[Ipopt] WSMP for IPOPT

[Stuart Rogers]

I have a few questions regarding WSMP and how it is used by IPOPT.

* The WSMP webpage (http://researcher.watson.ibm.
com/researcher/view_group.php?id=1426) says that version 15 includes
GPU-accelerated libraries. Which WSMP solvers (i.e. symmetric direct,
general direct, or iterative solvers) have been implemented to run on the
GPUs? Was the GPU acceleration of WSMP implemented in CUDA or OpenCL? Are
there any benchmark results showing a marked speed-up of IPOPT when using
the GPU-accelerated version of WSMP?

* The WSMP webpage seems to only offer downloads of WSMP. Is PWSMP included
in these downloads? Is IPOPT able to utilize the PWSMP linear solver? If
so, are there any benchmark results showing a marked speed-up of IPOPT when
using PWSMP rather than WSMP?

* The default value in IPOPT for the wsmp_num_threads option is 1:
https://www.coin-or.org/Ipopt/documentation/node62.html. This option
determines the number of threads to be used by WSMP. In the case when the
user doesn't set this option, why doesn't IPOPT or WSMP internally set this
option to the number of physical cores detected rather than defaulting to 1?

* wsmp_iterative is an IPOPT option that uses the iterative rather than
symmetric direct WSMP solver:
https://www.coin-or.org/Bonmin/option_pages/options_list_ipopt.html. Are
there any benchmark results showing a marked speed-up of IPOPT when using
this iterative WSMP option?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20170117/512fbcec/attachment.html>