83,111,229,122,60,211,133,230,220,105,92,41,55,46,245,40,244,
   102,143,54, 65,25,63,161, 1,216,80,73,209,76,132,187,208, 89,18,169,200,196,
   135,130,116,188,159,86,164,100,109,198,173,186, 3,64,52,217,226,250,124,123,
   5,202,38,147,118,126,255,82,85,212,207,206,59,227,47,16,58,17,182,189,28,42,
   223,183,170,213,119,248,152, 2,44,154,163, 70,221,153,101,155,167, 43,172,9,
   129,22,39,253, 19,98,108,110,79,113,224,232,178,185, 112,104,218,246,97,228,
   251,34,242,193,238,210,144,12,191,179,162,241, 81,51,145,235,249,14,239,107,
   49,192,214, 31,181,199,106,157,184, 84,204,176,115,121,50,45,127, 4,150,254,
   138,236,205,93,222,114,67,29,24,72,243,141,128,195,78,66,215,61,156,180
   ]

next :: Permutation -> Word8 -> Word8
next (Permutation !v) !idx'
  = v `V.unsafeIndex` (fromIntegral $ idx' .&. 0xFF)

而这一切都与 JuicyPixels 联系在一起:

main = do
    [target] <- getArgs
    let image = P.generateImage pixelRenderer 512 512
    P.writePng target image
  where
    pixelRenderer, pixelRenderer' :: Int -> Int -> Word8
    pixelRenderer !x !y
        = floor $ ((perlin3 permutation ((fromIntegral x - 256) / 32,
          (fromIntegral y - 256) / 32, 0 :: Double))+1)/2 * 128

    -- This code is much more readable, but also much slower.
    pixelRenderer' x y
        = (\w -> floor $ ((w+1)/2 * 128)) -- w should be in [-1,+1]
        . perlin3 permutation
        . (\(x,y,z) -> ((x-256)/32, (y-256)/32, (z-256)/32))
        $ (fromIntegral x, fromIntegral y, 0 :: Double)

我的问题是这perlin3对我来说似乎很慢。如果我对其进行概要分析,pixelRenderer 也会花费很多时间,但我现在将忽略它。不知道怎么优化perlin3。我试图用 bang 模式提示 GHC,这将执行时间缩短了一半,这很好。显式特化和内联对ghc -O. perlin3应该这么慢?


更新:这个问题的早期版本提到了我的代码中的一个错误。问题已经解决了; 原来我的旧版本有问题actuallyProperFraction。它隐式地将浮点数的整数部分四舍五入为Word8,然后从浮点数中减去它以获得小数部分。由于Word8只能取介于0255包含之间的值,因此对于该范围之外的数字(包括负数),这将无法正常工作。

4

2 回答 2

4

此代码似乎主要受计算限制。它可以改进一点,但不会改进很多,除非有一种方法可以使用更少的数组查找和更少的算术。

有两种有用的工具可用于测量性能:分析和代码转储。我添加了一个 SCC 注释,perlin3以便它显示在配置文件中。然后我用gcc -O2 -fforce-recomp -ddump-simpl -prof -auto. 该-ddump-simpl标志打印简化的代码。

Profiling:在我的电脑上,运行程序需要 0.60 秒,大约 20% 的执行时间(0.12 秒)perlin3根据 profile 花费在上面。请注意,我的个人资料信息的精度约为 +/-3%。

简化器输出:简化器产生相当干净的代码。 perlin3被内联到pixelRenderer中,所以这是您要查看的输出部分。大多数代码由未装箱的数组读取和未装箱的算术组成。为了提高性能,我们希望消除一些这种算法。

一个简单的更改是消除运行时检查SomeFraction(这不会出现在您的问题中,而是您上传的代码的一部分)。这将程序的执行时间减少到 0.56 秒。

-- someFraction t | 0 <= t, t < 1 = SomeFraction t
someFraction t = SomeFraction t

接下来,在简化器中显示了几个数组查找,如下所示:

                 case GHC.Prim.indexWord8Array#
                        ipv3_s23a
                        (GHC.Prim.+#
                           ipv1_s21N
                           (GHC.Prim.word2Int#
                              (GHC.Prim.and#
                                 (GHC.Prim.narrow8Word#
                                    (GHC.Prim.plusWord# ipv5_s256 (__word 1)))
                                 (__word 255))))

原始操作narrow8Word#用于从a 强制Int转换为 a Word8。我们可以通过在 的定义中使用Int而不是来摆脱这种强制。Word8next

next :: Permutation -> Int -> Int
next (Permutation !v) !idx'
  = fromIntegral $ v `V.unsafeIndex` (fromIntegral idx' .&. 0xFF)

这将程序的执行时间减少到 0.54 秒。仅考虑在 中花费的时间perlin3,执行时间已经(大约)从 0.12 秒下降到 0.06 秒。尽管很难衡量剩余时间的去向,但它很可能分散在剩余的算术和数组访问中。

于 2013-03-22T18:49:55.733 回答
2

在我的机器上,带有散热器优化的参考代码需要 0.19 秒。

首先,我已经使用JuicyPixels了我最喜欢的标志(它们在此处给出):yarryarr-image-io-Odph -rtsopts -threaded -fno-liberate-case -funbox-strict-fields -fexpose-all-unfoldings -funfolding-keeness-factor1000 -fsimpl-tick-factor=500 -fllvm -optlo-O3

import Data.Yarr as Y
import Data.Yarr.IO.Image as Y
...

main = do
    [target] <- getArgs
    image <- dComputeS $ fromFunction (512, 512) (return . pixelRenderer)
    Y.writeImage target (Grey image)
  where
    pixelRenderer, pixelRenderer' :: Dim2 -> Word8
    pixelRenderer (y, x)
        = floor $ ((perlin3 permutation ((fromIntegral x - 256) / 32,
          (fromIntegral y - 256) / 32, 0 :: Double))+1)/2 * 128

    -- This code is much more readable, but also much slower.
    pixelRenderer' (y, x)
        = (\w -> floor $ ((w+1)/2 * 128)) -- w should be in [-1,+1]
        . perlin3 permutation
        . (\(x,y,z) -> ((x-256)/32, (y-256)/32, (z-256)/32))
        $ (fromIntegral x, fromIntegral y, 0 :: Double)

这使程序速度提高了 30%,即 0.13 秒。

其次,我已将标准的使用替换floor

doubleToByte :: Double -> Word8
doubleToByte f = fromIntegral (truncate f :: Int)

这是已知问题(谷歌“haskell floor performance”)。执行时间减少到 52 毫秒(0.052 秒),几乎是 3 倍。

最后,为了好玩,我尝试并行计算噪声(dComputeP而不是在命令行运行中)dComputeS+RTS -N4程序花费了 36 毫秒,包括大约 10 毫秒的 I/O 常数。

于 2013-03-23T09:58:39.953 回答