Nestedparfor-Loops

You cannot use aparfor-loop inside anotherparfor循环。一个s an example, the following nesting ofparfor-loops is not allowed:

parfor一世= 1:10parforj = 1：5。。。结尾结尾

Code Analyzer in the MATLAB^®Editor flags the use ofparfor在另一个parfor-环形：

你不能筑巢parfor-loops because parallelization can be performed at only one level. Therefore, choose which loop to run in parallel, and convert the other loop to a为了循环。

Consider the following performance issues when dealing with nested loops:

You can also use a function that usesparforand embed it in aparfor循环。Parallelization occurs only at the outer level. In the following example, call a functionMyFun.m一世nside the outerparfor循环。The innerparfor-loop embedded inMyFun.mruns sequentially, not in parallel.

parfori = 1:10 myFun（i）结尾函数MyFun(i)parforj = 1：5。。。结尾结尾

转换嵌套为了-Loops toparfor-Loops

一个typical use of nested loops is to step through an array using a one-loop variable to index one dimension, and a nested-loop variable to index another dimension. The basic form is:

x =零（n，m）;为了a =1:n为了b = 1：m x（a，b）=有趣（a，b）结尾结尾

The following code shows a simple example. Use抽动andTOCto measure the computing time needed.

a = 100;抽动为了i = 1：100为了j= 1:100 a(i,j) = max(abs(eig(rand(A))));结尾结尾TOC

经过的时间为49.376732秒。

You can parallelize either of the nested loops, but you cannot run both in parallel. The reason is that the workers in a parallel pool cannot start or access further parallel pools.

If the loop counted by一世一世s converted to aparfor-loop, then each worker in the pool executes the nested loops using thejloop counter. Thej循环本身不能作为一个parforon each worker.

Because parallel processing incurs overhead, you must choose carefully whether you want to convert either the inner or the outer为了-loop to aparfor循环。The following example shows how to measure the parallel overhead.

第一个仅转换outer为了-loop to aparfor循环。利用抽动andTOCto measure the computing time needed. UseticbytesandTOCBytesto measure how much data is transferred to and from the workers in the parallel pool.

运行新代码，然后再次运行。第一次运行比后续运行慢，因为平行池需要一些时间来启动并使代码可用于工人。

a = 100;抽动ticbytes(gcp);parfori = 1：100为了j= 1:100 a(i,j) = max(abs(eig(rand(A))));结尾结尾TOCBytes(gcp) toc

BytesSentToWorkers BytesReceivedFromWorkers __________________ ________________________ 1 32984 24512 2 33784 25312 3 33784 25312 4 34584 26112 Total 1.3514e+05 1.0125e+05 Elapsed time is 14.130674 seconds.

Next convert only the一世nnerloop to aparfor循环。Measure the time needed and data transferred as in the previous case.

a = 100;抽动ticbytes(gcp);为了i = 1：100parforj= 1:100 a(i,j) = max(abs(eig(rand(A))));结尾结尾TOCBytes(gcp) toc

bytessenttoworkers bytesreceivedfromworkers __________________ __________________________________________________________________________________________________________________________________________________1.3458 6 5.4555+05 2 1.3495 2 1.3496e+06 5.4858e+05 3 1.3496e+05 3 1.3677E+06 5.6034E时间为48.631737秒。

If you convert the一世nnerloop to aparfor- 环，传输的数据的时间和数量都比平行的外循环大得多。在这种情况下，经过的时间几乎与嵌套为了-loop example. The speedup is smaller than running the outer loop in parallel, because you have more data transfer and thus more parallel overhead. Therefore if you execute the一世nner与运行串行相比，并行循环，您没有获得计算好处为了循环。

If you want to reduce parallel overhead and speed up your computation, run the outer loop in parallel.

If you convert the一世nner相反，循环，然后外循环的每次迭代都会启动一个单独的parfor循环。That is, the inner loop conversion creates 100parfor- 环。Each of the multipleparforexecutions incurs overhead. If you want to reduce parallel overhead, you should run the outer loop in parallel instead, because overhead only occurs once.

Nested为了-Loops: Requirements and Limitations

If you want to convert a nested为了-loop to aparfor- 环，您必须确保正确分类循环变量，请参阅Troubleshoot Variables in parfor-Loops。If your code does not adhere to the guidelines and restrictions labeled asRequired，您会遇到一个错误。MATLAB在读取代码时捕获其中一些错误。这些错误被标记为Required (static)。

In the following example, the code on the left does not work because you define the upper limit of the为了- 通过函数调用环。右侧的代码通过首先定义广播或常数变量来提供解决方法parfor-环形：

一个= zeros(100, 200);parfori = 1：大小（a，1）为了j= 1:size(A, 2) A(i, j) = i + j;结尾结尾

一个= zeros(100, 200); n = size(A, 2);parfori = 1：大小（a，1）为了j = 1：n a（i，j）= i + j;结尾结尾

遵循此限制。如果筑巢为了-loop变量在任何地方更改parfor-loop other than by its为了声明，该地区由为了-loop variable is not guaranteed to be available at each worker.

左侧的代码无效，因为它试图修改嵌套的值为了-loop variablej在循环的身体中。右侧的代码通过分配嵌套来提供解决方法为了-loop variable to a temporary variablet，，，，and then updatingt。

一个= zeros(10);parfor一世= 1:10为了j= 1:10 A(i, j) = 1; j = j+1;结尾结尾

一个= zeros(10);parfor一世= 1:10为了j= 1:10 A(i, j) = 1; t = j; t = t + 1;结尾结尾

遵循此限制。如果是嵌套为了- 环变量是索引的，迭代不保证是独立的。

左侧的示例无效，因为它试图索引嵌套为了-loop variablej。The example on the right removes this indexing.

一个= zeros(10);parfor一世= 1:10为了j = 1:10 j（1）;结尾结尾

一个= zeros(10);parfor一世= 1:10为了j= 1:10 j;结尾结尾

For example, the following code on the left does not work, but the code on the right does:

一个= zeros(4, 11);parfor一世= 1:4为了j = 1:10 a（i，j + 1）= i + j;结尾结尾

一个= zeros(4, 11);parfor一世= 1:4为了j= 2:11 A(i, j) = i + j - 1;结尾结尾

In the following example, the code on the left does not work because一个一世s sliced and indexed inside the nested为了循环。The code on the right works becausev一世s assigned to一个outside of the nested loop:

一个= zeros(4, 10);parfor一世= 1:4为了j = 1:10 a（i，j）= i + j;结尾disp(A(i, j))结尾

一个= zeros(4, 10);parfor一世= 1:4 v = zeros(1, 10);为了j = 1:10 v (j) = + j;结尾disp（v（j））a（i，:) = v;结尾

parfor-Loop Limitations

函数a = pfeg函数out = nfcn(in) out = 1 + in;结尾fcn = @nfcn;parfor一世dx = 1:10 A(idx) = feval(fcn, idx);结尾结尾

>> PFEG使用“本地”配置文件...连接到4名工人。ans = 2 3 4 5 6 7 8 9 10 11

x = 1:10;parfor我= 1:10 temp = x(i); anonymousFunction = @() 2*temp; x(i) = anonymousFunction() + i;结尾disp(x);

a ='a'; myFunction(a)函数X = myFunction(a) name = inputname(1);parfor一世=1:2 X(i).(name) = i;结尾结尾

myfunction（'a'，，，，'b'）函数X = myFunction(a,b) nin = nargin;parfor一世=1:2 X(i) = i*nin;结尾结尾