Description

The KLU L and U factors were synchronized to the device without first allocating device storage. The solution produced by the initial KLU solve was also used by CUDA operations without being synchronized to the device.

This caused invalid residuals, CUDA copy errors, and a later segmentation fault.

Closes #451

@shakedregev

Proposed changes

• Allocate device storage for the KLU L and U factors before synchronizing them.
• Synchronize the KLU solution to the device after a successful solve.

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 non-experimental examples and verified that residuals are close to machine precision. (In your build directory run: ./examples/<your_example>.exe -h to get instructions how to run examples). 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 gpuRefactor example reproduces 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

gpuRefactor still reports repeated allocation messages after the first system. That separate issue is tracked in #445.