Accelerating MATLAB Performance

273Implicit Parallelization (Vectorization and Indexing)

% Slow recursive version

fibonacci = [0, 1];

for idx = 3 : 1000

fibonacci(idx) = fibonacci(idx-1) + fibonacci(idx-2);

end

⇨ Elapsed time is 0.000562 seconds.

% Vectorized version: 10x faster

x = [0,1,zeros(1,98)];

fibonacci = filter(1, [1,-1,-1], x);

⇨ Elapsed time is 0.000059 seconds.

In another case, using lter improved a loop performance by 180× (!):

836

nobs = 1e5; % number of observations

p = 2; % number of lags

ro = [0.1 0.5]; % autoregressive coefficients

noise = rand(nobs,1)-p; % white noise

pseudoX = zeros(nobs,1); % vector to fill in

pseudoX(1:p) =