You may have heard that vectorizing your code is a cornerstone of achieving the best performance from both MATLAB® and Star-P code. But what does "vectorizing" mean? Simply put, vectorizing your code means that you avoid using “for” loops to operate on individual elements of a vector or array. Instead, you operate upon vectors and arrays as entire units using MATLAB's built-in functions. The following are two quick examples demonstrating ways to vectorize in common situations. (The lines which change upon vectorization have been highlighted in blue.)

First, consider a simple example illustrating what vectorization is about. Instead of calculating each element of a matrix using nested loops, here can calculate it quickly by operating on the entire matrix v at once:

NOT Vectorized:

% Bad: calculate matrix with for loop
% Note: may take hours if not vectorized!
v=rand(1000*p);
vs=zeros(1000*p);
for i=1:size(v,1)
  for j=1:size(v,2)
    vs(i,j) = sin(v(i,j));
  end
end

Vectorized:

% Good: calculate matrix with built-in
v=rand(1000*p);
vs_vec = sin (v);

Next, let’s consider a relatively common occurrence: a conditional statement inside a for loop.

1. Instead of checking the condition "less than zero" for each element by looping (top), a single vectorized sweep (bottom) is carried out. This yields a vector 'idx' which holds a value 1 (true) or 0 (false) for each element.
2. Vectorized logical indexing is performed (bottom). An element a(i) is set to zero if 'idx(i)' is true.

NOT Vectorized:

% Bad: conditional inside for loop
numpoints = 10000;
a = 2*rand(numpoints*p, 1)-1;
for i = 1:numpoints
if (a(i) < 0)
a(i) = 0;
end
end

Vectorized:

% Good: vectorized conditional
numpoints = 10000;
a = 2*rand(numpoints*p, 1)-1;
idx = (a < 0);
a(idx) = 0;