As the documentation tells, the input is expected to be a gray scale image. Then value over 1.0 are limit to 1.0 and you matrix is interpreted as [1, 1, 1; 1, 1, 1; 1, 1, 1] with zero entropy.
Converting the matrix by mat2gray divides the values by the larges element after subtracting the smalles element:
A = [2,3,4; 4,5,6; 10,12,1]
mA = min(A(:));
B = (A - mA) / (max(A(:)) - mA)
As far as I understand, this is the correct way to determine the entropy.