Migrate magnetic response table to Eigen3 AoS layout; consolidate nested-vector overloads as wrappers

[Copilot]

ComputeMagneticFieldResponseTable and ComputeVectorPotentialCache used std::vector<std::vector<double>> as an intermediary between Eigen and ABSCAB, causing ~1M small heap allocations per call and strided memory access in the hot loops. Additionally, the same ABSCAB call logic was duplicated across two independent sets of overloads.

Memory layout: 3×N → N×3 (AoS)

RowMatrix3Xd is changed from Matrix<double, 3, Dynamic, RowMajor> to Matrix<double, Dynamic, 3, RowMajor>. In N×3 row-major, .data() is directly the flat AoS array [x0,y0,z0, x1,y1,z1, …] that ABSCAB expects, and per-point coordinate access (i, 0/1/2) is contiguous. Fixes the TODO(jurasic) bad loop order in MakeCylindricalGrid.

New flat-pointer AoS overloads (canonical implementations)

Added MagneticField / VectorPotential overloads for all filament types and MagneticConfiguration taking (int N, const double* eval_pos, double* result). These pass the Eigen .data() pointer directly to ABSCAB with zero copies.

Eliminated all STL intermediates in the hot path

ComputeMagneticFieldResponseTable and ComputeVectorPotentialCache no longer allocate cylindrical_grid_stl or magnetic_field_stl. The field buffer is allocated once outside the circuit loop and reused via setZero(). The MagneticConfiguration copy is also hoisted out of the loop.

// Before: per-circuit-per-filament nested-STL allocation + copy round-trip
std::vector<std::vector<double>> grid_stl(N);   // N heap allocs per circuit
std::vector<std::vector<double>> field_stl(N);  // N heap allocs per circuit
MagneticField(config, grid_stl, field_stl);
magnetic_field = ToEigenMatrix(field_stl).transpose();

// After: one Eigen N×3 buffer, passed directly to ABSCAB
magnetic_field.setZero();
MagneticField(config, N, cylindrical_grid.data(), magnetic_field.data());

Nested-vector overloads consolidated as thin wrappers

The six existing MagneticField / VectorPotential overloads taking vector<vector<double>> (used by LinkingCurrent and all tests) are now thin conversion wrappers that build a flat eval-positions buffer once, call the AoS implementation, and scatter the result back. This eliminates the duplicated ABSCAB setup logic while keeping the public API intact.

Result

test_bench_response_table_from_coils: ~1.32 s vs ~1.41 s baseline (~6% improvement). The remaining cost is the ABSCAB Biot-Savart integration itself (already OpenMP-parallelised internally).

[jurasic-pf]

• Migrate magnetic response table to Eigen3 AoS layout; consolidate nested-vector overloads as wrappers #490 👈 (View in Graphite)
• main

---

How to use the Graphite Merge Queue

Add the label merge-queue to this PR to add it to the merge queue.

You must have a Graphite account in order to use the merge queue. Sign up using this link.

An organization admin has enabled the Graphite Merge Queue in this repository.

Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue.

This stack of pull requests is managed by Graphite. Learn more about stacking.

[jons-pf]

LGTM, except that I would suggest to also migrate the magnetic field computation for InfiniteStraightFilament to accept the AoS data directly for consistency.