Reusable interpolation state for callers evaluating many curves. More...

#include <MotionCurves.hpp>

Collaboration diagram for MayaFlux::Kinesis::CurveEvaluator:

Public Member Functions
void	configure (InterpolationMode mode, double tension)
	Re-resolve the kernel.

	CurveEvaluator (InterpolationMode mode=InterpolationMode::CATMULL_ROM, double tension=0.5)
	Construct and resolve the evaluation kernel.

void	evaluate (const Eigen::MatrixXd &control_points, Eigen::Index num_samples, Eigen::MatrixXd &out)
	Evaluate a curve into a caller-owned matrix.

void	evaluate_planar (std::span< const double > control_points, size_t dim, Eigen::Index num_samples, std::vector< double > &out)
	Evaluate a curve into a coordinate-major buffer.

InterpolationMode	mode () const
	Currently configured mode.

void	reparameterize (const Eigen::MatrixXd &points, Eigen::Index num_samples, Eigen::MatrixXd &out)
	Resample a polyline to uniform arc length into a caller-owned matrix.

void	reparameterize_planar (std::span< const double > points, size_t dim, Eigen::Index point_count, Eigen::Index num_samples, std::vector< double > &out)
	Resample a polyline to uniform arc length, coordinate-major.

double	tension () const
	Currently configured tension.

Private Member Functions
void	build_chunks (Eigen::Index num_samples, Eigen::Index num_segments, Eigen::Index active_count)

const double *	extend (std::span< const double > control_points, size_t dim, Eigen::Index &count)

void	rebuild_kernel ()

Private Attributes
std::vector< double >	m_arc

std::vector< double >	m_basis

std::vector< CurveChunk >	m_chunks

std::vector< double >	m_extended

std::vector< double >	m_folded

std::vector< double >	m_frac

std::vector< size_t >	m_lower

InterpolationMode	m_mode

Eigen::Index	m_overlap { 0 }

std::vector< double >	m_planar

std::vector< double >	m_planar_alt

Eigen::Index	m_points_per_segment { 0 }

std::vector< Eigen::Index >	m_seg_first

std::vector< Eigen::Index >	m_seg_total

bool	m_supports_multi { false }

std::vector< double >	m_tbuf

double	m_tension

bool	m_trigonometric { false }

Detailed Description

Reusable interpolation state for callers evaluating many curves.

The evaluation core operates on plain double buffers with no Eigen types and no allocation after the first call at given dimensions. Eigen appears only at the API boundary and in the basis matrix declarations.

Layouts

Control points are point-major: the coordinate index varies fastest, so point q occupies control_points[q * dim .. q * dim + dim). This matches the memory of a column-major Eigen matrix whose columns are control points, so matrix.data() can be passed directly.

Planar output is coordinate-major: out[d * num_samples + i] is coordinate d of sample i. Consecutive samples of one coordinate are contiguous, which is what lets the sample axis occupy the SIMD lanes. Callers wanting an Eigen matrix pay one transpose in the boundary overloads.

Evaluation

Mode and tension resolve once to a single basis matrix, applied to the control block once per chunk rather than per sample. Each coordinate is then a Horner evaluation over the sample parameter, four samples per vector under AVX2 and two under NEON.

Not thread safe. One instance per thread, or one per node.

Kinesis::CurveEvaluator eval(Kinesis::InterpolationMode::CATMULL_ROM, 0.5);
std::vector<double> out;
 
eval.evaluate_planar(controls, 3, 32, out);
const double* x = out.data();
const double* y = x + 32;
const double* z = y + 32;

Definition at line 144 of file MotionCurves.hpp.

The documentation for this class was generated from the following files:

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Layouts

Evaluation