Graphics argument validation for function in parfor loop
Show older comments
I have a function that I'd like to run serially but also sometimes inside a parfor loop, with the ability to optionally plot output. Understandably, I can't plot or modify graphics within a parpool thread, but I find even having unused relevant argument validation that draws from matlab.graphics.axis.Axes class properties causes a parallel:threadpool:NonConcurrentLibrary error to be thrown during parallel execution. Attached is a code snippet demonstrating what I mean.
I suppose I have two questions. First, is this expected behaviour, given that args is not even invoked in the body of the function? Second, is there any way to retain automatic argument validation for an axis object that plays nicely with parfor, in cases where these arguments are used conditionally and never during parallel execution?
function ArgValidationParForTest()
pool = gcp();
fprintf('Parallel pool initialised with %d workers\n', pool.NumWorkers)
ntests = 10;
parfor i = 1:ntests
TestFunc()
end
end
function TestFunc(args)
arguments
args.?matlab.graphics.axis.Axes
end
fprintf('Called\n')
end
Accepted Answer
If you use a process-based pool, it will work --
parpool('Processes')
More Answers (1)
Second, is there any way to retain automatic argument validation for an axis object that plays nicely with parfor, in cases where these arguments are used conditionally and never during parallel execution?
You would probably have to hand off to a separate validation routine,
function TestFunc(varargin)
fprintf('Called\n')
args=handOff(varargin{:});
end
function args=handOff(varargin)
p = gcp('nocreate');
args=[];
if isempty(p) || ~isa(p,'parallel.ThreadPool')
args=validateGraphics(varargin{:});
elseif ~isempty(varargin)
error 'Handle graphics aren''t supported in threadpools'
end
end
function args=validateGraphics(args)
arguments
args.?matlab.graphics.axis.Axes
end
end
2 Comments
Jake
25 minutes ago
You can recover some of the benefits of CLI by adding an explicit argument list, i.e., by copy-pasting the following auto-generated argument list into your TestFunc arguments block. This block knows only the names of the name/value parameters, and not any of the defaut values and type restrictions. So you would still need to use the handOff function to do the validation proper. However, it at least it will complete the parameter names for you.
disp(char( " args."+properties("matlab.graphics.axis.Axes") ))
Categories
Find more on Parallel Computing Fundamentals in Help Center and File Exchange
