Matlab timer is not executed asynchronously
12 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
Yu-Chih Tung
el 29 de Mzo. de 2017
Comentada: raym
el 11 de Jul. de 2018
Hi, I am developing a Matlab UI program where the callback of a button pressed might take lots of time to process. So I basically want the callback to be executed asynchronously and ask the UI to update the callback result when it is ready.
Some documents suggests doing this by using Matlab timer callback but I found the callback is still not executed asynchronously.
Here is a simple code for testing:
function [ ] = ThreadTest( )
t = timer('StartDelay',0.1, 'TimerFcn',@utilityFcn);
start(t);
for i=1:10000;
pause(0.1);
fprintf('x');
end
end
function utilityFcn(src,evt)
% some lengthy calculation/update done here
for i=1:100;
temp = rand(1000,1000)*rand(1000,1000);
end
fprintf('done');
end
I expect this program will continuously print 'x' in every 0.1 sec and there will be a print of 'done' when the callback is finished. However, in this example, the 'x' is only printed when the all lengthy computation is done (i.e., always printed after the 'done' is printed)
0 comentarios
Respuesta aceptada
Walter Roberson
el 29 de Mzo. de 2017
The details are more complicated than Jan indicates. Multiple threads are used in some cases.
"For example, a timer object is in a different thread than the MATLAB main thread. But it cannot interrupt a single MATLAB command i.e. a variable copy command.
A block of MATLAB code can be interrupted by a timer object. Thus if a block of code operates on a variable, there is a possibility of data corruption.
However, Just-in-Time compilation (JIT) can execute several lines of code all at once in certain circumstances, in which case the timer will not be able to interrupt execution. If the interruption must happen in a particular place, a DRAWNOW command will ensure that one can occur. Otherwise, the DRAWNOW is not needed."
In your example possibly everything in your "for" loop got compiled into a single JIT block.
We are positive that internal calls to BLAS or LINPAK cannot be interrupted by timers. We believe (but are not certain) that no one source line of MATLAB code can be interrupted by timers. We are rather less certain about when multiple-line blocks can be interrupted by timers.
2 comentarios
raym
el 11 de Jul. de 2018
Even a timerfunc of single line: system('notepad') will block the cmd window of matlab.
Más respuestas (1)
Jan
el 29 de Mzo. de 2017
Editada: Jan
el 29 de Mzo. de 2017
Note what would happen, if the timer would be asynchronous:
function [ ] = ThreadTest( )
global x
t = timer('StartDelay', 1, 'TimerFcn', @utilityFcn);
start(t);
for i = 1:100;
pause(0.1);
x = x + 1;
fprintf('%d ', x);
end
end
function utilityFcn(src,evt)
global x
x = x + 1000;
end
Now there would be a racing condition between the main thread and the timer callback: The main thread gets the value of x and increases it by 1. If now the timer thread does exactly the same, it is not predictable which thread writes back the new value at first.
Unpredictable states are extremely bad for programming. And therefore Matlab is single threaded currently. [EDITED: added] This means that there cannot run code asynchronously, neither as a timer callback nor by using C-Mex tricks like mexCallMATLAB from several threads.
But what about parfor loops? [EDITED end]
The example might look artifical, but the real world code is much more complicated and collisions are harder to find, but this does not mean that they do not occur. It can concern persistent variables, ApplicationData of GUI objects, the state of mutual exclusive radio buttons, writing to files, etc.
2 comentarios
Jan
el 29 de Mzo. de 2017
Editada: Jan
el 29 de Mzo. de 2017
@Yu-Chih Tung: Matlab is single-threaded. You cannot run code asynchronusly. If you try this using tricks in the MEX interface, Matlab crashs as expected.
So my answer is: This will not work. Matlab is not prepared to handle e.g. the mentioned racing conditions in its internal functions.
parfor loops run asynchronously. I do not have teh parallel processing toolbox, so wait if anybody else posts a corresponding solution.
Ver también
Categorías
Más información sobre Startup and Shutdown en Help Center y File Exchange.
Productos
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!