How to reduce the overall cost(Storage as well as communication) of transfer a dense matrix in nested functions

Suppose I have a dense matrix in the very first stage of my code and I need to tranfer that matrix into the function again and again and just need a matrix vector multiplication inside the function as shown below

clc

Matrix A= 100000×100000 (dense)

Program starts

{

Function1

{

Function2

{

Here I need Matrix multiplication with some vector (M.a = b)....... [a,b = vectors]

}

}

}

How can I tranfer this matrix efficiently in the code and get matrix-vector multiplication and How can I store it efficiently?

Answer 1

[Arjun]

Hi @Nadeem Ahmed,

I understand that you want to efficiently store and transfer dense matrices in your programs.

Here are some suggestions which might help you:

• To avoid repeated transfers, you can make the matrix a “global” variable; however, this must be used sparingly and specific to the application. Refer to the documentation of “global” : https://www.mathworks.com/help/matlab/ref/global.html
• You can also use “persistent” storage to preserve a variable inside a function between repeated function calls. To use “persistent” efficiently, refer to the documentation of “persistent”: https://www.mathworks.com/help/matlab/ref/persistent.html
• To store matrix efficiently, you must ensure the correct datatype to use which can save space since storing a double is costlier compared to “int32” data type, if entries are single precision, then use “int32” or “single” instead of “double”.
• You must use optimized function like “A*a” for efficient matrix vector multiplication to boost up speed as the internal implementations are highly optimized for performance.

You can try implementing the above suggestions to the problem.

I hope this will help!