Add golden file regression tests and code cleanup

[sbryngelson]

User description

Summary

• Add golden file regression tests for numerical reproducibility across CPU/GPU builds
• Remove orphan functions (Phase 1 cleanup: -260 lines)
• Add strain rate helpers and unify turbulence transport code

Changes

• Golden file tests: Reference data for channel (k-ω, EARSM, MLP, TBNN), mixing length, and laminar 3D cases
• Code cleanup: Remove unused functions from timing, fields, solver, poisson_solver, nn_core, and turbulence_gep
• Turbulence transport: Refactor for better code sharing and add strain rate computation helpers
• CI: Update scripts to run golden file tests

Test plan

• CI passes with new golden file tests
• Verify CPU and GPU builds produce consistent results against reference data

🤖 Generated with Claude Code

---

PR Type

Tests, Enhancement

---

Description

• Add comprehensive golden file regression tests for numerical reproducibility

  • 6 test cases (channel k-ω, EARSM, MLP, TBNN, mixing length, 3D laminar)
  • Near-bitwise tolerance (1e-10) verification against reference files
  • Regeneration script for intentional numerical changes

• Unify CPU/GPU SST k-ω transport implementation with identical arithmetic

  • Single code path using device-callable helpers (dmin, dmax, dclamp)
  • Jacobi iteration via snapshot arrays for CPU consistency
  • Pre-computed wall distance and F1 blending at each cell

• Add strain rate magnitude helper functions to gpu_utils.hpp

  • Frobenius norm and squared strain computations
  • GPU-callable inline functions for consistent turbulence models

• Remove orphan functions from multiple files (Phase 1 cleanup)

  • Unused methods: scale_input, sor_iteration, print_velocity_profile, min_interior
  • Removed duplicate GPU transport kernel (advance_turbulence_gpu)

---

Diagram Walkthrough

flowchart LR
  A["Golden File Tests<br/>6 cases"] -->|verify| B["Numerical<br/>Reproducibility<br/>tol=1e-10"]
  C["Unified SST<br/>Transport"] -->|identical| D["CPU/GPU<br/>Arithmetic"]
  E["Helper Functions<br/>dmin/dmax/strain"] -->|enable| D
  F["Cleanup<br/>Remove orphans"] -->|reduce| G["Code<br/>Complexity"]
  B -->|enable| H["Safe<br/>Refactoring"]

Loading

File Walkthrough

	Relevant files
Tests	3 files test_golden.cpp Add comprehensive golden file regression test suite            +637/-0  regenerate_golden.sh Add script to regenerate golden reference files                    +86/-0    test_transport_trace.cpp Add minimal trace test for CPU/GPU transport debugging      +109/-0
Enhancement	4 files turbulence_transport.cpp Unify CPU/GPU SST k-ω with identical arithmetic kernel      +415/-555 turbulence_transport.hpp Add snapshot fields for Jacobi iteration consistency          +3/-11    gpu_utils.hpp Add GPU-callable strain rate magnitude helper functions    +49/-0    turbulence_baseline.cpp Use strain magnitude helper for mixing length model            +3/-2
Bug fix	1 files gpu_kernels.cpp Fix MLP feature computation to match CPU reference velocity +13/-9
Miscellaneous	12 files fields.hpp Remove unused min_interior method from ScalarField              +1/-9      nn_core.hpp Remove unused scale_input method from MLP class                    +1/-4      poisson_solver.hpp Remove unused sor_iteration method from PoissonSolver        +1/-4      solver.hpp Remove unused print_velocity_profile method from RANSSolver +1/-4      timing.cpp Remove orphan timing utility functions                                      +0/-52    timing.hpp Remove orphan timing utility declarations                                +0/-8      fields.cpp Remove min_interior implementation                                              +0/-34    poisson_solver.cpp Remove sor_iteration implementation                                            +0/-36    solver.cpp Remove print_velocity_profile implementation                          +0/-24    nn_core.cpp Remove scale_input implementation                                                +0/-8      turbulence_gep.cpp Remove orphan turbulence GEP functions                                      +0/-73    turbulence_gep.hpp Remove orphan turbulence GEP declarations                                +0/-4
Configuration changes	3 files run_ci_local.sh Add golden test suite execution and run_golden_test helper +69/-1    CMakeLists.txt Add test_golden executable and 6 parameterized test cases +11/-0    dns_unsteady.cfg Reduce time step and add perturbation amplitude parameter +3/-1

---

Summary by CodeRabbit

• New Features

  • Added golden file regression test suite for enhanced validation coverage across channel flow and mixing length configurations.

• Improvements

  • Enhanced GPU-CPU consistency in turbulence transport calculations with unified kernel implementation.
  • Refined computational kernel feature normalization and strain-rate calculations.
  • Streamlined solver API interfaces by removing internal-use methods.

• Configuration Updates

  • Adjusted time-stepping and perturbation amplitude parameters in duct flow example.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR adds golden file regression testing infrastructure, removes deprecated public APIs (scaling helpers, diagnostic methods, solver utilities), introduces GPU-compatible strain-rate calculation functions, and refactors turbulence transport to unify CPU/GPU execution paths using precomputed wall distances and device-view parameters.

Changes

Cohort / File(s)	Summary
Test Infrastructure & Build CMakeLists.txt, scripts/run_ci_local.sh	Adds test_golden executable with six golden test cases (Channel K-Omega, EARSM, MLP, TBNN, Mixing Length, Laminar 3D) and introduces run_golden_test() function for CI execution with timeout handling, skip detection, and result reporting.
Configuration examples/08_duct_flow/dns_unsteady.cfg	Reduces time step from 0.005 to 0.001 and adds new perturbation_amplitude parameter (0.1).
Public API Removals – Headers include/fields.hpp, include/nn_core.hpp, include/poisson_solver.hpp, include/solver.hpp, include/timing.hpp, include/turbulence_gep.hpp	Removes min_interior() from ScalarField; removes fill() overloads from TensorField; removes scale_input() from MLP; removes sor_iteration() from PoissonSolver; removes print_velocity_profile() from RANSSolver; removes print_gpu_utilization_summary() and assert_gpu_dominant() from TimingStats; removes ensure_initialized() and compute_gep_factor() declarations from TurbulenceGEP.
Public API Removals – Implementation src/fields.cpp, src/nn_core.cpp, src/poisson_solver.cpp, src/solver.cpp, src/timing.cpp, src/turbulence_gep.cpp	Implements removal of corresponding methods; inlines GEP factor computation into update path; inlines input scaling into forward paths.
GPU Utilities include/gpu_utils.hpp	Adds five GPU-compatible inline functions: strain_magnitude_frobenius(), strain_magnitude_frobenius_3d(), strain_magnitude_simple(), strain_squared(), and strain_squared_3d() with OpenMP offload guards and documentation.
Turbulence Transport Refactoring include/turbulence_transport.hpp	Adds k_old_, omega_old_, and wall_dist_ as private state members for Jacobi iteration; removes advance_turbulence_gpu declaration; adjusts apply_wall_bc_omega formatting.
Turbulence Baseline & Kernels src/turbulence_baseline.cpp, src/gpu_kernels.cpp	Includes gpu_utils.hpp; replaces direct strain-rate calculation with call to strain_magnitude_frobenius(); changes strain-rotation ratio to conditional form; uses clamped local velocity magnitude for feature normalization.
Turbulence Transport Implementation src/turbulence_transport.cpp	Major refactoring: unifies CPU/GPU paths into single shared loop with device-agnostic helpers (dmin, dmax, dclamp); precomputes wall distances; introduces compute_F1_local helper; updates advance_turbulence signature to accept TurbulenceDeviceView* parameter; eliminates separate GPU kernel code in favor of inlined, unified logic.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• Add proper config files for 3D examples and standardize run.sh scripts #16: Adds identical test_golden executable and six golden test cases to CMakeLists.
• 3D #15: Modifies SSTKOmegaTransport::advance_turbulence signature to accept TurbulenceDeviceView and introduces k_old_, omega_old_, wall_dist_ state members.

Suggested labels

Review effort 4/5, testing, refactoring, GPU, turbulence-modeling

Poem

🐰 Golden tests now guide the way,
CPU and GPU learn to play,
APIs cleaned, no clutter here,
Strain-rates unified, crystal clear—
A rabbit's refactor, tight and bright! 🌟

Pre-merge checks and finishing touches

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 60.87% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'Add golden file regression tests and code cleanup' accurately reflects the main changes in the PR: it adds a golden file regression test suite and removes orphaned/unused functions across multiple modules. The title is concise, clear, and captures the primary objectives.

✨ Finishing touches

• 📝 Generate docstrings

---

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

[qodo-code-review]

PR Compliance Guide 🔍

Below is a summary of compliance checks for this PR:

Security Compliance
⚪	Path traversal Description: The case_name CLI argument is directly concatenated into output paths (e.g., prefix = ref_dir + case_name + "_") and, when --regenerate is used, can be exploited for path traversal (e.g., ../../...) to overwrite arbitrary files accessible to the user running the test. test_golden.cpp [443-574] Referred Code int run_test(const std::string& case_name, bool regenerate) { std::string golden_dir = get_golden_dir(); TestCase tc = get_test_case(case_name); std::cout << "\n"; std::cout << "================================================================\n"; std::cout << " GOLDEN FILE TEST: " << case_name << "\n"; std::cout << "================================================================\n"; // Check for NN weights if needed std::string nn_path; if (!tc.nn_model.empty()) { nn_path = find_nn_weights(tc.nn_model); if (nn_path.empty()) { std::cout << "[SKIP] NN weights not found for " << tc.nn_model << "\n"; return 0; // Skip is not a failure } std::cout << "Using NN weights: " << nn_path << "\n"; } // Setup mesh ... (clipped 111 lines)
Ticket Compliance
⚪	🎫 No ticket provided Create ticket/issue
Codebase Duplication Compliance
⚪	Codebase context is not defined Follow the guide to enable codebase context checks.
Custom Compliance
🟢	Generic: Comprehensive Audit Trails Objective: To create a detailed and reliable record of critical system actions for security analysis and compliance. Status: Passed Learn more about managing compliance generic rules or creating your own custom rules
	Generic: Meaningful Naming and Self-Documenting Code Objective: Ensure all identifiers clearly express their purpose and intent, making code self-documenting Status: Passed Learn more about managing compliance generic rules or creating your own custom rules
	Generic: Secure Logging Practices Objective: To ensure logs are useful for debugging and auditing without exposing sensitive information like PII, PHI, or cardholder data. Status: Passed Learn more about managing compliance generic rules or creating your own custom rules
🔴	Generic: Robust Error Handling and Edge Case Management Objective: Ensure comprehensive error handling that provides meaningful context and graceful degradation Status: Missing null checks: New GPU-offload code dereferences device_view without validating it is non-null and valid, risking crashes when GPU offload is enabled but device_view is absent/invalid. Referred Code #ifdef USE_GPU_OFFLOAD // GPU: use device pointers (parallel loop = Jacobi automatically) size_t u_total_size = (size_t)mesh.total_Ny() * (mesh.total_Nx() + 1); size_t v_total_size = (size_t)(mesh.total_Ny() + 1) * mesh.total_Nx(); u_ptr = device_view->u_face; v_ptr = device_view->v_face; k_read_ptr = device_view->k; omega_read_ptr = device_view->omega; k_write_ptr = device_view->k; omega_write_ptr = device_view->omega; nu_t_ptr = device_view->nu_t; wall_dist_ptr = device_view->wall_distance; u_stride_local = device_view->u_stride; v_stride_local = device_view->v_stride; #else (void)device_view; // CPU: Jacobi snapshot - copy k, omega to snapshot arrays first for (int jj = 0; jj < mesh.total_Ny(); ++jj) { for (int ii = 0; ii < mesh.total_Nx(); ++ii) { k_old_(ii, jj) = k(ii, jj); omega_old_(ii, jj) = omega(ii, jj); ... (clipped 243 lines) Learn more about managing compliance generic rules or creating your own custom rules
⚪	Generic: Secure Error Handling Objective: To prevent the leakage of sensitive system information through error messages while providing sufficient detail for internal debugging. Status: Verbose failure output: Test failures print detailed file paths and computed field values which may be acceptable for test binaries but should be verified as non-user-facing and non-sensitive for your environment. Referred Code if (!f) { std::cerr << " [FAIL] Missing golden file: " << golden_path << "\n"; return false; } double max_diff = 0.0; int max_i = 0, max_j = 0, max_k = 0; double computed_val = 0.0, golden_val = 0.0; if (mesh.is2D()) { for (int j = mesh.j_begin(); j < mesh.j_end(); ++j) { for (int i = mesh.i_begin(); i < mesh.i_end(); ++i) { double gval; if (!(f >> gval)) { std::cerr << " [FAIL] " << name << ": Unexpected EOF at (" << i << "," << j << ")\n"; return false; } double diff = std::abs(field(i, j) - gval); if (diff > max_diff) { max_diff = diff; ... (clipped 44 lines) Learn more about managing compliance generic rules or creating your own custom rules
	Generic: Security-First Input Validation and Data Handling Objective: Ensure all data inputs are validated, sanitized, and handled securely to prevent vulnerabilities Status: CLI input not constrained: The case_name argument is used to construct output/reference file paths without explicit allowlisting at the point of use, so confirm it cannot be abused outside the intended test harness (even though get_test_case() currently throws for unknown cases). Referred Code std::string case_name = argv[1]; bool regenerate = false; Learn more about managing compliance generic rules or creating your own custom rules

Compliance status legend

🟢 - Fully Compliant
🟡 - Partial Compliant
🔴 - Not Compliant
⚪ - Requires Further Human Verification
🏷️ - Compliance label

[qodo-code-review]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
High-level	Refactor monolithic transport kernel for modularity The unified advance_turbulence function is a monolithic kernel with redundant computations. It should be refactored into smaller, modular kernels for better readability, maintainability, and performance. Examples: src/turbulence_transport.cpp [577-849] void SSTKOmegaTransport::advance_turbulence( const Mesh& mesh, const VectorField& velocity, double dt, ScalarField& k, ScalarField& omega, const ScalarField& nu_t_prev, const TurbulenceDeviceView* device_view) { TIMED_SCOPE("sst_transport"); ... (clipped 263 lines) Solution Walkthrough: Before: void advance_turbulence(...) { // Monolithic kernel for CPU/GPU #pragma omp target ... for each cell (i, j) { // ... // Compute F1 for the current cell double F1_c = compute_F1_local(k(i,j), ...); // Re-compute F1 for all neighbors inside the loop, causing redundant work double F1_ip = compute_F1_local(k(i+1,j), ...); double F1_im = compute_F1_local(k(i-1,j), ...); // ... and so on for other neighbors // Compute face diffusivities using the just-computed F1 values double nu_k_e = 0.5 * (nu_k_from(F1_c) + nu_k_from(F1_ip)); // ... // Compute RHS and update k, omega for cell (i,j) } } After: void advance_turbulence(...) { // Kernel 1: Compute F1 for all cells and store in a field compute_blending_functions(k, omega, F1_field); // Kernel 2: Compute effective diffusivities using the F1 field compute_effective_diffusivities(F1_field, nu_t, nu_k_field, nu_omega_field); // Kernel 3: Main update loop, now much simpler #pragma omp target ... for each cell (i, j) { // Read pre-computed face diffusivities double nu_k_e = 0.5 * (nu_k_field(i,j) + nu_k_field(i+1,j)); // ... // Compute RHS and update k, omega } } Suggestion importance[1-10]: 9 __ Why: The suggestion correctly identifies a major design flaw in the new unified advance_turbulence kernel, which is monolithic, complex, and performs significant redundant computations, impacting maintainability and performance.	High
Possible issue	Prevent potential division by zero Restore the small epsilon regularization term in the denominator when calculating the local Reynolds number feature to prevent potential division-by-zero errors. src/gpu_kernels.cpp [164] -features[feat_base + 4] = S_mag * delta * delta / nu; +features[feat_base + 4] = S_mag * delta * delta / (nu + 1e-10); Apply / Chat Suggestion importance[1-10]: 7 __ Why: The suggestion correctly points out a potential division-by-zero if nu is 0, which is a valid concern for numerical stability, although nu is unlikely to be zero in practice.	Medium
More

[qodo-code-review]

CI Feedback 🧐

A test triggered by this PR failed. Here is an AI-generated analysis of the failure:

Action: build-and-test (ubuntu-latest, Debug)

Failed stage: Build and Test (CPU-only) [❌]

Failed test name: Golden: Channel EARSM

Failure summary: The action failed because the golden-file regression test Golden: Channel EARSM exited with code 1 after its output no longer matched the stored golden results within the required tolerance (tol=1e-10). The channel_earsm comparison reported multiple fields exceeding tolerance: - u: max_diff = 7.624e-04 at (16,1) (computed 5.783e-03 vs golden 6.545e-03) - v: max_diff = 1.953e-05 at (32,2) (computed -9.607e-06 vs golden 9.922e-06) - p: max_diff = 4.463e-03 at (32,32) (computed -5.428e-03 vs golden -9.649e-04) - k: max_diff = 5.137e-06 at (32,32) (computed 2.083e-03 vs golden 2.078e-03) - omega: max_diff = 1.281e+00 at (2,31) (computed 6.587e+00 vs golden 5.306e+00) All other tests passed, but this single golden mismatch caused the overall workflow to fail (Failed: 1), producing Process completed with exit code 1.

Relevant error logs: 1: ##[group]Runner Image Provisioner 2: Hosted Compute Agent ... 453: �[0;32m[PASS]�[0m Code sharing paradigm 454: �[0;34m========================================�[0m 455: �[0;34mFast Tests (~1-2 minutes)�[0m 456: �[0;34m========================================�[0m 457: �[1;33m[INFO]�[0m Running 3D Quick Validation... 458: �[0;32m[PASS]�[0m 3D Quick Validation 459: === Fast 3D Validation Tests === 460: Test 1: Divergence-free after projection... PASSED (div=7.265878e-10) 461: Test 2: Z-invariant flow preservation... PASSED (max z-variation=1.583471e-06) 462: Test 3: Degenerate 3D (Nz=1) matches 2D... PASSED (max diff=0.000000e+00) 463: === Results: 3/3 tests passed === 464: [SUCCESS] All quick 3D validation tests passed! 465: �[1;33m[INFO]�[0m Running 3D Gradients... 466: �[0;32m[PASS]�[0m 3D Gradients 467: === 3D Gradient Tests === 468: Test 1: Linear u=z field (du/dz should be 1)... PASSED (max error = 0.000000e+00) 469: Test 2: Sinusoidal w=sin(x) field (dw/dx = cos(x))... PASSED (max error = 6.382268e-03) 470: Test 3: All nine gradient components (polynomial field)... PASSED (max error = 0.000000e+00) 471: Test 4: Divergence accuracy (divergence-free field)... PASSED (max div = 4.329870e-15) 472: Test 5: Z-gradient symmetry (parabolic profile)... PASSED (max error = 0.000000e+00) 473: === Results: 5/5 tests passed === 474: [SUCCESS] All 3D gradient tests passed! 475: �[1;33m[INFO]�[0m Running 3D W-Velocity... 476: �[0;32m[PASS]�[0m 3D W-Velocity 477: === 3D W-Velocity Tests === 478: Test 1: W-velocity storage and indexing... PASSED 479: Test 2: W-velocity staggering (z-face locations)... PASSED (w has 5 z-faces for 4 cells) 480: Test 3: W contribution to divergence... PASSED (max divergence error = 0.000000e+00) 481: Test 4: Pressure gradient in z affects w... PASSED (mean w = 5.000000e-06 > 0) 482: Test 5: W-velocity boundary conditions... PASSED (max w at walls = 0.000000e+00) 483: Test 6: W-velocity cell-center interpolation... PASSED (max interpolation error = 0.000000e+00) 484: === Results: 6/6 tests passed === 485: [SUCCESS] All w-velocity tests passed! 486: �[1;33m[INFO]�[0m Running 3D BC Application... 487: �[0;32m[PASS]�[0m 3D BC Application 488: === 3D Boundary Condition Tests === 489: Test 1: No-slip walls enforced on y-boundaries... PASSED (max wall v = 0.000000e+00) 490: Test 2: Periodic z-direction consistency... PASSED (max w diff at periodic boundary = 0.000000e+00) 491: Test 3: Mass conservation (divergence-free)... PASSED (max divergence = 8.196994e-10) 492: Test 4: All boundary types can be set independently... PASSED (solver ran without errors, max vel = 9.999200e-03) 493: === Results: 4/4 tests passed === ... 536: [SUCCESS] All golden file comparisons passed! 537: �[1;33m[INFO]�[0m Running Golden: Channel EARSM... 538: �[0;31m[FAIL]�[0m Golden: Channel EARSM (exit code: 1) 539: Output (last 30 lines): 540: ================================================================ 541: GOLDEN FILE TEST: channel_earsm 542: ================================================================ 543: Running 10 time steps... 544: Simulation complete. 545: Comparing against golden files (tol=1e-10): 546: u : max_diff = 7.624e-04 [FAIL] at (16,1) computed=5.783e-03 golden=6.545e-03 547: v : max_diff = 1.953e-05 [FAIL] at (32,2) computed=-9.607e-06 golden=9.922e-06 548: p : max_diff = 4.463e-03 [FAIL] at (32,32) computed=-5.428e-03 golden=-9.649e-04 549: k : max_diff = 5.137e-06 [FAIL] at (32,32) computed=2.083e-03 golden=2.078e-03 550: omega : max_diff = 1.281e+00 [FAIL] at (2,31) computed=6.587e+00 golden=5.306e+00 551: [FAILURE] Golden file comparison failed! 552: �[1;33m[INFO]�[0m Running Golden: Mixing Length... ... 570: [SUCCESS] All golden file comparisons passed! 571: �[1;33m[INFO]�[0m Running Golden: Channel TBNN... 572: �[0;32m[PASS]�[0m Golden: Channel TBNN 573: u : max_diff = 2.776e-17 [OK] 574: v : max_diff = 6.426e-18 [OK] 575: p : max_diff = 1.278e-15 [OK] 576: [SUCCESS] All golden file comparisons passed! 577: �[0;34m========================================�[0m 578: �[0;34mMedium Tests (~2-5 minutes)�[0m 579: �[0;34m========================================�[0m 580: �[1;33m[INFO]�[0m Running 3D Poiseuille Fast... 581: �[0;32m[PASS]�[0m 3D Poiseuille Fast 582: === Fast 3D Poiseuille Tests === 583: Test 1: Fast Poiseuille convergence (init at 0.95x analytical)... PASSED 584: Iterations: 101, Residual: 7.589790e-06 585: Max error: 2.720375e-03 (5.4% of U_max=5.0e-02) 586: Test 2: Larger grid Poiseuille (48x48x8)... PASSED 587: Centerline velocity: 4.5e-02 (analytical: 5.0e-02, error: 9.4%) 588: Test 3: W-velocity stays zero for channel flow... PASSED ... 704: PASSED 705: Testing basic computation... PASSED (CPU computed correctly) 706: Testing MLP execution... PASSED (CPU execution verified) 707: Testing TurbulenceNNMLP execution... PASSED (CPU path executed) 708: Testing TurbulenceNNTBNN execution... PASSED (CPU path executed) 709: [PASS] All CPU backend tests passed! 710: �[0;34m========================================�[0m 711: �[0;34mLonger Tests (~3-5 minutes)�[0m 712: �[0;34m========================================�[0m 713: �[1;33m[INFO]�[0m Running 2D/3D Comparison... 714: �[0;32m[PASS]�[0m 2D/3D Comparison 715: === 2D vs 3D Solver Comparison Tests === 716: Test 1: Degenerate case (Nz=1 vs 2D)... 717: Running 2D solver... done (iters=501, res=4.031548e-06) 718: Running 3D solver (Nz=1)... done (iters=501, res=4.031548e-06) 719: u L2 error: 0.000000e+00 720: u Linf error: 0.000000e+00 721: v L2 error: 0.000000e+00 722: v Linf error: 0.000000e+00 723: 2D max div: 1.578489e-10 724: 3D max div: 1.578489e-10 725: PASSED 726: Test 2: Z-invariant Poiseuille (Nz=4 vs 2D)... 727: Running 2D solver... done (iters=501, res=4.0e-06) 728: Running 3D solver (Nz=4)... done (iters=501, res=3.1e-04) 729: z-plane 0: u_err=1.0e-04, v_err=6.6e-05 730: �[1;33m[INFO]�[0m Running Solver... 731: �[0;32m[PASS]�[0m Solver 732: === Solver Unit Tests === 733: Testing laminar Poiseuille flow... PASSED (error=1.70531%, iters=3001) 734: Testing solver convergence behavior... PASSED (residual=7.751314e-07, iters=3001) 735: Testing divergence-free constraint (staggered grid)... PASSED (max_div=9.447799e-12, rms_div=7.493713e-13) 736: Testing incompressibility (periodic flux balance)... PASSED (flux_var=1.163160e-11, mean=6.667459e-02) 737: Testing single timestep accuracy (discretization)... PASSED (err: 10.00% -> 10.00%, delta=6.92e-03%) 738: Testing momentum balance (Poiseuille)... residual=5.14e-07, iters=3001, L2_error=1.62%... PASSED 739: Testing energy dissipation rate... residual=5.14e-07, iters=3001, energy_error=1.70%... PASSED 740: All solver tests passed! ... 751: ======================================== 752: [SUCCESS] All BC combinations validated! 753: �[1;33m[INFO]�[0m Running Physics Validation... 754: �[0;32m[PASS]�[0m Physics Validation 755: ======================================================== 756: ======================================================== 757: ======================================== 758: Test 7: Quick Sanity Checks 759: ======================================== 760: Checking for NaN/Inf... [OK] 761: Checking realizability... [OK] 762: [PASS] All sanity checks passed 763: ======================================== 764: ======================================== 765: Results: 766: L2 profile error after 1 step: 0.0111526% 767: [PASS] Analytical profile preserved to 0.0111526% 768: ======================================== 769: ======================================== 770: �[1;33m[INFO]�[0m Running Taylor-Green... 771: �[0;32m[PASS]�[0m Taylor-Green 772: ======================================================== 773: ======================================================== 774: Step Time KE KE_theory Error (%) 775: ======================================================== ... 780: [OK] Periodic BCs working correctly 781: ======================================================== 782: �[1;33m[INFO]�[0m Running NN Integration... 783: �[0;32m[PASS]�[0m NN Integration 784: === NN Integration Tests === 785: Testing NN-MLP model validity... PASSED 786: Testing NN-TBNN model validity... PASSED 787: Testing NN-TBNN solver integration... PASSED 788: Testing repeated NN updates... PASSED 789: Testing NN with different grid sizes... PASSED 790: �[0;34m========================================�[0m 791: �[0;34mTest Summary�[0m 792: �[0;34m========================================�[0m 793: Results: 794: Passed: 33 795: Failed: 1 796: Skipped: 0 797: Total time: 7m 23s 798: �[0;31mFAILED TESTS:�[0m 799: - Golden: Channel EARSM 800: ##[error]Process completed with exit code 1. 801: Post job cleanup.

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (5)

examples/08_duct_flow/dns_unsteady.cfg (1)

69-70: Document the purpose of perturbation_amplitude parameter.

The new perturbation_amplitude = 0.1 parameter is added without explanation. Consider adding a comment describing how this parameter affects the simulation (e.g., initial condition perturbation, forcing amplitude, etc.).

scripts/run_ci_local.sh (1)

165-215: Consider adding parameter validation for case_name.

The run_golden_test() function is well-structured and handles edge cases appropriately (missing binary, skipped tests). Consider adding validation to ensure case_name is not empty to catch invocation errors early.

Optional: Add parameter validation

 run_golden_test() {
     local test_name=$1
     local case_name=$2
     local timeout_secs=${3:-60}
+
+    if [ -z "$case_name" ]; then
+        log_failure "$test_name (case_name not provided)"
+        FAILED=$((FAILED + 1))
+        return 1
+    fi
 
     local test_binary="${BUILD_DIR}/test_golden"

src/gpu_kernels.cpp (1)

97-97: Unused parameter u_ref after refactoring.

The u_ref parameter is passed to compute_mlp_scalar_features_gpu but is no longer used in the feature calculations (replaced by local u_ref_local). It's only cast to (void) on line 175. Consider removing it from the function signature if it's no longer needed.

Also applies to: 175-175

src/turbulence_baseline.cpp (1)

282-284: Consider using strain_magnitude_frobenius for consistency.

The GPU path at line 284 and CPU path at line 365 still use inline strain magnitude calculations. For consistency with mixing_length_cell_kernel (line 49) and to centralize the formula, consider using strain_magnitude_frobenius(Sxx, Syy, Sxy) here as well.

🔎 Suggested change for line 284

-            const double S_mag = sqrt(2.0 * (Sxx*Sxx + Syy*Syy + 2.0*Sxy*Sxy));
+            const double S_mag = strain_magnitude_frobenius(Sxx, Syy, Sxy);

src/turbulence_transport.cpp (1)

840-848: Performance: GPU-CPU-GPU roundtrip for wall boundary conditions.

The wall boundary conditions require syncing k and omega from device to host (line 842), applying BCs on the host (lines 844-845), then syncing back to device (line 847). This adds two full field transfers per time step.

Consider implementing GPU-resident wall BC kernels in a future optimization pass to eliminate this roundtrip.

📜 Review details

Configuration used: defaults

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 4aa91fb and 2780fc2.

⛔ Files ignored due to path filters (23)

• tests/golden/reference/channel_earsm_k.dat is excluded by !**/*.dat
• tests/golden/reference/channel_earsm_omega.dat is excluded by !**/*.dat
• tests/golden/reference/channel_earsm_p.dat is excluded by !**/*.dat
• tests/golden/reference/channel_earsm_u.dat is excluded by !**/*.dat
• tests/golden/reference/channel_earsm_v.dat is excluded by !**/*.dat
• tests/golden/reference/channel_komega_k.dat is excluded by !**/*.dat
• tests/golden/reference/channel_komega_omega.dat is excluded by !**/*.dat
• tests/golden/reference/channel_komega_p.dat is excluded by !**/*.dat
• tests/golden/reference/channel_komega_u.dat is excluded by !**/*.dat
• tests/golden/reference/channel_komega_v.dat is excluded by !**/*.dat
• tests/golden/reference/channel_mlp_p.dat is excluded by !**/*.dat
• tests/golden/reference/channel_mlp_u.dat is excluded by !**/*.dat
• tests/golden/reference/channel_mlp_v.dat is excluded by !**/*.dat
• tests/golden/reference/channel_tbnn_p.dat is excluded by !**/*.dat
• tests/golden/reference/channel_tbnn_u.dat is excluded by !**/*.dat
• tests/golden/reference/channel_tbnn_v.dat is excluded by !**/*.dat
• tests/golden/reference/laminar_3d_p.dat is excluded by !**/*.dat
• tests/golden/reference/laminar_3d_u.dat is excluded by !**/*.dat
• tests/golden/reference/laminar_3d_v.dat is excluded by !**/*.dat
• tests/golden/reference/laminar_3d_w.dat is excluded by !**/*.dat
• tests/golden/reference/mixing_length_p.dat is excluded by !**/*.dat
• tests/golden/reference/mixing_length_u.dat is excluded by !**/*.dat
• tests/golden/reference/mixing_length_v.dat is excluded by !**/*.dat

📒 Files selected for processing (23)

• CMakeLists.txt
• examples/08_duct_flow/dns_unsteady.cfg
• include/fields.hpp
• include/gpu_utils.hpp
• include/nn_core.hpp
• include/poisson_solver.hpp
• include/solver.hpp
• include/timing.hpp
• include/turbulence_gep.hpp
• include/turbulence_transport.hpp
• scripts/run_ci_local.sh
• src/fields.cpp
• src/gpu_kernels.cpp
• src/nn_core.cpp
• src/poisson_solver.cpp
• src/solver.cpp
• src/timing.cpp
• src/turbulence_baseline.cpp
• src/turbulence_gep.cpp
• src/turbulence_transport.cpp
• tests/golden/regenerate_golden.sh
• tests/golden/test_golden.cpp
• tests/test_transport_trace.cpp

💤 Files with no reviewable changes (8)

• src/turbulence_gep.cpp
• src/nn_core.cpp
• src/solver.cpp
• include/timing.hpp
• src/timing.cpp
• src/poisson_solver.cpp
• include/turbulence_gep.hpp
• src/fields.cpp

🧰 Additional context used

📓 Path-based instructions (4)

**/*.{cpp,cc,cxx}

📄 CodeRabbit inference engine (.cursorrules)

**/*.{cpp,cc,cxx}: Do not add platform-specific tolerances with #ifdef __APPLE__ or similar platform checks in tests
Include code comments explaining WHY, not WHAT; document numerical algorithms; note non-obvious optimizations

Files:

• src/gpu_kernels.cpp
• src/turbulence_transport.cpp
• src/turbulence_baseline.cpp

**/*.{cpp,cc,cxx,h,hpp}

📄 CodeRabbit inference engine (.cursorrules)

**/*.{cpp,cc,cxx,h,hpp}: Avoid overly strict floating-point comparisons (e.g., == for doubles); use appropriate tolerances based on algorithm
Fix all compiler warnings before pushing; use [[maybe_unused]] for intentionally unused variables in assertions
Use const for read-only references and mark methods const if they don't modify state
Use RAII for resource management; use smart pointers or RAII wrappers instead of manual new/delete or malloc/free
Use OmpDeviceBuffer wrapper for GPU buffer management instead of manual allocation
Check return values and use exceptions for error conditions with informative error messages

Files:

• src/gpu_kernels.cpp
• include/poisson_solver.hpp
• include/solver.hpp
• include/gpu_utils.hpp
• include/nn_core.hpp
• src/turbulence_transport.cpp
• include/fields.hpp
• include/turbulence_transport.hpp
• src/turbulence_baseline.cpp

include/**/*.{h,hpp}

📄 CodeRabbit inference engine (.cursorrules)

Every public function needs documentation explaining parameters, return values, and side effects; include usage examples for complex functions

Files:

• include/poisson_solver.hpp
• include/solver.hpp
• include/gpu_utils.hpp
• include/nn_core.hpp
• include/fields.hpp
• include/turbulence_transport.hpp

**/*.{sh,slurm,batch}

📄 CodeRabbit inference engine (.cursor/rules/lean-changes.mdc)

Always submit slurm jobs using the embers QOS

Files:

• scripts/run_ci_local.sh

🧠 Learnings (9)

📓 Common learnings

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Applies to tests/test_physics_validation.cpp : Navier-Stokes solver validation tests must verify: viscous terms, pressure gradient, parabolic velocity profiles, divergence-free constraint, momentum balance, channel symmetry, cross-model consistency, and numerical stability

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: For GPU-related changes, also run `./test_before_ci_gpu.sh` to run the complete GPU CI test suite locally

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Applies to tests/test_physics_validation.cpp : Navier-Stokes solver validation tests must verify: viscous terms, pressure gradient, parabolic velocity profiles, divergence-free constraint, momentum balance, channel symmetry, cross-model consistency, and numerical stability

Applied to files:

• CMakeLists.txt
• src/gpu_kernels.cpp
• include/solver.hpp
• include/gpu_utils.hpp
• scripts/run_ci_local.sh
• src/turbulence_transport.cpp
• include/turbulence_transport.hpp
• src/turbulence_baseline.cpp

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Applies to **/*.{cpp,cc,cxx,h,hpp} : Use `OmpDeviceBuffer` wrapper for GPU buffer management instead of manual allocation

Applied to files:

• include/gpu_utils.hpp
• src/turbulence_transport.cpp

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Explicitly upload data to GPU; check `USE_GPU_OFFLOAD` is defined; verify `omp_get_num_devices() > 0` at runtime for GPU offload code

Applied to files:

• include/gpu_utils.hpp
• src/turbulence_transport.cpp
• src/turbulence_baseline.cpp

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Applies to tests/test_tg_validation.cpp : Taylor-Green Vortex test must validate energy decay with <5% error over 100 timesteps using viscous terms, time integration, and periodic BCs

Applied to files:

• examples/08_duct_flow/dns_unsteady.cfg
• scripts/run_ci_local.sh
• src/turbulence_transport.cpp
• include/turbulence_transport.hpp
• src/turbulence_baseline.cpp

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: For GPU-related changes, also run `./test_before_ci_gpu.sh` to run the complete GPU CI test suite locally

Applied to files:

• scripts/run_ci_local.sh

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Run `./test_before_ci.sh` before every push to repository to test Debug build, Release build, and all unit tests

Applied to files:

• scripts/run_ci_local.sh

📚 Learning: 2025-12-25T20:38:56.549Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursor/rules/lean-debug.mdc:0-0
Timestamp: 2025-12-25T20:38:56.549Z
Learning: For GitHub CI debugging, run `./tools/ci_packet.sh <runid>` via terminal tool.

Applied to files:

• scripts/run_ci_local.sh

📚 Learning: 2025-12-25T20:38:43.379Z

Learnt from: CR
Repo: comp-physics/cfd-nn PR: 0
File: .cursorrules:0-0
Timestamp: 2025-12-25T20:38:43.379Z
Learning: Create feature branches from latest main; run `./test_before_ci.sh` to verify starting state before development

Applied to files:

• scripts/run_ci_local.sh

🧬 Code graph analysis (4)

src/gpu_kernels.cpp (7)

include/turbulence_transport.hpp (4)

• delta (124-124)
• delta (124-124)
• nu (120-120)
• nu (120-120)

include/turbulence_gep.hpp (4)

• delta (49-49)
• delta (49-49)
• nu (48-48)
• nu (48-48)

include/turbulence_nn_mlp.hpp (2)

• delta (50-50)
• delta (50-50)

include/turbulence_nn_tbnn.hpp (2)

• delta (52-52)
• delta (52-52)

include/turbulence_model.hpp (4)

• delta (193-193)
• delta (193-193)
• nu (189-189)
• nu (189-189)

include/turbulence_baseline.hpp (4)

• delta (40-40)
• delta (40-40)
• delta (85-85)
• delta (85-85)

include/features.hpp (1)

• nu (129-129)

src/turbulence_transport.cpp (2)

include/gpu_kernels.hpp (1)

• compute_gradients_from_mac_gpu (30-46)

src/features.cpp (2)

• compute_gradients_from_mac (26-59)
• compute_gradients_from_mac (26-30)

include/turbulence_transport.hpp (3)

include/turbulence_nn_mlp.hpp (3)

• mesh (35-43)
• mesh (63-63)
• mesh (96-96)

include/turbulence_nn_tbnn.hpp (6)

• mesh (37-45)
• mesh (63-63)
• mesh (106-106)
• mesh (108-108)
• mesh (113-120)
• mesh (123-123)

include/turbulence_earsm.hpp (7)

• mesh (115-115)
• mesh (119-126)
• mesh (158-158)
• mesh (170-170)
• mesh (320-320)
• mesh (325-325)
• mesh (329-337)

src/turbulence_baseline.cpp (1)

include/gpu_utils.hpp (2)

• strain_magnitude_frobenius (28-30)
• strain_magnitude_frobenius (28-28)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)

• GitHub Check: gpu-tests
• GitHub Check: build-and-test (ubuntu-latest, Debug)

🔇 Additional comments (18)

include/solver.hpp (1)

144-144: This function is unused and safe to remove.

print_velocity_profile has no references in the codebase, confirming it is a debug helper that can be safely removed as part of the cleanup.

examples/08_duct_flow/dns_unsteady.cfg (1)

45-45: Verify timestep reduction impact on simulation runtime.

The timestep dt was reduced from 0.005 to 0.001 (5× smaller). With max_iter = 30000, this yields T_final ≈ 30, which is inconsistent with the comment on line 48 stating "T_final ~ 150, about 12 flow-throughs."

Either update max_iter to maintain the intended simulation duration or verify that the shorter duration is acceptable for this DNS case.

include/fields.hpp (3)

33-34: LGTM - Comment correctly reflects API change.

The comment has been appropriately updated to reflect that only max_interior() remains after min_interior() was removed.

---

33-34: min_interior() removal is safe and complete.

No remaining callers of ScalarField::min_interior() exist in the repository. The method removal has no dangling references.

---

158-228: TensorField::fill() removal verified — no remaining callers found.

Confirmation that the component-wise fill() overloads for TensorField have been successfully removed with no remaining callers in the codebase. All .fill() calls target ScalarField or VectorField objects only.

CMakeLists.txt (1)

330-339: LGTM - Golden test infrastructure well-integrated.

The golden file regression test suite is properly configured with individual test cases for each scenario. The test binary and test case registration follow established patterns in the CMakeLists.txt file.

scripts/run_ci_local.sh (2)

8-10: LGTM - Usage documentation updated correctly.

The usage instructions properly document the new golden test suite option.

---

376-388: LGTM - Golden tests properly integrated into CI flow.

The golden file regression tests are appropriately included in the all, golden, and full test suites. The test cases match those registered in CMakeLists.txt, and timeout values are reasonable for regression tests.

include/nn_core.hpp (1)

91-91: Removal of scale_input() is safe; no remaining callers found.

No references to scale_input() exist elsewhere in the codebase. The public API removal is complete.

include/poisson_solver.hpp (1)

70-70: No remaining callers of sor_iteration() found. The method has been successfully removed from the public API, and only sor_rb_iteration() remains available. The codebase contains no references to the removed method.

src/gpu_kernels.cpp (1)

147-165: Verify feature 5 behavior is intentional.

With the change to u_ref_local = max(u_mag, 1e-10), feature 5 (u_mag / u_ref_local) will essentially always be 1.0 for any non-stagnant flow. This reduces feature informativeness. The comment states this matches CPU behavior, so verify this is the intended normalization approach.

src/turbulence_baseline.cpp (1)

48-49: Good refactor to use centralized strain magnitude helper.

Using strain_magnitude_frobenius from gpu_utils.hpp improves code consistency across the codebase.

include/turbulence_transport.hpp (1)

267-268: New state fields for CPU/GPU consistency.

The addition of k_old_, omega_old_, and wall_dist_ supports the unified Jacobi iteration approach. These enable identical arithmetic between CPU and GPU paths by using snapshot arrays rather than in-place updates during the parallel loop.

include/gpu_utils.hpp (1)

21-61: Well-designed GPU-compatible strain rate helpers.

Good addition of centralized, documented strain rate functions. The separation between Frobenius norm, simple magnitude, and squared variants is useful, and the 3D versions provide good coverage. Proper use of #pragma omp declare target ensures GPU compatibility.

src/turbulence_transport.cpp (4)

26-50: Good addition of device-callable helper functions.

The dmin, dmax, dclamp, and dmax0 helpers provide GPU-compatible alternatives to std::min/std::max since standard library functions may not be available in GPU target regions. These are properly placed within the #pragma omp declare target block.

---

239-244: Pre-computed wall distance for unified CPU/GPU kernel.

Good approach to pre-compute wall distances once during initialization, avoiding repeated mesh queries in the hot loop.

---

545-571: Well-designed device-callable F1 blending function.

The compute_F1_local helper is a pure function that computes the SST blending function using only local cell data. This design ensures identical arithmetic between CPU and GPU paths.

---

938-975: Unified closure kernel is correct.

The update() function's closure computation is embarrassingly parallel since each cell's nu_t depends only on local values. The unified CPU/GPU kernel correctly handles this case.

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

Critical: GPU path reads and writes to same buffer, causing race condition.

The GPU path sets k_read_ptr = device_view->k and k_write_ptr = device_view->k (same for omega). In a parallel loop, this means cells may read updated values from neighboring cells depending on execution order, leading to non-deterministic results.

The CPU path correctly uses snapshot arrays (k_old_, omega_old_) for Jacobi iteration, but the GPU path does not. Either:

1. Allocate separate read/write buffers for GPU, or
2. Use device_view's existing structure to provide separate input/output buffers

🔎 Conceptual fix

The GPU path should use separate read and write buffers, similar to the CPU path:

     // GPU: use device pointers (parallel loop = Jacobi automatically)
+    // Need separate read buffers for true Jacobi iteration
+    // Option 1: Use k_old/omega_old device buffers if available
+    // Option 2: Double-buffer scheme with device_view
     u_ptr = device_view->u_face;
     v_ptr = device_view->v_face;
-    k_read_ptr = device_view->k;
-    omega_read_ptr = device_view->omega;
+    k_read_ptr = device_view->k_prev;  // Separate snapshot buffer
+    omega_read_ptr = device_view->omega_prev;
     k_write_ptr = device_view->k;
     omega_write_ptr = device_view->omega;

Committable suggestion skipped: line range outside the PR's diff.

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

🧩 Analysis chain

🏁 Script executed:

# Search for Nghost/Ng definitions and usage
rg -n "Nghost|Ng\s*=" --type cpp -A 2 -B 2

Repository: comp-physics/cfd-nn

Length of output: 26078

---

🏁 Script executed:

# Check the specific file and context around lines 716-719
cat -n src/turbulence_transport.cpp | sed -n '700,830p'

Repository: comp-physics/cfd-nn

Length of output: 10316

---

🏁 Script executed:

# Look for mesh initialization and ghost cell configuration
rg -n "ghost" --type cpp -i | head -30

Repository: comp-physics/cfd-nn

Length of output: 2582

---

🏁 Script executed:

# Check if there's validation or enforcement of Nghost >= 2
rg -n "Nghost\s*[<>=]+" --type cpp -B 2 -A 2

Repository: comp-physics/cfd-nn

Length of output: 396

---

🏁 Script executed:

# Check test setups to see what Nghost values are used
rg -n "Nghost\s*=" tests/ --type cpp | head -20

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check mesh initialization to see if Nghost has constraints
rg -n "class Mesh|Nghost" include/mesh.hpp -A 5 -B 2

Repository: comp-physics/cfd-nn

Length of output: 1933

---

🏁 Script executed:

# Check how the transport update function is called and if there are guards
rg -n "update_transport_rk2_sst_internal" --type cpp -B 3 -A 3

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check if Nghost is validated anywhere in solver or mesh setup
rg -n "assert.*Nghost|Nghost.*assert|Nghost.*<|Nghost.*>" --type cpp

Repository: comp-physics/cfd-nn

Length of output: 488

---

🏁 Script executed:

# Check the exact loop bounds to confirm the analysis
sed -n '703,710p' src/turbulence_transport.cpp

Repository: comp-physics/cfd-nn

Length of output: 390

---

🏁 Script executed:

# Check if the function is always called with Nghost >= 2
rg -n "update_transport_rk2_sst" src/solver.cpp -B 5 -A 5

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check default mesh creation to see if Nghost is always set to 2
rg -n "Mesh\(" --type cpp -A 3 | head -40

Repository: comp-physics/cfd-nn

Length of output: 682

---

🏁 Script executed:

# Check Mesh constructor to see what default Nghost value it uses
rg -n "Mesh::" include/mesh.hpp -A 10 | head -50

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check if there's a Mesh() default constructor defined
rg -n "Mesh()" --type cpp -A 3 -B 1 | head -40

Repository: comp-physics/cfd-nn

Length of output: 2171

---

🏁 Script executed:

# Check what happens in actual solver initialization
rg -n "initialize\|setup" src/solver.cpp -A 2 | grep -i "mesh\|nghost" -A 2 -B 2 | head -30

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Find Mesh constructor definition in mesh.hpp
rg -n "Mesh(" include/mesh.hpp -B 2 -A 8 | head -60

Repository: comp-physics/cfd-nn

Length of output: 123

---

🏁 Script executed:

# Check mesh.cpp for any default Mesh constructor
rg -n "Mesh::" src/mesh.cpp | head -20

Repository: comp-physics/cfd-nn

Length of output: 701

---

🏁 Script executed:

# Check the create_uniform factory method
sed -n '211,230p' src/mesh.cpp

Repository: comp-physics/cfd-nn

Length of output: 407

---

🏁 Script executed:

# Check if there's a default Mesh() constructor
grep -n "Mesh::Mesh\|Mesh()" include/mesh.hpp

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Look at the complete Mesh class definition for constructors
sed -n '1,50p' include/mesh.hpp

Repository: comp-physics/cfd-nn

Length of output: 1837

---

🏁 Script executed:

# Check if Nghost is initialized anywhere by default in Mesh struct
grep -n "Nghost\s*=" include/mesh.hpp

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check actual usages in tests to see what Nghost values are used
rg "create_uniform\|init_uniform" tests/ --type cpp -A 1 | head -40

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Search for actual mesh creation in tests with Nghost values
rg -n "Mesh.*create_uniform\|init_uniform" tests/ --type cpp -B 2 -A 1 | head -50

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Check test files more directly
find tests -name "*.cpp" -exec grep -l "Mesh\|create_uniform\|Nghost" {} \;

Repository: comp-physics/cfd-nn

Length of output: 1089

---

🏁 Script executed:

# Look at a specific test file to see mesh setup
head -100 tests/test_mesh.cpp

Repository: comp-physics/cfd-nn

Length of output: 2784

---

🏁 Script executed:

# Verify the exact init_uniform signature and default Nghost
sed -n '31,50p' src/mesh.cpp

Repository: comp-physics/cfd-nn

Length of output: 725

---

🏁 Script executed:

# Check what actual tests use for mesh initialization
rg -n "init_uniform|create_uniform" tests/test_solver_cpu_gpu.cpp -B 2 -A 2 | head -30

Repository: comp-physics/cfd-nn

Length of output: 1056

---

🏁 Script executed:

# Check if there's a default parameter for nghost in the header
sed -n '1,150p' include/mesh.hpp | grep -n "init_uniform"

Repository: comp-physics/cfd-nn

Length of output: 154

---

🏁 Script executed:

# Check the exact function signature in mesh.hpp
rg -n "init_uniform" include/mesh.hpp -A 3

Repository: comp-physics/cfd-nn

Length of output: 460

---

🏁 Script executed:

# Verify what nghost value is passed when omitted - check if compilation even works
rg -n "mesh.init_uniform.*;" tests/test_solver_cpu_gpu.cpp -A 1 -B 1 | head -20

Repository: comp-physics/cfd-nn

Length of output: 45

---

🏁 Script executed:

# Confirm the 3D version also has default nghost = 1
sed -n '137,142p' include/mesh.hpp

Repository: comp-physics/cfd-nn

Length of output: 292

---

Verify ghost cell count is sufficient for 2-cell stencil.

The indices idx_ip2, idx_im2, idx_jp2, idx_jm2 at lines 716-719 access cells two positions away for computing gradients at neighbor cells (lines 770-806). This requires at least Ng >= 2 ghost cells. However, the default nghost parameter in init_uniform() is 1 (mesh.hpp lines 134, 140), and test code calls mesh.init_uniform() without passing the nghost argument (e.g., test_solver_cpu_gpu.cpp lines 150, 263, 383), which means the code runs with only 1 ghost cell by default. With Ng=1, accessing ii+2 or ii-2 will cause out-of-bounds access since the valid range is only [0, Nx+1] but ii+2 reaches Nx+2.