parallel.pool.Constant

Build parallel.pool.Constant from data or function handle

Syntax

C = parallel.pool.Constant(X) C = parallel.pool.Constant(FH) C = parallel.pool.Constant(FH,CLEANUP) C = parallel.pool.Constant(COMP)

Description

C = parallel.pool.Constant(X) copies the value X to each worker and returns a parallel.pool.Constant object, C, which allows each worker to access the value X within a parallel language construct (parfor, spmd, parfeval) using the property C.Value. This can improve performance when you have multiple parfor-loops accessing the same constant set of data, because X is transferred only once to the workers.

C = parallel.pool.Constant(FH) evaluates function handle FH on each worker and stores the result in C.Value. This is also useful for creating and using any handle-type resources on the workers, such as file handles and database connections.

C = parallel.pool.Constant(FH,CLEANUP) evaluates function handle FH on each worker and stores the result in C.Value. When C is cleared, the function handle CLEANUP is evaluated with a single argument C.Value on each worker.

C = parallel.pool.Constant(COMP) uses the values stored in the Composite COMP, and stores them in C.Value on each worker. This is especially useful when the data that you need to use inside a parfor-loop can be constructed only on the workers, such as when the data is too large to conveniently fit in the client, or when it is being loaded from a file that only the workers can access. If COMP does not have a defined value on every worker, an error results.

Tips

parallel.pool.Constant must be called in the MATLAB client session.

You can use a parallel.pool.Constant with an already running parallel pool or subsequent parallel pools.

Examples

Make parallel.pool.Constant from array in client

This example shows how to create a numeric parallel.pool.Constant, and use it in multiple parfor-loops on the same pool.

First, create some large data on the client, then build a parallel.pool.Constant, transferring the data to the pool only once.

data = rand(1000);
c = parallel.pool.Constant(data);
for ii = 1:10
    % Run multiple PARFOR loops accessing the data.
    parfor jj = 1:10
        x(ii,jj) = c.Value(ii,jj);
    end
end

Make parallel.pool.Constant from function handle

This example shows how to create a parallel.pool.Constant with a function handle and a cleanup function.

Create a temporary file on each worker. By passing @fclose as the second argument, the file is automatically closed when c goes out of scope.

c = parallel.pool.Constant(@() fopen(tempname(pwd),'wt'),@fclose);
spmd
   disp(fopen(c.Value)); % Displays the temporary filenames.
end

parfor idx = 1:1000
   fprintf(c.Value,'Iteration: %d\n',idx);
end
clear c;   % Closes the temporary files.

Make parallel.pool.Constant from Composite

This example shows how to build large data sets as a Composite on pool workers inside an spmd block, and then use that data as a parallel.pool.Constant inside a parfor-loop.

spmd
  if labindex == 1
    x = labBroadcast(1,rand(5000));
  else
    x = labBroadcast(1);
  end
end
xc = parallel.pool.Constant(x);
parfor idx = 1:10
  s(idx) = sum(xc.Value(:,idx));
end
s

s =
   1.0e+03 *

    2.5108    2.5031    2.5123    2.4909    2.4957    2.5462    2.4859    2.5320    2.5076    2.5432

Documentation