memmapfile in parfor loops
8 views (last 30 days)
Show older comments
Hello, I'm trying to use mem mapped files for input and output of data from the body of a parloop.
If I understand correctly, this seems impossible, as the way memmapfile data is accessed (struct) violates the 'first indexing' rule for sliced variables in parallel loops.
As the memmapfile can be accessed by concurrent processes (as in the demo), is there a workaround to use it in parfor loops?
Thanks. L
0 Comments
Answers (1)
Sarah Wait Zaranek
on 11 Nov 2011
You should be able to use memmapfile in a parfor loop. The trick is to pass m into a function within the parfor loop. You can get around the restriction that way. See simple example below:
function parMemMap
% randData = gallery('uniformdata', [100, 1], 0, 'double');
% fid = fopen('myfile.dat');
% fwrite(fid, randData, 'double');
% fclose(fid);
m = memmapfile('myfile.dat','Format', 'double','Writable',true);
parfor ii=1:length(m.Data)
output(ii,1) = myTestFunc(m,ii);
end
data = m.Data(:);
disp(output)
isequal(output,data)
end
function test=myTestFunc(m,ii)
m.Data(ii) = m.Data(ii)*rand;
test = m.Data(ii);
end
3 Comments
Keith
on 10 Feb 2024
11/11/11! Suppose there's a need to share data not only inside a parfor but between ML instances? Will a memmapfile, specifiying the desired disk file, in one thread facilitate reading that disk file as written by another(cooperating) thread?
Walter Roberson
on 10 Feb 2024
Will a memmapfile, specifiying the desired disk file, in one thread facilitate reading that disk file as written by another(cooperating) thread?
No.
However, a memmapfile in one thread could obtain data that you exchange with the client using parallel.pool.DataQueue and your client could then make that data available to another thread using a different set of parallel data queues. It is a somewhat ackward process.
See Also
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!