GPU computing monte carlo simulation

i don't know why this code calculation time is same with cpu calculation

How can I make it calculate at gpu?

t=tic(); % measure the calculation time
P = [0.1 0.2 0.3 0.4 0.5]; % Probability of basic events
m = 0;
N = 1000; % Num of iteration
A = ones(N,1,'gpuArray'); % GPU calculation
for n = 1:N % iteration
    for i = 1:numel(P)
        r = rand();
    if r <= P(i) % sampling
        S(i) = true;
    else
        S(i) = false;
    end
    end
    % gate # GAB
    Sg = DetermineState (20,S(2),S(3));
    Sg1 = Sg;
    % gate # GAC
    Sg = DetermineState (20,S(4),S(5));
    Sg2 = Sg;
    % gate # A
    Sg = DetermineState2 (10,S(1),Sg1,Sg2);
    St = Sg;
if St == true
    m = m+1;
end
end
m = gather(count);
P = m/N;
value = P
gpuTime = toc(t)

DetermineState function is like this. But I think it is not problem.

function [Sg] = DetermineState(varargin)
% OR Gate
if varargin{1} == 10
    if max(varargin{2},varargin{3}) == 0
        u = 0;
    else
        u = 1;
    end
    % AND Gate
elseif varargin{2} == 20
    if min(varargin{2},varargin{3}) == 0
        u = 0;
    else
        u = 1;
    end
    % NOT Gate
else
    u = 0;
end
Sg = u;
end

[Walter Roberson]

    for i = 1:numel(P)
        r = rand();
    if r <= P(i) % sampling
        S(i) = true;
    else
        S(i) = false;
    end
    end

Is there a reason you did not vectorize that as

    S = reshape(rand(size(P)) <= P, 1, []);

[Raymond Norris]

My sense is that this might not be the complete code, for several reasons.

1. A is a gpuArray, but nowhere are you using it. Therefore, everything is on the CPU.
2. Towards the end is the code

if St == true
    m = m+1;
end
end
m = gather(count);

However, count is not assigned anywhere. I'm also puzzled why you'd assign m in the for-loop, but then override it when you assign it to the gathering of count (which presumable is on the GPU).

3. I'm gathering that this

Sg = DetermineState2 (10,S(1),Sg1,Sg2);

should be

Sg = DetermineState (10,S(1),Sg1,Sg2);

DetermineState2 isn't provided.

One last note (and again, it might be because the code sample isn't complete), the code as written takes almost no time (I realize that's all relative and perhaps this needs to be called many time), so this might not perform well on a GPU. GPU calculations perform much better when lots of (somewhat large) work is being assigned to it.