memory allocated with mxMalloc

Question

rick 2011-9-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/16208-memory-allocated-with-mxmalloc

Suppose within an m-file I use a mex file to allocate persistent memory using mxMalloc and mxMakeMemoryPersistent.

When the mex file returns, I then start a batch job and pass the pointer returned by the mex file to the batch job.

Will the batch job be able to access this memory?

More info: A GUI is used to control hardware components which write to the mxMalloc'ed memory block, I would like the batch job to save the data written there in parallel. Don't worry, synchronization will be taken care of.

3 个评论
显示 1更早的评论隐藏 1更早的评论

rick 2011-9-20

yes, my confusion arises with the matlab memory manager, I couldn't find any documentation as to how it manages multiple matlab processes, i.e. does each process have its own address space managed by separate managers or does one memory manager take care of all processes...thereby allowing memory to be shared (if it was 'smart' and realized a pointer to allocated memory was being passed to another process)

Peter Li 2011-9-23

Ah sorry, I didn't realize this was specific to interprocess calls. In that case Edric's suggestion sounds like the way to go.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Edric Ellis 2011-9-20

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/16208-memory-allocated-with-mxmalloc#answer_21949

Do you mean a 'batch job' created using the BATCH function in Parallel Computing Toolbox? If so, then no, the workers executing the batch job will not have access to the memory as they run in a completely separate MATLAB process.

You might wish to check out 'sharedmatrix' from the File Exchange:

http://www.mathworks.com/matlabcentral/fileexchange/28572-sharedmatrix

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

rick 2011-9-20

That's what I was thinking, however I couldn't find documentation on how separate matlab processes were handled by the memory manager. Unfortunately it seems that at the moment Matlab's pct is only setup for simple parallelism.

请先登录，再进行评论。