Rejection isn't new - what's new is using a variation of his "use the high order bits of the 64x64 multiply by a range to get a random number efficiently" trick

Rejection isn't new - what's new is using a variation of his "use the high order bits of the 64x64 multiply by a range to get a random number efficiently" trick (which he published two or three years ago) to substantially improve the efficiency of a rejection sampling method. The efficient random number in a range trick is really cool. I've used it in a couple of data structures that we've designed - it's substantially faster than using modulus, or even using pre-computed division by reciprocal multiplication.