Re: [Help-glpk] Patches to improve pseudocost initialisatiion speed

[Chris Matrakidis]

Andrew,

>> I have a draft patch that introduces an internal API for keeping the
>> dual simplex state between calls and adjusting it for new bounds. It
>> works fine and speeds up pseudocost initialisation for all
>> configurations of the solver (i.e. combinations of USE_AT, EXCL and
>> SHIFT). However, I'm not happy with the API design, so I don't
>> consider it ready for submission. If you think it may be helpful, I
>> will send it.
>
> Please post your patch to the list with minimal comments for archiving.

Here it is, broken down in two pieces.
The first patch defines the API and all necessary changes and uses it
in pseudocost initialisation. This is on top of my previous pseudocost
initialisation patches but not the other simplex patches (but should
apply without conflicts).
The second patch is a tentative initial effort to update the row wise
representation of N. A more worked out version, should probably have
reset_nt() in spxnt.c and choose between build_nt() and reset_nt()
depending on the number of simplex iterations since the last call.


Best Regards,

Chris Matrakidis

dual_API1.patch
Description: Binary data

dual_API2.patch
Description: Binary data