Re: amber-developers: parallel random number generators

From: Volodymyr Babin <vbabin.ncsu.edu>
Date: Wed, 7 May 2008 17:14:39 -0400 (EDT)

> This is already in sander: see mt19937.f. There is even a parallel
> routine called mt19937_bcast, but it doesn't look(?) like it is really at
> parallel rng in the sense we are talking about.

No, it is not a parallel mt19937. The mt19937_bcast method simply
calls mpi_bcast (3 times instead of just 1) [it simply broadcasts
the mt19937 state; used for initialization -- master loads the state
and then it gets broadcasted to everyone else].

Have a great day,
  Volodymyr
Received on Sun May 11 2008 - 06:07:15 PDT
Custom Search