filter function - delay at the first run

Hi all, I have a question related to the filter function. I noticed a delay the first time this function is called for instance in a for loop.

Let's start from this code:

 function y = dummy(x)

    y = zeros(size(x));

    [b, a] = cheby1(4, 0.5, 0.1);

    for i = 1:size(x, 2)
       tic
       y(:, i) = filter(b, a, x(:, i));
       toc
    end
 end

with x = rand(10000, 64);

Every time the function dummy is called, the computational time at the first iteration (i==1) is higher (e.g. ~50 ms in my workstation).

Do you any idea why it happens? There is a way to avoid it?

Thanks

Luca

Answer 1

[Sean de Wolski]

That is likely the JIT compiler taking a little extra time on the first iteration but then being much slower.

What does the timing look like if you turn the JIT off?

feature accel off

Answer 2

[Jan]

In the line:

y(:, i) = filter(b, a, x(:, i));

memory is reserved for x(:,1). Perhaps this memory is reused later by the smart JIT? This is a pure guessing.

Just for curiosity: Do you know that the loop over i is not necessary and that:

y = filter(b, a, x)

would use all cores for multi-threading, because x has more than 16 columns?

If this task is time critical, you could use the faster FEX: FilterM