[Ipopt-tickets] [Ipopt] #195: non-deterministic results of IPOPT with MA57 from Matlab

[Ipopt]

#195: non-deterministic results of IPOPT with MA57 from Matlab
----------------------+------------------------
  Reporter:  huigang  |      Owner:  ipopt-team
      Type:  defect   |     Status:  new
  Priority:  highest  |  Component:  Ipopt
   Version:  3.10     |   Severity:  normal
Resolution:           |   Keywords:
----------------------+------------------------

Comment (by kelman):

 Not sure whether the gotoblas multi-threading is deterministic or not. The
 mex file included in Opti Toolbox uses Matlab's MA57 (I helped the author
 work out how to do so) so that makes sense.

 Did you use the compilation instructions from Opti Toolbox with MSVC? I
 think he always uses MKL in those instructions. I believe if you compiled
 the Netlib BLAS into a dll and linked against it that way, Matlab will
 actually override BLAS calls from inside the mex file with its own MKL
 versions. Which Fortran compiler did you use, ifort or f2c or ? I've only
 ever built an Ipopt Windows mex file using MinGW, so I don't know what
 happens with static libraries vs dll's when using MSVC.

-- 
Ticket URL: <https://projects.coin-or.org/ticket/195#comment:4>
Ipopt <http://projects.coin-or.org/Ipopt>
Interior-point optimizer for nonlinear programs.