Fwd: [enzo/enzo-dev] Change default setting for (the inaccurately named) bitwise-identicality setting to no (pull request #47)

[Matthew Turk]

Hi all,

I wanted to make sure this got seen on the list, and discussed if need be.

-Matt


---------- Forwarded message ----------
From: Matthew Turk <pullreques...@bitbucket.org>
Date: Thu, May 10, 2012 at 3:10 PM
Subject: [enzo/enzo-dev] Change default setting for (the inaccurately
named) bitwise-identicality setting to no (pull request #47)
To: co...@enzo-project.org


A new pull request has been opened by Matthew Turk.

MatthewTurk/enzo-dev has changes to be pulled into enzo/enzo-dev.

https://bitbucket.org/enzo/enzo-dev/pull-request/47/change-default-setting-for-the

Title: Change default setting for (the inaccurately named)
bitwise-identicality setting to no

The setting "bitwise" in the Enzo config system refers to ensuring
sequential pairwise averaging in the potential solve, when computing
boundary conditions.  While this does reduce variance in the solve
from timestep to timestep, it is not (necessarily) a method of
ensuring bitwise identicality; compiler variance is in my tests the
dominant factor in ensuring identicality.  I implemented this
functionality.

In recent tests, I have found that for nested grid simulations,
turning this off sped up calculations by *more than an order of
magnitude*.

I am issuing this pull request to leave the option in, but to disable
it by default.  I believe this is safe and appropriate, and it should
pay off for anyone doing AMR calculations.

Changes to be pulled:

cf97564c6b34 by Matthew Turk: "Change default setting for bitwise to
bitwise-no."


--
This is an issue notification from bitbucket.org.
You are receiving this either because you are the participating
in a pull request, or you are following it.