parfor loop
Parfor loops behave similarly to regular for loops except that they attempt to use multiple processors or threads to divide the loop space into chunks. The chunks are equally-sized (or as equal as possible).
The number of chunks is determined by the value set by the parcluster command. Your computer configuration will determine the number of chunks that can be processed in parallel. For example, if your machine only has two cores, setting parcluster(64) will cause the loop to be executed in 64 chunks, but only two of those will occur simultaneously.
Display of variable values is disabled during a parfor loop.
Special types of variables are inside a parfor loop:
Loop Variable
parfor j=1:64
end
Where j
must be continuous.Broadcast Variable
b = 5;
parfor j=1:64
b = j;
end
b
Where b will be printed in the command window as
5, because broadcast variables are readable within the
parfor loop, but any modifications to them are discarded once the loop
ends. Local Variable
parfor j=1:64
a = j;
end
a
In this case, a will cause an error as an undefined variable,
because local variables only exist within the parfor loop and are not
accessible after the loop completes itself. Sliced Variable
a = zeros(1,64);
parfor j=1:64
a(j) = j;
end
a
Where a will be printed with each of its elements having the
value that has been assigned to it within the parfor loop. Sliced variables
are indexed by the loop variable and each loop then processes a different part of the sliced
variable.For speed and safety, preallocate sliced variables using the
zeros
function or a similar method.
Reduced Variable
sum = 0;
parfor j=1:64
sum = sum + j;
end
sum
Where sum will be printed in the command window as 2080.
Reduced variables are computed via calculations using the loop variable (directly or
indirectly). Only + and * are supported operations for reduced variables (as they are
commutative).sum = 0;
parfor j=1:64
sumsq = sumsq + times(j, j);
end
sumsq
Where sumsq will be printed in the command window as
89440. The reduced operation(s) can be functions of the loop variable. Limitations
The behavior of global variables in a parfor loop is not well-defined since multiple loops may be modifying the variable concurrently.Certain functions are not thread-safe and will cause issues in a parfor loop. Currently there is not an exhaustive list of such functions, but one example is readmultvectors and any other function that accesses a shared resource (for example, xlswrite) also has potential to be problematic.