A simple method for rejection sampling efficiency improvement on SIMT architectures

We derive a probability distribution for the possible number of iterations required SIMT (single instruction multiple thread) program using rejection sampling to finish creating sample across all threads. This is found match recently proposed from Chakraborty and Gupta (in: Communications in statistics: theory methods, 2015) that was shown as good approximation certain datasets. work demonstrates an exact application this distribution. The can be used evaluate relative merit some methods on GPU without resort numerical tests. reduces expected geometric single thread per warp limit. A simplified formula approximate obtain iteration samples provided. With new result, simple, efficient layout assigning tasks threads function recourse more complicated methods.