How to resolve warning message about broadcast variable while using 'parfor'?
7 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
Atanu
el 14 de Jul. de 2024
Comentada: Atanu
el 16 de Jul. de 2024
males = {'animal1', 'animal2', 'animal3'};
treatmentGrps = {'T1', 'T2', 'T3'};
nrows = size(featureLists, 1);
ncols = size(featureLists, 2);
featureLists = cell(nrows, ncols);
parfor grp = 1:nrows
tempRow = cell(1, ncols);
for animal = 1:ncols
tempRow{animal} = individualPsychValuesPerSession('approachavoid', ...
treatmentGrps{grp}, males{animal});
end
featureLists(grp, :) = tempRow;
end
When I use 'parfor' like above I get an warning message
The entire array or structure 'males' is a broadcast variable.
This might result in unnecessary communication overhead.
But, when I change 'ncols' to 'numel(males)' like following
for animal = 1:numel(males)
tempRow{animal} = individualPsychValuesPerSession('approachavoid', ...
treatmentGrps{grp}, males{animal});
end
for the inner loop I don't get any warning.
Can someone please explain this behavior of 'parfor'?
0 comentarios
Respuesta aceptada
Edric Ellis
el 15 de Jul. de 2024
I think you should ignore this message in this case. It looks like you really need all the values of males for each iteration of the parfor loop, and that's OK. The message is there to warn about cases where the data is being "accidentally" broadcast.
The fact is that the array males is very small (in terms of bytes), so it should not cause significant overhead. You can check using whos to see the size of the array in the workspace, and also use ticBytes and tocBytes to see how much data is being transferred to run the loop. My bet is that the overall data transfer is not being significantly changed by the size of males.
Más respuestas (1)
Drew
el 14 de Jul. de 2024
Editada: Walter Roberson
el 14 de Jul. de 2024
See https://www.mathworks.com/help/parallel-computing/broadcast-variable.html It includes an example with numel().
If this answer helps you, please remember to accept the answer
2 comentarios
Walter Roberson
el 14 de Jul. de 2024
It was due to the ending period being interpreted as part of the URL. I have corrected the link.
Ver también
Categorías
Más información sobre Parallel for-Loops (parfor) en Help Center y File Exchange.
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!