Support the new precompiled Ipopt with LBT

[amontoison]

@odow

[codecov]

Codecov Report

Patch coverage: 100.00% and no project coverage change.

Comparison is base (f66f888) 94.02% compared to head (8923b17) 94.02%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #369   +/-   ##
=======================================
  Coverage   94.02%   94.02%           
=======================================
  Files           4        4           
  Lines         921      921           
=======================================
  Hits          866      866           
  Misses         55       55           

Impacted Files	Coverage Δ
src/Ipopt.jl	100.00% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[odow]

Looks like an issue with spral:

In SpralSolverInterface::Factorization: Unhandled error. info.flag = -53.
Maybe one forgot to set environment variable OMP_CANCELLATION to TRUE.

I'd rather not need to set environment variables. have you tested spral? Perhaps we could just stick with mumps as the default?

[odow]

Is this going to fix our Windows binary issue?

[amontoison]

Looks like an issue with spral:

In SpralSolverInterface::Factorization: Unhandled error. info.flag = -53.
Maybe one forgot to set environment variable OMP_CANCELLATION to TRUE.

I'd rather not need to set environment variables. have you tested spral? Perhaps we could just stick with mumps as the default?

I thought that MUMPS was still the default linear solver:
coin-or/Ipopt#654 (comment)

The plan is to keep MUMPS as the default linear solver.
I can easily do a patch if it's needed but I opened a PR upstream:
https://github.com/coin-or/Ipopt/pull/677/files

SPRAL is working if we define the following environment variables before that we start Julia:

export OMP_CANCELLATION=TRUE
export OMP_NESTED=TRUE
export OMP_PROC_BIND=TRUE

but it's not user-friendly (ralna/spral#25 (comment)).

This new artifact is compiled with LBT and without some link flags that were not relevant so It could solve the issue with the binary on Windows.
This Ipopt_jll is also compiled with the optimization level -O3 so It should be faster than before.

[odow]

I can easily do a patch if it's needed but I opened a PR upstream:

Seems like we could also set the solver ourselves inside Ipopt.jl?

[amontoison]

Can we do that by default everytime that we create a JuMP model?

[odow]

You could do it here in CreateIpoptProblem:

Ipopt.jl/src/C_wrapper.jl

Lines 274 to 292 in f66f888

	prob = IpoptProblem(
	ipopt_problem,
	n,
	m,
	zeros(Float64, n),
	zeros(Float64, m),
	zeros(Float64, m),
	zeros(Float64, n),
	zeros(Float64, n),
	0.0,
	0,
	eval_f,
	eval_g,
	eval_grad_f,
	eval_jac_g,
	eval_h,
	nothing,
	)
	finalizer(FreeIpoptProblem, prob)

[amontoison]

Stefan released Ipopt 3.14.13 this morning. MUMPS will be the default linear solver now if SPRAL is installed.