How can I set the bandwidth for multivariate kernel regression using ksrmv?

I have values in y coming from a function depending on x1 and x2. I applied ksrmv to get fitted values. I want the bandwidth to be h = 0.1. However, when I tried mhat_nw = ksrmv(x1(:),x2(:),0.1, y(:)); I got fitted values that cannot be correct.

When I tried mhat_nw = ksrmv([x1(:), x2(:)],0.1,y(:)); I got the following error:

Error using ksrmv (line 58) x and y have different rows.

Thank you, Marius

The code without bandwidth:

N = 900;

grid = linspace(0,1,30);

[x2 x1] = meshgrid(grid, grid);

x12 = [x1(:) x2(:)];

true = @(x1,x2) (sin(2*pi*x1)+x2);

y = true(x1,x2)+vec2mat(normrnd(0,1,N,1),30);

% The Nadaraya-Watson smoother

mhat_nw = ksrmv([x1(:), x2(:)],y(:));

mhat_nw_fit = y;

mhat_nw_fit(:)= mhat_nw.f;