How increase calculate speed in for loop

Hi all.

I have a problem in my code about too long calculate time.

This is my part of code

=============================================

Depth=5000;
Num_Alines=400;
Num_Bscan=300;
Alines=180
half=Alines/2;
ReconImage=zeros(Depth,Num_Alines,Num_Bscan);

for i=1:Depth
    Sampledist2=Sampledist;
    Sampledist2(Sampledist2==Sampledist2(i,round(half),round(half)))=1;
    Sampledist2(Sampledist2~=1)=0;
    Sampledist2=Sampledist2*D313; %D313 is one data not array.
    for j=round(half):Num_Alines-round(half)
        for k=round(half):Num_Bscan-round(half)
            ReconImage(:,1+j-round(half):j+round(half),1+k-round(half):k+round(half))=ReconImage(:,1+j-round(half):j+round(half),1+k-round(half):k+round(half))+fftimage(i,j,k).*Sampledist2/1000;
        end
    end
end

=============================================================================

In my code, Sampledist is (Depth,Alines,Alines) size aray.

Ultimately, I want sum "fftimage(i,j,k).*Sampledist2/1000" to "ReconImage" array. But in my code, 'for k=~' part spend 0.8 sec during one cycle. So actually, expected spend time is up to ten thousands hours.

I never agree my code is the best.

When i search to increase calculation, i found 'parfor' but i can't apply that because for loop refer the result.(underline)

Anyone have nice idea?

Thank you.

Kim.

[Jan]

What is "Sampledist"?

Answer 1

[Jan]

Start with a simplification of the code:

Depth      = 5000;
Num_Alines = 400;
Num_Bscan  = 300;
Alines     = 180
h          = round(Alines / 2);  % Call round() once only
R          = zeros(Depth, Num_Alines, Num_Bscan);
for i = 1:Depth
    S = Sampledist;
    S(S == S(i, h, h)) = 1;
    S(S ~= 1) = 0;
    S = S * D313 / 1000;
    for j = h:Num_Alines - h
        for k = h:Num_Bscan - h
            R(:, 1+j-h:j+h, 1+k-h:k+h) = ...
                R(:, 1+j-h:j+h, 1+k-h:k+h) + fftimage(i,j,k) .* S;
        end
    end
end

Please provide some values for Sampledist and fftimage - maybe produced by rand().

[kisik KIM]

actually, Sampledist is distance value like below code

x = 400; 

for i=0:Depth-1 % from 0 to m-1
    for j=1:Alines % from 1 to n
        for l=1:Alines
            t=round(sqrt((7.5*(i+x))^2+(100*(j-round(half)))^2+(100*(l-round(half)))^2),-2);
            Sampledist(i+1,j,l) = t;
           
        end
    end
end

And fftimage is measured data using transducer. So both can be regarded rand() to easier now.

[kisik KIM]

Above code calculate each array but below code calculate each elementals.

for i=1:DataLength
    for j=round(half):Num_Alines-round(half)
        for k=round(half):Num_Alines-round(half)
            for a=1:Alines
                
            y = find( Sampledist(:,round(half),a) == Sampledist(i,round(half),a));
            
            for l=1:numel(y)
                array(l,1)=fix(y(l)/m)+1;
                array(l,2)=rem(y(l),m);
                if  j - array(l,1) + round(half) > 0
                    ReconImage(i,j,k-round(half)+a) = ReconImage(i,j,k-round(half)+a) + fftimage(i+(array(l,2)-1),j+(array(l,1)-round(half)),a)*D313(array(l,2),array(l,1),a)/10000;
                else

                end
            end
            end
        end
    end
end

Surprisingly, This code is faster. I guess the problem is array size.(As you can see, full size of "S" array are don't needed because some of part have data(others are zero).

So, I'm not sure which one is better.

[Jan]

Sorry, I tried to guess the sized of the inputs, but the code still stops with errors.

Please do not let the readers guess, what your input arguments are. If I cannot run your code, it is very hard to improve its speed.