How can I adjust the number of threads for functions based on fftw?

As far as I know, embadded matlab function 'dct' is based on fftw library, and it supports 'Thread-based Environment'.

And I noticed MATLAB proccess automatically use CPU about ~1000% when I use dct function, in thread-based parallel environment, without any other instruction.

How can the function decide the proper number of threads in fft ?

Can I set the number of threads manually to improve the performance?

Thanks.

Answer 1

[Angelo Yeo]

You can determine the maximum number of threads with the function "maxNumCompThreads".

See the doc below for more information.

https://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html

[현탁 박]

Thanks.

But I want to know how to set 'exact' threads which the dct function used, not the maximum threads.

When I use fftw in c code, I have to set the number of threads manually via

void fftw_plan_with_nthreads(int nthreads);

And FFTW team suggest that test various trial nthreads to find a effective thread number.

I want to follow that routine, but I cannot even configure the exact threads fftw used, and that's the problem.

[Angelo Yeo]

@현탁 박: Would you try to set the number of workers for thread environment with the following commands?

numWorkers = 2;
parpool("Threads", numWorkers)

Note that this feature was introduced from R2022b, and details can be found in the link below.

Release Notes for Parallel Computing Toolbox - MATLAB & Simulink (mathworks.com)