Add external ReSolve interface

[tamar-dewilde]

Description

Add an experimental EVLOSER sparse linear solver interface to HiOp that uses an installed public ORNL/ReSolve package.

Enable the interface at configure time with:

HIOP_USE_EVLOSER=ON

Runtime solver selection:

linear_solver_sparse=evloser

The interface supports CPU KLU solves and RF refactorization on CUDA and HIP. The existing embedded resolve solver implementation remains unchanged.

@pelesh

Proposed changes

• Add HIOP_USE_EVLOSER as an independent sparse solver option.

• Add the HiOp EVLOSER wrapper and external ReSolve provider.

• Discover public ReSolve with find_package(ReSolve CONFIG REQUIRED).

• Link through ReSolve::ReSolve.

• Propagate the ReSolve dependency through the installed HiOp CMake configuration.

• Use public ReSolve components for:

  • CPU KLU factorization and solve
  • CUDA RF refactorization and solve
  • HIP RF refactorization and solve
  • CSR matrix and vector storage

• Convert HiOp symmetric sparse matrices to the CSR representation expected by ReSolve.

• Support host and device matrix, right-hand-side, and solution transfers.

• Return failed factorizations and refactorizations to HiOp for regularization and retry.

• Register EVLOSER in sparse KKT solver selection and dual initialization.

• Require the inertia-free factorization acceptor for EVLOSER.

• Add driver options:

  • -evloser
  • -evloser_cuda_rf
  • -evloser_hip_rf

• Add EVLOSER RF regression tests: NlpSparseRaja2_EVLOSER_RF (CUDA) and NlpSparseRaja2_EVLOSER_HIP_RF (HIP).

• Keep the existing embedded resolve solver sources unchanged.

Additional cleanup in files touched by this branch

• Update the sparse example self-check to reject non-finite objectives and errors.
• Correct the sparse RAJA driver help text to match the existing ReSolve option names.

Checklist

• All tests pass (make test and make test_install per testing [instructions](https://github.com/ORNL/ReSolve?tab=readme-ov-file#test-and-deploy)). Code tested on

  • CPU backend
  • CUDA backend
  • HIP backend

• I have manually run the non-experimental examples and verified that residuals are close to machine precision. Code tested on:

  • CPU backend
  • CUDA backend
  • HIP backend

• Code compiles cleanly with flags -Wall -Wpedantic -Wconversion -Wextra.

• Regression tests were added and run for the new code.

• The new code is documented.

• The feature branch is rebased with respect to the target branch.

• I have updated CHANGELOG.md to reflect the changes in this PR.

Further comments

The HIP RF path runs without iterative refinement because the current iterative-refinement implementation depends on CUDA-only Krylov kernels.

SuiteSparse/KLU was supplied through the active Spack environment with spack -e. The HiOp Spack recipe was not changed.

Detailed CPU, CUDA, and Frontier validation logs will be attached in a follow-up comment.

This remains a draft PR until the external EVLOSER interface has been reviewed and is ready for integration.

[tamar-dewilde]

External ReSolve validation completed using ReSolve commit:

a93fb6571e3542b2fa779983fef0186a034c53c9

• CPU: 35/35 tests passed
• CUDA: 41/41 tests passed
• Frontier HIP: 37/37 tests passed
• Direct CPU, CUDA RF, and HIP RF runs passed the six-digit self-check
• Runtime linkage was verified against the expected external ReSolve libraries

Attached are the complete CPU, CUDA, and Frontier validation logs:

• local-cpu-external-evloser-validation.log
• local-cuda-external-evloser-validation.log
• frontier-external-evloser-hip-validation.log