K_UL and Kcel are matrices.
M1=4;
M2=4;
nK = size(K_UL,2);
Us=zeros(ndof,nK);
for x=-M1:1:0
for y=-M2:1:M2-1
fx=1i*pi*x/M1;
fy=1i*pi*y/M2;
u=Myfun(Kcel,M1,M2,fx,fy);
Us = Us+(1+(x>-M1&x
Thanks for your suggestion Harald Kirchsteiger
But, here I am using a function Myfun inside the loop. Which depends on a matrix (Kcell). Can we use such type of predefine function inside the parallel loop?
I would start from vectorizing your code, vectorization often works better than parfor
https://stackoverflow.com/questions/14035365/why-does-vectorized-code-run-faster-than-for-loops-in-matlab
The vector version (needs testing) is as follows:
fx=1i*pi*[-M1:1:0]/M1;
fy=1i*pi*[-M2:1:M2-1]/M2;
u=Myfun(Kcel,M1,M2,fx,fy);
calculate values of
1+(x>-M1&x
As a primer for the further work please consider the following code prepared for testing of a single element from your loop:
tic
for x=-M1:1:0
for y=-M2:1:M2-1
fx=1i*pi*x/M1;
fy=1i*pi*y/M2;
end
end
toc
time_loop=toc
tic
fx=1i*pi*[-M1:1:0]/M1;
fy=1i*pi*[-M2:1:M2-1]/M2;
toc
time_vec=toc
speedup=time_loop/time_vec
The speedup is of the order of
5 for M1=4, M2=4
260 for M1=1000, M2=500
19300 for M1=10000, M2=5000
- Badges
- Science topic