How to reduce execution time of for loops in MATLAB?

Question

NANDANA B T 2018-1-9

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/376276-how-to-reduce-execution-time-of-for-loops-in-matlab

评论： NANDANA B T 2018-1-11

I am new to Matlab. I am doing project in communication. While doing convolutional encoding with for loop it takes around 200 sec to execute. How to reduce the execution time?Do any of you guys know how to make the code run faster?

for i=1:size(re_bin,1)/100
    coded_data1(i,:)=convenc(re_bin(i,:),trellis);
end
encodedData= coded_data1;
for j=1:100
    for k=(j*size(re_bin,1)/100)+1: (j+1)*size(re_bin,1)/100  
        coded_data(k-(j*size(re_bin,1)/100),:)=convenc(re_bin(j*size(re_bin,1)/100,:),trellis);
    end
    encodedData1(:,:,j)= coded_data;
end
codedata=encodedData1;
finalCodeddata=cat(3,encodedData,codedata);

4 个评论
显示 2更早的评论隐藏 2更早的评论

Stephen23 2018-1-9

"I initialized the variable before for loop"

MATLAB has no concept of initializing variables, as variables are all dynamically typed (and sized).

Did you preallocate the output array to its final size?

NANDANA B T 2018-1-10

Yes sir I preallocated the output array

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jan 2018-1-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/376276-how-to-reduce-execution-time-of-for-loops-in-matlab#answer_299330

在 MATLAB Online 中打开

s100 = size(re_bin,1) / 100;
% Pre-allocate:
encodedData = zeros(s100, 0);  % Better define 2nd dimension correctly
for i = 1:s100
  encodedData(i, :) = convenc(re_bin(i,:), trellis);
end
encodedData1 = zeros(100, ???, 100);  % Sorry, no idea what to insert for ???
for j = 1:100
  tmp = convenc(re_bin(j * s100, :), trellis);
  for k = (j*s100)+1:(j+1)*s100  
    encodedData1(k - j * s100, :, j) = tmp;
  end
end
finalCodeddata = cat(3,encodedData, encodedData1);

The pre-allocation is essential for speed. It looks like "encodedData1" is very redundant. You should replace the loop over k by a repmat command:

for j = 1:100
  tmp = convenc(re_bin(j * s100, :), trellis);
  encodedData1(:, :, j) = repmat(tmp, 100, 1);
end

Do I see correctly, that the indices in "encodedData1(k - j * s100, :, j)" are 1:100 in each iteration? Sorry, I cannot test this by running your code, because I do not have the required toolbox.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Jan 2018-1-10

The please post the current version of the code. How could we suggest further changes otherwise?

NANDANA B T 2018-1-11

在 MATLAB Online 中打开

function finalCodeddata=convolEncoder(data_bin)

binary=dec2twos(data_bin,16);

%convolutional encoder trellis=poly2trellis(7,[133,171]);

finalCodeddata=zeros(80,32,100); s100 = size(binary,1) / 100; % Pre-allocate: coded_data1 = zeros(s100, 32); % Better define 2nd dimension correctly encodedData1 = zeros(s100, 32,99);

for i=1:s100

coded_data1(i,:)=convenc(binary(i,:),trellis);

end

for j=1:99 tmp = convenc(binary(j * s100, :), trellis);

    encodedData1(:, :, j) = repmat(tmp, 80, 1);
 end

finalCodeddata=cat(3,coded_data1,encodedData1);

endfunction

This is my modified code which takes almost 4 sec for execution

请先登录，再进行评论。

Answer 2

Steven Lord 2018-1-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/376276-how-to-reduce-execution-time-of-for-loops-in-matlab#answer_299326

在 MATLAB Online 中打开

If you profile your code, my guess is that the calls to convenc inside the nested loop take most of the time.

coded_data(k-(j*size(re_bin,1)/100),:)=convenc(re_bin(j*size(re_bin,1)/100,:),trellis);

Since the expression on the right side does not involve the inner loop variable k at all, don't recompute it every time. If you need to have that inner loop over k at all, compute that convenc result once and use that loop simply to assign the result you computed into a section of coded_data. You may need or want to use repmat to eliminate the loop entirely.