In some cases, you must modify the code to convert
parfor-loops. This example shows how to diagnose and fix
parfor-loop problems using a simple nested
for-loop. Run this code in MATLAB® and examine the results.
for x = 0:0.1:1 for y = 2:10 A(y) = A(y-1) + y; end end
To speed up the code, try to convert the
parfor-loops. Observe that this code produces
parfor x = 0:0.1:1 parfor y = 2:10 A(y) = A(y-1) + y; end end
In this case you cannot simply convert the
parfor-loops without modification. To make this work, you must
change the code in several places. To diagnose the problems, look for Code Analyzer
messages in the MATLAB Editor.
This code shows common problems when you try to convert
To solve these problems, you must modify the code to use
The body of the
parfor-loop is executed in a parallel pool using
multiple MATLAB workers in a nondeterministic order. Therefore, you have to meet these
requirements for the body of the
The body of the
parfor-loop must be independent. One
loop iteration cannot depend on a previous iteration, because the iterations
are executed in parallel in a nondeterministic order. In the
A(y) = A(y-1) + y;
parfor. For next steps in dealing with independence issues, see Ensure That parfor-Loop Iterations are Independent.
You cannot nest a
parfor-loop inside another
parfor-loop. The example has two nested
for-loops, and therefore you can replace only one
for-loop with a
Instead, you can call a function that uses a
parfor-loop inside the body of the other
parfor-loop. However, such nested
parfor-loops give you no computational benefit,
because all workers are used to parallelize the outermost loop. For help
dealing with nested loops, see Nested parfor and for-Loops and Other parfor Requirements.
parfor-loop variables must be consecutive increasing
integers. In the example,
parfor x = 0:0.1:1
parforhere. You can solve this problem by changing the value of the loop variable to integer values required by the algorithm. For next steps in troubleshooting
parfor-loop variables, see Ensure That parfor-Loop Variables Are Consecutive Increasing Integers.
You cannot break out of a
parfor-loop early, as you
can in a
for-loop. Do not include a return or break
statement in the body of your
communication, the other MATLAB instances running the loop do not know when
to stop. As an alternative, consider
If you still have problems converting
parfor-loops, see Troubleshoot Variables in parfor-Loops.
You can profile a
toc to measure the speedup
compared to the corresponding
tocBytes to measure how much
data is transferred to and from the workers in the parallel pool. For more
information and examples, see Profiling parfor-loops.