MATLAB Answers

how to parallelize separate matrix-matrix multiplications?

8 views (last 30 days)
Yang on 25 Jan 2018
Commented: Yang on 26 Jan 2018
There are three matrix-matrix multiplications: C1=A1*B1, C2=A2*B2, C3=A3*B3.
If I wrote the code like above, the three matrix-matrix multiplications will be calculated in a sequential way. Is there an easy way to parallelize the three matrix-matrix multiplications (apart from using parfor)?
Apology if this question has been raised&answered before. I didn't find such a question after some quick search.


Sign in to comment.

Accepted Answer

Brendan Hamm
Brendan Hamm on 25 Jan 2018
Edited: Brendan Hamm on 25 Jan 2018
Unless you are working on a cluster, I would not expect to see any speed up from parallelizing such operations. The reason for this is that a matrix multiplication is multi-threaded, meaning that each of these computations is already utilizing multiple cores on your machine.
If you are working on a cluster, you can assign each of these variables in an SPMD block and perform the computation on each machine separately.


Jan on 25 Jan 2018
For small matrices only one thread is used. If you have to multiply many of them, a parfor can be useful.
Brendan Hamm
Brendan Hamm on 25 Jan 2018
This is a good point and useful for others who might happen upon this question, as according to Yang, there may not be similar posts. I was considering his question alone, but this is a better generalization of the concept.
Also, I edited the work hyper-threaded to multi-threaded, so as not to be hardware specific.

Sign in to comment.

More Answers (0)