Calling matlab::engine::MATLABEngine::feval from multiple threads
9 vues (au cours des 30 derniers jours)
I just discovered the new mex C++ API and immediately fell in love with it! But what's about thread safety of the matlab::engine::MATLABEngine::feval call in particular and the whole API in general? The docs don't say anything regarding this topic.
I have an algorithm that filters a lot of signals, which could gain a lot from parallelization. So ideally I'd like to create as much std::threads as there are independent signals and process them all in parallel with the MATLAB engine. Is that possible? I assume that the interpreter invoked by feval will be probably single-threaded, however maybe the execution of the function after parsing is capable of multithreading? Or should I go on and perform multithreaded filtering completely in the C++ domain using my own filter classes and go back to the MATLAB interface after having joined all threads?
What about the Data API? Is it safe to access (readonly) a matlab::data::Array from multiple threads in parallel?
I'm looking forwards to your answers!
Sangeetha Jayaprakash le 25 Mai 2018
The matlab::engine::MATLABEngine::fevalAsync function is thread safe i.e. if there are multiple threads in the C++ application, all of them can access the "fevalAsync" function.
Please note that this still means that only one of those requests(from any one of the threads) will be actually run on the MATLAB engine instance at any given time.