Modifying large arrays slower than modifying structure arrays?

7 ビュー (過去 30 日間)
Joshua  Robinson
Joshua Robinson 2019 年 9 月 11 日
編集済み: Joshua Robinson 2019 年 9 月 11 日
I'm having issues with my code where I noticed manipulating large matrix arrays is extremely slow.
In my original implementation, I have to read in some data from a binary file using a for loop. I noticed that if I inserted the data into an array of structures instead of putting the data directly into an initialized matrix array, my performance was much better. I'm not sure why this is happening? Could someone provide me with some insight?
Below is a representative example of the performance issues I'm seeing.
% set up raw data
n = 10000;
rawNum = (1:n);
rawData = magic(n); % if n==10000, data is about 0.75 GB
% struct example
simpleStruct.num = [];
simpleStruct.data = [];
S(n,1) = simpleStruct;
tic
for i=1:n
S(i).num = rawNum(i);
S(i).data = rawData(i,:);
end
fprintf("Time took to populate struct array: %1.3f seconds.\n",toc);
% array example
arrayNum = zeros(n,1);
arrayData = zeros(n,n);
tic
for i=1:n
arrayNum(i) = rawNum(i);
arrayData(i,:) = rawData(i,:);
end
fprintf("Time took to populate plain ole array: %1.3f seconds.\n",toc);

採用された回答

Walter Roberson
Walter Roberson 2019 年 9 月 11 日
Row versus column ordering effects.
>> tic;for i = 1:n; arrayNum(i) = rawNum(i); arrayData(i,:) = rawData(i,:); end;toc
Elapsed time is 2.254735 seconds.
>> tic;for i = 1:n; arrayNum(i) = rawNum(i); arrayData(:,i) = rawData(:,i); end;toc
Elapsed time is 0.196039 seconds.
  2 件のコメント
Joshua  Robinson
Joshua Robinson 2019 年 9 月 11 日
編集済み: Joshua Robinson 2019 年 9 月 11 日
Thanks!
EDIT: I guess however, if I update my original question to now include the effects of row vectors vs column vectors, it's still true that matlab handles the structs better then either the column vector or row vector.
% set up raw data
n = 10000;
rawNum = (1:n);
rawData = magic(n); % if n==10000, data is about 0.75 GB
% struct example
simpleStruct.num = [];
simpleStruct.data = [];
S(n,1) = simpleStruct;
tic
for i=1:n
S(i).num = rawNum(i);
S(i).data = rawData(:,i);
end
fprintf("Time took to populate struct array: %1.3f seconds.\n",toc);
% array column example
arrayNum = zeros(n,1);
arrayData = zeros(n,n);
tic
for i=1:n
arrayNum(i) = rawNum(i);
arrayData(:,i) = rawData(:,i);
end
fprintf("Time took to populate columns in array: %1.3f seconds.\n",toc);
% array row example
arrayNum = zeros(n,1);
arrayData = zeros(n,n);
tic
for i=1:n
arrayNum(i) = rawNum(i);
arrayData(i,:) = rawData(i,:);
end
fprintf("Time took to populate rows in array: %1.3f seconds.\n",toc);
Time took to populate struct array: 0.387 seconds.
Time took to populate columns in array: 0.534 seconds.
Time took to populate rows in array: 2.307 seconds.
Bruno Luong
Bruno Luong 2019 年 9 月 11 日
Test is not correct; Correct is row-vs-row or col-vs-col,

サインインしてコメントする。

その他の回答 (0 件)

カテゴリ

Help Center および File ExchangeMatrix Indexing についてさらに検索

製品


リリース

R2017b

Community Treasure Hunt

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

Start Hunting!

Translated by