MATLAB codes to CUDA MATLA (MATLAB Image Processing Toolbox Functions on GPUs)

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello everyone,
I want convert my MATLAB codes to CUDA MATLAB, can someone tell how can I do that ?

this is a code below in MATLAB as a series, so to use same algorithm but in parallel how can i do that in CUDA MATLAB ?

clc;
clear all;
close all;
tic
img=imread(‘cat.bmp’);
[M N]=size(img);
data1=zeros(M,N);

for i=1:M;
for j=1:N
if img(i,j)>150
data1(i,j)=1;
end
end
end

data=data1;

[x,y] = size(data);
% expand dataset to avoid crash when searching:
data = [zeros(1,y+2);[zeros(x,1) data zeros(x,1)]];
[x,y] = size(data);

labels = zeros(size(data));
nextlabel = 1;
linked = ;

for i = 2:x % for each row
for j = 2:y-1 % for each column
if data(i,j) ~= 0 % not background
% find binary value of neighbours
neighboursearch = [data(i-1,j-1), data(i-1,j), data(i-1,j+1),data(i,j-1)];
% for 4-connectivity, replace with:
% neighboursearch = [data(i-1,j),data(i,j-1)];

        % search for neighbours with binary value 1
        [~,n,neighbours] = find(neighboursearch==1);
        
        % if no neighbour is allready labeled: assign new label
        if isempty(neighbours)
            linked{nextlabel} = nextlabel; %#ok<*AGROW>
            labels(i,j) = nextlabel;
            nextlabel = nextlabel+1;                
        
        % if neighbours is labeled: pick the lowest label and store the
        % connected labels in "linked"
        else
            neighboursearch_label = [labels(i-1,j-1), labels(i-1,j), labels(i-1,j+1),labels(i,j-1)];
            L = neighboursearch_label(n);
            labels(i,j) = min(L);
            for k = 1:length(L)
                label = L(k);
                linked{label} = unique([linked{label} L]);
            end                
        end
    end
end

end

% remove the previous expansion of the image
labels = labels(2:end,2:end-1);

%% join linked areas

% for each link, look through the other links and look for common labels.
% if common labels exist they are linked → replace both link with the
% union of the two. Repeat until there is no change in the links.

change2 = 1;
while change2 == 1
change = 0;
for i = 1:length(linked)
for j = 1:length(linked)
if i ~= j
if sum(ismember(linked{i},linked{j}))>0 && sum(ismember(linked{i},linked{j})) ~= length(linked{i})
change = 1;
linked{i} = union(linked{i},linked{j});
linked{j} = linked{i};
end
end
end
end

if change == 0
    change2 = 0;
end

end

% removing redundat links
linked = unique(cellfun(@num2str,linked,‘UniformOutput’,false));
linked = cellfun(@str2num,linked,‘UniformOutput’,false);

K = length(linked);
templabels = labels;
labels = zeros(size(labels));

% label linked labels with a single label:
for k = 1:K
for l = 1:length(linked{k})
labels(templabels == linked{k}(l)) = k;
end
end

toc

2

If you just want your code to execute on the gpu, make your data a gpu array by using gdata = gpuArray(data) and do your operations with the gpuArray gdata.

If you want to unwind your loops so they execute in parallel, you’ll need something more sophisticated.