Hypergeometric distribution with parameters l, s, and m. If M - kL > 40 and kL < kH, where M = ⌊min(s + 1,l - s + 1)⋅min(m + 1,l - m + 1)/(l + 2)⌋, kL = max(0,min(s.l - s) - (max(m,l - m)), kH = min(min(m,l - m), min(s,l - s)), the random numbers are generated by the H2PE method (see [Kach85] for details). Otherwise, they are produced using the inverse transformation method in combination with the table lookup method. The H2PE method is a variation of the acceptance/rejection method that uses constant (on the fraction close to the distribution mode) and exponential (at the distribution tails) functions as majorizing functions. To avoid time-consuming acceptance/rejection checks, a squeezing technique is applied.
See Intel® MKL Vector Statistics Random Number Generator Performance Data for test results summary and performance graphs.