Description

Fixes the missing host allocation in the KLU examples reported in #448.

Both kluFactor.cpp and kluRefactor.cpp create the solution vector but do not allocate its host storage before calling the KLU solve. This causes the solve to report:

[ERROR] Trying to copy from external vector, but destination (host) is not allocated!
KLU solve status: 1

and kluRefactor.exe then terminates with a segmentation fault.

I also checked the remaining examples and found similar memory handling issues in four experimental KLU examples. These changes allocate host and device storage once, synchronize updated data before device use, and clean up the allocated objects.

The FGMRES example stored a pointer to a local preconditioner that was destroyed at the end of the iteration. The preconditioner now remains valid until cleanup.

Closes #448

@shakedregev

Proposed changes

• Allocate the solution vector in host memory after it is created in kluFactor.cpp.
• Allocate the solution vector in host memory after it is created in kluRefactor.cpp.
• Fix repeated and missing allocations in four experimental KLU examples.
• Synchronize updated RHS and solution data before device use.
• Fix the preconditioner lifetime in the FGMRES example.
• Update the cleanup blocks to delete vec_rhs, the solver objects, and the handlers in the correct order.

Checklist

• All tests pass (make test and make test_install per testing instructions). Code tested on
  • CPU backend
  • CUDA backend
  • HIP backend
• I have manually run the affected KLU examples and verified that the fixed paths complete without allocation errors. Code tested on:
  • CPU backend
  • CUDA backend
  • HIP backend
• Code compiles cleanly with flags -Wall -Wpedantic -Wconversion -Wextra.
• The new code follows Re::Solve style guidelines.
• N/A: The existing KLU examples reproduce the bug.
• N/A: This fix does not add a new feature or public API that needs documentation.
• The feature branch is rebased with respect to the target branch.
• N/A: This is a small follow-up fix to the explicit memory management change already listed in CHANGELOG.md.

Further comments

On the current develop branch, kluRefactor.exe later reaches a separate allocation failure in ExampleHelper::printSummary(). That issue is addressed in #447. With the pending #447 fix temporarily applied, the 20-system ACTIVSg10k run completes with exit code 0.

The experimental CUDA RF examples complete their initial KLU solves but still fail in the later RF path because of the separate factor transfer issue tracked in #451.

The experimental GLU example no longer reports allocation errors and its initial solve succeeds, but the later updated systems still produce inaccurate residuals. That issue is separate from the memory handling changes in this PR.