有人可以解释一下为什么会有显着的时间差异吗?
function [] = maincalc ()
ak=am();
t1=tic;
[out] = myfun (ak.aa, ak.b, ak.c, ak.d, ak.e, ak.f, ak.g, ak.h);
to1end=toc(t1)
t2=tic;
[out2] = myfun2 (ak);
to2end=toc(t2)
结果:
to1end =
0.047520231560659
to2end =
12.490895284055467
class am(我知道有人可能会说没有理由为此使用类,但整个代码是更复杂和长的代码的简化,并且类是必要的):
classdef am
properties
aa = 1;
b = 2;
c = 3;
d = 4;
e = 2.3;
f = 4.2;
g = 5.09;
h = 12.3;
end
end
函数myfun:
function [out] = myfun (aa, b, c, d, e, f, g, h)
n = 500000;
i = 0; j = 0; k = 0; l = 0;
for s = 1:n
i = aa/b + j*k - i;
j = c/d ^ 0.5 - j / i;
k = e*f + aa/3 - k/8;
l = g + exp (h) + l ^ -1;
end
out.i = i;
out.j = j;
out.k = k;
out.l = l;
函数myfun2:
function [out] = myfun2 ( ak )
n = 500000;
i = 0; j = 0; k = 0; l = 0;
for s = 1:n
i = ak.aa/ak.b + j*k - i;
j = ak.c/ak.d ^ 0.5 - j / i;
k = ak.e*ak.f + ak.aa/3 - k/8;
l = ak.g + exp (ak.h) + l ^ -1;
end
out.i = i;
out.j = j;
out.k = k;
out.l = l;
我在某处读过有人解释 MATLAB 的写时复制,但这并不真正适用于此,因为没有对该类的任何成员进行任何更改。
================================================== ===================================== 此行下方的详细信息最近于 2013 年 8 月 2 日添加
Marcin 回应说,这与 MATLAB 将参数传递给函数的方式没有太大关系(顺便说一下,这是一个伟大的发现!),但我认为它仍然有一定关系。我编写了另一个代码,这次所有三种方法都需要多次访问该类:
function [] = maincalc3 ()
inputvar=inputclass();
to1end = 0;
to2end = 0;
to3end = 0;
j = 100;
for i = 1:j;
t1=tic;
[out] = func1 (inputvar);
to1end=toc(t1) + to1end;
t2=tic;
[out2] = func2 (inputvar.s);
to2end=toc(t2) + to2end;
t3=tic;
[out3] = func3 (inputvar);
to3end=toc(t3) + to3end;
end
......................................
classdef inputclass
properties
s = 1;
end
end
......................................
function f = func1 (inputvar)
f = inputvar.s;
end
......................................
function f = func2 (s)
f = s;
end
......................................
function [f] = func3 (inputvar)
s=inputvar.s;
f = s;
end
结果:
to1end =
0.002419525505078
to2end =
0.001517134538850
to3end =
0.002353777529397
func1()
和 func3()
花费的时间大约相同,但 func2
花费的时间大约少 60%。这是否意味着 MATLAB 将参数传递给函数的方式(通过值或对象)确实会影响性能?
最佳答案
我认为这与按值或引用将对象传递给函数没有太大关系。这只是因为在您的循环中,在 myfun2() 中,matlab 需要多次访问对象及其字段。就是这样。
例如,如果您创建名为 myfun3()
的第三个函数,如下所示:
function [out] = myfun3 (ak)
n = 500000;
i = 0; j = 0; k = 0; l = 0;
% FOLLOWING LINE IS NEW <-----
% REMOVE object refencese from within the loop and create local variables
% to use in the loop, instead of referencing object properties all the time.
aa = ak.aa; b = ak.b; c=ak.c; d=ak.d; e=ak.e; f=ak.f; g=ak.g; h=ak.h;
for s = 1:n
i = aa/b + j*k - i;
j = c/d ^ 0.5 - j / i;
k = e*f + aa/3 - k/8;
l = g + exp (h) + l ^ -1;
end
out.i = i;
out.j = j;
out.k = k;
out.l = l;
该函数的执行速度甚至比 myfun1()
还要快一些。在我的电脑上,结果是:
to1end =
0.0533
to2end =
23.9410
to3end =
0.0526 % RESULT for myfun3() function
关于performance - MATLAB 按值传递与按对象传递,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/17982911/