Re: Skript takes in matlab 6s in Octave never ends

[Kai Torben Ohlhus]

On Fri, Mar 8, 2019 at 6:36 AM khalil2535 <address@hidden> wrote:

After few debugging that's what works fast in matlab that never ends in
octave

clear;close;clc;

load('data.mat');
tic
for i = 2:nrTime
  v_i(:, i) = v_i(:, i-1) + a_i(:, i) * sampleTime; % euler integrator
  p_i(:, i) = p_i(:, i-1) + v_i(:, i) * sampleTime; % euler integrator
  end
toc



i don't have an idea how do matlab make loop-based code works that fast ,
maybe it vertorize it somehow

As you say, in Octave you should in general try to avoid for-loops in favor of vectoriztation.

Octave has kind of support for JIT-optimization as Matlab has, but in addition Octave

has many overlooked vectorization functions.

For example your code runs for me t1/t2 = 22 times faster by using `cumsum`

and even in Matlab you gain  between 1.5 or 2 speedup ;-)

clear;close;clc;

M = 5e2;

nrTime = 2e2;

sampleTime = 5;

v_i_orig = rand (M, nrTime);

p_i_orig = rand (M, nrTime);

a_i = rand (M, nrTime);

v_i = v_i_orig;

p_i = p_i_orig;

tic

for i = 2:nrTime

  v_i(:, i) = v_i(:, i-1) + a_i(:, i) * sampleTime; % euler integrator

  p_i(:, i) = p_i(:, i-1) + v_i(:, i) * sampleTime; % euler integrator

end

t1 = toc;

v_i_ref = v_i;

p_i_ref = p_i;

v_i = v_i_orig;

p_i = p_i_orig;

tic

i = 2:nrTime;

v_i(:, i) = a_i(:, i) * sampleTime; % euler integrator

v_i = cumsum (v_i, 2);

p_i(:, i) = v_i(:, i) * sampleTime; % euler integrator

p_i = cumsum (p_i, 2);

t2 = toc;

% Ensure both computation result in the same result.

assert (isequal (v_i_ref, v_i))

assert (isequal (p_i_ref, p_i))

t1/t2

HTH,

Kai