How to execute for-loop iterations in parallel

1 次查看（过去 30 天）
Vinit Nagda
Vinit Nagda 2022-2-2
评论： Ive J 2022-2-2
I want to implement parfor to speed up the calculation but I am facing errors.
How can I implement parfor for the following for loop:
I have a 3D matrix X (DIM: mxnxp)
trues=find(X);
for i=1:size(trues,1)
[x,y,z]=ind2sub([m n p],trues(i,1))
if (condition true)
X(x,y,z)=0;
end
end
Is there any other way I can speed up and optimize execution?
Thank you.
  2 个评论
Benjamin Thompson
Benjamin Thompson 2022-2-2
Can you post a working example of sequential MATLAB code that implements your algorithm? You can always set all of X to zeros before the for loop to simplify part of your algorithm. parfor usually cannot write to different locations in a matrix in parallel. You may be able to write the result to an output vector in parfor, then redim the output vector to mxnxp.
Vinit Nagda
Vinit Nagda 2022-2-2
m=300;n=300;p=50;
X=ones(m,n,p);
trues=find(X);
for i=1:size(trues,1)
[x,y,z]=ind2sub([m n p],trues(i,1));
if (x>y+z)
X(x,y,z)=0;
end
end
@Benjamin Thompson Here is the working example. Although the if condition is different in real problem, but algorithm remains same.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（1 个）

Ive J
Ive J 2022-2-2
编辑：Ive J 2022-2-2
You don't need even a loop (parfor aside) for this (and I guess you don't even need ind2sub depending on your true purpose here):
m=300; n=300; p=50;
X = ones(m, n, p);
trues = find(X > 0);
[x, y, z] = ind2sub([m n p], trues);
idx = x > y + z;
X(trues(idx)) = 0;
  2 个评论
Vinit Nagda
Vinit Nagda 2022-2-2
@Ive J Thank you for your response. Apparently, in the real problem the if condition check is bit complex rather than simple x > y + z; and therefore requires a loop.
Ive J
Ive J 2022-2-2
Maybe you can explain the probelm itself then :)

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Loops and Conditional Statements 的更多信息

标签

产品


版本

R2020b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by