Re: README.MacOS

[Jarno Rajahalme]

On Jan 27, 2011, at 15:32 , ext Ben Abbott wrote:

> On Jan 27, 2011, at 5:11 AM, Jarno Rajahalme wrote:
> 
>>> 
>>> Jarno/Lukas, pls respond at the *bottom* of the email thread, as it makes 
>>> it much easier for those arriving late to follow along. Thanks!
>>> 
>>> Jarno, is MacResearch a part of Apple?
>>> 
>> 
>> Sorry for the laziness, takes effort to scroll to the bottom :-)
>> 
>> On MacResearch, I don't know. I have no other connection to them, other than 
>> finding relevant discussion to problems using Apple's version of atlas in 
>> 64-bit. According to http://www.macresearch.org/about_macresearch they are 
>> independent.
>> 
>> Jarno
> 
> What I was wondering is if this bug has been reported to Apple? From your 
> prior comment I'd inferred you had, but the link provided was to MacResearch 
> ... hence my confusion.
> 


The bugreport.apple.com Problem ID is 7850167, originally filed in 11-Apr-2010 
12:04 AM.

However, I don't think these are publicly visible, hence the link to the 
discussion at macresearch.

I made yesterday a wrapper for the Apple BLAS and LAPACK. Today I have been 
running tests. The complete BLAS test suite passes all tests, as does LAPACK 
linear test suite. However, there are some failures (fails to pass, but does 
not crash) in lapack eigenvalue tests. Many of these failures stay even when 
testing against lapack 3.1.1 itself, without using the wrapper for Apple lapack 
routines. Some failures were resolved by using blas from atlas.

So it seems there are some bugs left in the apple blas/lapack that are not 
related to the function calling convention at all. With the wrapper these seem 
to affect eigenvalue routines only.

  Jarno