MATLAB Answers

Fast Element-Wise power

25 views (last 30 days)
Thomas Bauer
Thomas Bauer on 30 Oct 2017
Hello all,
I am trying to optimize some code that contains element wise calculations with rather large matrices. I have already shaved off about 50% with things like the bsxfun (which in my case works wonders). But now I am stuck at an operation where a simple line of code value=matrix.^3 takes up most of the runtime of the script. is there any way to perform this element-wise calculation faster?
Best regards

  1 Comment

Thomas Bauer
Thomas Bauer on 30 Oct 2017
Hello, i think i just answered my own question. Appearently using
value=matrix.*matrix.*matrix
runs significantly faster.
BTW i also figured out bsxfun was NOT the reason for my faster running script. It actually was the termination of several operations by precalculating another variable.

Sign in to comment.

Accepted Answer

KL
KL on 30 Oct 2017
Edited: KL on 30 Oct 2017
a=1:10000000;
b=repmat(3,size(a));
tic
c1=a.^b;
toc
tic
c2=a.^3;
toc
tic
c3=a.*a.*a;
toc
And then the results are,
Elapsed time is 0.118509 seconds.
Elapsed time is 0.129281 seconds.
Elapsed time is 0.006726 seconds.

  2 Comments

Thomas Bauer
Thomas Bauer on 30 Oct 2017
Thank you for this example!
Johannes Kalliauer
Johannes Kalliauer on 1 Apr 2019
sometimes if you have small integers you can speed it up by useing integers
a=randi([0,6],1,10000000);
b=repmat(3,size(a));
int8a=uint8(a);
tic
c1=a.^b;
toc
tic
c2=a.^3;
toc
tic
c3=a.*a.*a;
toc
tic
c4=int8a.*int8a.*int8a;
toc
And then the results are,
Elapsed time is 0.199174 seconds.
Elapsed time is 0.153210 seconds.
Elapsed time is 0.014326 seconds.
Elapsed time is 0.002569 seconds.

Sign in to comment.

More Answers (0)