You can use discretize (which didn't exist when this question was asked originally) to do this. Generate uniform random numbers, bin them using bins whose widths are given by P, and for each bin return the corresponding element of X.
P = [0.1 0.3 0.4 0.2];
X = [1 2 5 9];
values = discretize(rand(1, 1e4), cumsum([0 P]), X);
histogram(values, 'Normalization', 'probability')
The probabilities shown in the histogram should agree pretty closely with the values in P.