Tests using Exact fail with Intel compiler versions > 11

[Ian Hinder]

The Intel compiler 12.1 and later versions seems to be slightly less accurate than Intel 11 by default, and produces roundoff-level differences which are amplified by Exact. The affected tests are:

>>> Exact.KS-tilted/1procs
>>> RotatingSymmetry180.Kerr/1procs 
>>> RotatingSymmetry180.Kerr-rotating-180/1procs 
>>> RotatingSymmetry180.Kerr-rotating-180-staggered/1procs  
>>> RotatingSymmetry180.Kerr-staggered/1procs 
>>> RotatingSymmetry90.Kerr-rotating-90/1procs
>>> RotatingSymmetry90.Kerr-rotating-90-staggered/1procs

>>> RotatingSymmetry180.Kerr/2procs 
>>> RotatingSymmetry180.Kerr-rotating-180/2procs 
>>> RotatingSymmetry180.Kerr-rotating-180-staggered/2procs  
>>> RotatingSymmetry180.Kerr-staggered/2procs   
>>> RotatingSymmetry180.KerrSchild-rotating-180/2procs  
>>> RotatingSymmetry90.Kerr-rotating-90/2procs  
>>> RotatingSymmetry90.Kerr-rotating-90-staggered/2procs    
>>> RotatingSymmetry90.KerrSchild-rotating-90/2procs

These should be replaced by EinsteinExact tests where possible. Peter is working on this.

Keyword:

[Ian Hinder]

The new EinsteinExact tests have been implemented and committed, and they all pass on Datura (Intel 11), Stampede (Intel 13) and build (gcc). OK to remove the corresponding Exact tests?

[Roland Haas]

Fine with me.