dirac_to_dirac_approx_short_thread

Threaded Dirac-to-Dirac reduction.

Overview

dirac_to_dirac_approx_short_thread<T> is the multi-threaded variant of dirac_to_dirac_approx_short.

It provides the same functionality, interface, and behavior as the single-threaded implementation, but evaluates the distance metric and its gradient in parallel.

The optimization result is equivalent to the non-threaded version.

Use this implementation for:

• Large M (input components)
• Large L (reduced components)
• High-dimensional problems
• Performance-critical applications

Interface

Inherits:

• dirac_to_dirac_approx_i<T>

Template parameter:

• T ∈ {float, double}

All overloads of:

• approximate(...)
• modified_van_mises_distance_sq(...)
• modified_van_mises_distance_sq_derivative(...)

are identical to the non-threaded implementation.

Supported Input Formats

The same overload families as in dirac_to_dirac_approx_short are available:

• Raw pointer interface (T*)
• GSL vector interface (gsl_vector / gsl_vector_float)
• GSL matrix interface (gsl_matrix / gsl_matrix_float)

Memory layout and parameter semantics are unchanged.

Weights

• wY: weights of input Dirac mixture
• wX: weights of reduced mixture

If nullptr:

• Uniform weights are assumed

Example (Raw Pointer)

dirac_to_dirac_approx_short_thread<double> reducer;
 
bool ok = reducer.approximate(
    y,        // input points (M × N)
    M,
    L,
    N,
    bMax,
    x,        // initial guess / output
    wX,       // reduced weights (optional)
    wY,       // input weights (optional)
    &result,
    options
);

Example (GSL Vector)

dirac_to_dirac_approx_short_thread<double> reducer;
 
bool ok = reducer.approximate(
    yVector,
    L,
    N,
    bMax,
    xVector,
    wXVector,
    wYVector,
    &result,
    options
);

Example (GSL Matrix)

dirac_to_dirac_approx_short_thread<double> reducer;
 
bool ok = reducer.approximate(
    yMatrix,
    L,
    bMax,
    xMatrix,
    wXVector,
    wYVector,
    &result,
    options
);

Notes

• Produces the same results as dirac_to_dirac_approx_short
• Parallelizes internal distance and gradient evaluation
• Interface-compatible replacement for the non-threaded version

If threading is not required, use dirac_to_dirac_approx_short.