Re: Seqfault with single complex matrices on Mac

[Ben Abbott]

On Apr 6, 2016, at 08:05, Ben Abbott <address@hidden> wrote:

>> On Apr 6, 2016, at 07:43, Sebastian <address@hidden> wrote:
>> 
>> 
>>>> Am 06.04.2016 um 13:36 schrieb Ben Abbott <address@hidden>:
>>>> 
>>>> On Apr 6, 2016, at 2:42 AM, Sebastian Schöps <address@hidden> wrote:
>>>> 
>>>> Dear all, 
>>>> 
>>>> there must be a change from 4.0.0 to 4.0.1 that leads to crashes on Mac OS
>>>> using clang and Apple's BLAS lib. For example single complex matrices are
>>>> affected, e.g.
>>>> 
>>>> single ([1+i 2+i 3+i]) * single ([ 4+i ; 5+i ; 6+i])
>>>> 
>>>> segfaults octave. Since the problem was not present in 4.0.0, maybe someone
>>>> has an idea? If not, I would have to start bisecting, but this will take
>>>> forever...
>>>> 
>>>> Thanks, 
>>>> Sebastian
>>> 
>>> This works for me on both the default and stable branches.
>>> 
>>> clang --version
>>> Apple LLVM version 7.0.2 (clang-700.1.81)
>>> Target: x86_64-apple-darwin14.5.0
>>> Thread model: posix
>> 
>> Thanks for helping. Do you use openblas or Apple's implementation via 
>> vecLibFort, [1]? My problems occur only in the latter case, i.e., it works 
>> perfectly with openblas.
>> 
>> I have currently no clue what's going on. The problem might be restricted to 
>> homebrew. Nonetheless, on my mac (10.11) octave 4.0.0 works while octave 
>> 4.0.1 segfaults. Since all other settings are identical some change in 
>> Octave should be causing the trouble.
>> 
>> Sebastian
>> 
>> [1] https://github.com/mcg1969/vecLibFort
> 
> Veclib had a lot of bugs. You can try a "make check" to get an impression. 
> I'm not in front of my computer, and don't remember which blas/ lapack I'm 
> using, but it's neither one of those.

I'm using Atlas 3.10

Ben