我试图找出我在使用 Haskell 时遇到的一些性能问题。作为其中的一部分,我编写了一个小的比较程序来比较 C 和 Haskell。具体来说,我将 C 程序翻译为 Haskell,并尽可能少地进行更改。然后,Haskell 程序的速度测量部分以非常命令式的方式编写。
该程序在某个范围内制作两个随机数列表,然后计算通过简单连接这些点形成的图形的积分,其中一个列表是 x 值,一个列表是 y 值。本质上,它是梯形规则。
下面是两个代码:
主程序
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#define N 5000000
#define maxY 1e5f/N
#define maxXgap 1
int main(){
int i;
float *y, *x;
float xaccum, area;
clock_t begin, end;
double time_spent;
y = (float*)malloc(sizeof(float)*N);
x = (float*)malloc(sizeof(float)*N);
srand(50546345); // change seed for different numbers
//populate y and x fields with random points
for(i = 0; i < N; i++){
y[i] = ((float)rand())/((float)RAND_MAX)*maxY;
}
xaccum = 0;
for(i = 0; i < N; i++){
x[i] = xaccum;
xaccum += ((float)rand())/((float)RAND_MAX)*maxXgap;
}
begin = clock();
//perform a trapezoidal integration using the x y coordinates
area = 0;
for(i = 0; i < N-1; i++){
area += (y[i+1]+y[i])/2*(x[i+1]-x[i]);
}
end = clock();
time_spent = (double)(end - begin) / CLOCKS_PER_SEC * 1000;
printf("%i points\n%f area\n%f ms\n", N, area, time_spent);
}
主文件
{-# LANGUAGE BangPatterns #-}
module Main where
import Data.Array.Unboxed
import Data.Array.IO
import Data.List
import System.Random
import System.CPUTime
import Text.Printf
import Control.Exception
main :: IO ()
main = do
(x,y) <- initArrays
area <- time $ integrate x y
print area
n :: Int
n = 5000000
maxY :: Float
maxY = 100000.0/(fromIntegral n)
maxXgap :: Float
maxXgap = 1
--initialize arrays with random floats
--this part is not measured in the running time (very slow)
initArrays :: IO (IOUArray Int Float, IOUArray Int Float)
initArrays = do
y <- newListArray (0,n-1) (randomList maxY n (mkStdGen 23432))
x <- newListArray (0,n-1) (scanl1 (+) $ randomList maxXgap n (mkStdGen 5462))
return (x,y)
randomList :: Float -> Int -> StdGen -> [Float]
randomList max n gen = map (abs . ((*) max)) (take n . unfoldr (Just . random) $ gen)
integrate :: IOUArray Int Float -> IOUArray Int Float -> IO Float
integrate x y = iterative x y 0 0
iterative :: IOUArray Int Float -> IOUArray Int Float -> Int -> Float -> IO Float
iterative x y !i !accum = do if i == n-1
then return accum
else do x1 <- readArray x i
x2 <- readArray x (i+1)
y1 <- readArray y i
y2 <- readArray y (i+1)
iterative x y (i+1) (accum + (y2+y1)/2*(x2-x1))
time :: IO t -> IO t
time a = do
start <- getCPUTime
v <- a
end <- getCPUTime
let diff = (fromIntegral (end-start)) / (10^9)
printf "Computation time %0.5f ms\n" (diff :: Double)
return v
在我的系统上,C 集成大约需要 7 毫秒,Haskell 集成大约需要 60 毫秒。当然 Haskell 版本会更慢,但我想知道为什么它会慢得多。显然 Haskell 代码有很多低效率的地方。
为什么 Haskell 代码这么慢?怎么可能修复它?
感谢您的任何回答。