MATLAB Answers

Is it possible to accelerate the speed of saving data into files with parallel way?

2 views (last 30 days)
wei zhang
wei zhang on 18 Jul 2020
Edited: Bruno Luong on 21 Jul 2020
I am trying to save data to files, i.e. txt file. The speed is very low. This step takes almost 80% time over my whole script. My computer has many cores. Could I use them, i.e. parallel way, to help to accelerate the speed in Matlab? I don't know much about know this. May it is impossible. Please just tell me some idea or suggestion about it? Thank you very much.
(edited as Rik suggested)
My code uses fprintf function in a big loop. This is the most time consuming step.
i.e. I need to save a large matrix a, which is 4000000*4
p = fopen(filename,'w');
for i=1:length(a)
fprintf(fp,'%d %d %d %d\n',a(i,1),a(i,2),a(i,3),a(i,4));
end
Is there any method to accelerate it?

  4 Comments

Show 1 older comment
wei zhang
wei zhang on 19 Jul 2020
@Rik I am using fopen and fprintf functions. Thank you for introducing the bottleneck causes. Maybe it is not the problem for Matlab.
Rik
Rik on 19 Jul 2020
Without knowing more about your code, this is only conjecture. You need to post exact code if you want specific advice. You can also run the profiler to find bottlenecks in your code.
wei zhang
wei zhang on 21 Jul 2020
@Rik I had add a mini example of my fprintf codes. I haven't find any method to improve it. Maybe it is impossible.

Sign in to comment.

Accepted Answer

Bruno Luong
Bruno Luong on 21 Jul 2020
Edited: Bruno Luong on 21 Jul 2020
You might try to remove the for-loop
fp = fopen(filename,'w');
fprintf(fp,'%d %d %d %d\n', a(:,1:4).'); % remove the indexing with your array has 4 columns
fclose(fp);
A file is a sequential access device. There is no use of parallel CPU. When you write a file a hard disk head must go sequentially in tracks and put a little electrical signal there to store your data. More or less samething happens with all other HW, even SSD.
Another thing that takes time is convert the internal binary format to digital of your fprintf('%d').
If you can, just write file in binary format. That is the best method (fatest).

  2 Comments

wei zhang
wei zhang on 21 Jul 2020
I have tested your codethat your way could accelerate the process remarkably. Really thankful. The inputs should be matrix, not number. Could you give another advice to the code below to remove the for loop? I have no ideas about it with only symbol.
for i = 1:n_face
fprintf(fp,'0 0\n');
end
Bruno Luong
Bruno Luong on 21 Jul 2020
Try this:
fp = fopen('test.txt','w')
fwrite(fp, repmat(sprintf('0 0\n'),1,n_face));
fclose(fp);
Or you could do with the same method as I have showed
fp = fopen('test.txt','w');
a = repmat([0 0], n_face, 1);
fprintf(fp,'%d %d\n', a); % no need for transpose since it's all 0s
fclose(fp);

Sign in to comment.

More Answers (1)

Mohammad Sami
Mohammad Sami on 21 Jul 2020
Edited: Mohammad Sami on 21 Jul 2020
Is there a reason why you want to use a for loop to write this ?
You can write the entire matrix to file in one go using writematrix function.
a = randi([0 10],1000000,4);
tic;
writematrix(a,'abc.txt','FileType','text','Delimiter',' ');
toc;

  1 Comment

Rik
Rik on 21 Jul 2020
I'm on mobile so I can't write and test the code, but you can also use fprintf. You just need to make sure the orientation of a is correct (it will read the input column by column when writing the lines in the text file).

Sign in to comment.

Products


Release

R2019a

Community Treasure Hunt

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

Start Hunting!