Fast pseudorandom-number generators with modulus 2^k or 2^k - 1 using fused multiply-add

Abstract

Several illustrations of a general technique called the Algorithm and Architecture approach was presented. The programmer controlled unrolling of loops was demonstrated equivalent to customized vectorization of RISC-type code. Its use was illustrated to show that RS/6000 processors could compute the distribution (-1, 1) at the rate of 3.25 multiply-adds. A linear congruential generators, related to the multiplicative congruential generators was also specified.