Unlike a parallel for-loop (parfor
), which takes the iterations of a loop and distributes them among multiple threads, a single program, multiple data (spmd
) statement takes a series of commands and distributes them to all the threads, so that each thread performs the command and stores the results. Consider this:
poolobj = parpool(2); % open a parallel pool with two workers
spmd
q = rand(3); % each thread generates a unique 3x3 array of random numbers
end
q{1} % q is called like a cell vector
q{2} % the values stored in each thread may be accessed by their index
delete(poolobj) % if the pool is closed, then the data in q will no longer be accessible
It is important to note that each thread may be accessed during the spmd
block by its thread index (also called lab index, or labindex
):
poolobj = parpool(2); % open a parallel pool with two workers
spmd
q = rand(labindex + 1); % each thread generates a unique array of random numbers
end
size(q{1}) % the size of q{1} is 2x2
size(q{2}) % the size of q{2} is 3x3
delete(poolobj) % q is no longer accessible
In both examples, q
is a composite object, which may be initialized with the command q = Composite()
. It is important to note that composite objects are only accessible while the pool is running.