Fix segfault

[alexkorovko]

Summary by CodeRabbit

• Bug Fixes

  • Fixed optimization failures when reusing a minimizer after switching to a problem with fewer state batches.
  • Updated GPU state and pointer buffers to resize correctly for both growth and shrinkage, preventing stale entries.

• Tests

  • Added regression coverage for consecutive Gauss-Newton optimizations where the state batch count decreases between runs.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Enterprise

Run ID: 7ae93168-349f-4af5-8db8-74728f5fe82e

📥 Commits

Reviewing files that changed from the base of the PR and between d4dd454 and 9c750b3.

📒 Files selected for processing (2)

• cunls/minimizer/minimizer_state.cu
• tests/gauss_newton_test.cpp

🚧 Files skipped from review as they are similar to previous changes (2)

• cunls/minimizer/minimizer_state.cu
• tests/gauss_newton_test.cpp

---

📝 Walkthrough

Walkthrough

The CUDA minimizer now resizes state and pointer buffers on both growth and shrinkage of problem topology. A regression test reuses one Gauss-Newton minimizer across problems with decreasing state-batch counts.

Changes

Minimizer state batch reuse

Layer / File(s)	Summary
Exact-size buffer management cunls/minimizer/minimizer_state.cu	State, state-pointer, and problem-pointer buffers resize when their required batch counts change, including topology shrinkage; surrounding formatting changes preserve behavior.
Decreasing batch regression coverage tests/gauss_newton_test.cpp	Tests reuse a Gauss-Newton minimizer across two-state-batch and one-state-batch problems, checking consistency, synchronizing CUDA execution, and validating that the second minimization does not throw.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title is directly related to the change set, which fixes a segfault caused by buffer/state reuse issues.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dev/ak/fix_segfault

Warning

Tools execution failed with the following error:

Failed to run tools: 14 UNAVAILABLE: read ECONNRESET

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

cunls/minimizer/minimizer_state.cu (1)

70-90: 🚀 Performance & Scalability | 🔵 Trivial | 🏗️ Heavy lift

Avoid freeing batch buffers on the shrink path. states_.resize(state_batches.size()) only makes the outer vector shorter, but that still destroys the dropped DeviceVectors, and DeviceVector::~DeviceVector() calls cudaFree. The inner state_vec.resize(size) is cheap when the new size fits capacity, so the stall risk here comes from recurring topology shrink itself; if that path is common, keep the removed buffers around or move DeviceVector to an async/pool-backed allocator.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@cunls/minimizer/minimizer_state.cu` around lines 70 - 90, Update
MinimizerState::CreateStates so shrinking the topology does not destroy dropped
DeviceVectors or invoke cudaFree through their destructors. Preserve removed
batch buffers for reuse, or replace their allocation with an async/pool-backed
strategy, while retaining existing resizing behavior for active batches.

Source: Path instructions

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@cunls/minimizer/minimizer_state.cu`:
- Around line 70-90: Update MinimizerState::CreateStates so shrinking the
topology does not destroy dropped DeviceVectors or invoke cudaFree through their
destructors. Preserve removed batch buffers for reuse, or replace their
allocation with an async/pool-backed strategy, while retaining existing resizing
behavior for active batches.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Enterprise

Run ID: 62fab16a-e1f5-43f4-83bb-caa2ab233f22

📥 Commits

Reviewing files that changed from the base of the PR and between d0aa5a2 and d4dd454.

📒 Files selected for processing (2)

• cunls/minimizer/minimizer_state.cu
• tests/gauss_newton_test.cpp