Optimized Mapping

Enable automatic optimization with MappingConfig (optimize_mapping=true).

Why It Helps

POLFED is commonly memory-bound. For Hamiltonians with few distinct offdiagonal values, reducing memory traffic in mapping can bring large speedup.

Note: the keyword name is optimize_mapping in MappingConfig.

Typical Use

mapping = MappingConfig(
    optimize_mapping=true,
    parallel_strategy=MulColsParallel(),
)

vals, vecs, report = polfed(mat, x0, howmany, target;
    produce_report=true,
    mapping=mapping,
)
display_report(report)

This workflow uses polfed with display_report for diagnostics.

What does it do?

With matrix input, POLFED inspects matrix structure and builds optimized mapping kernels. The central idea is:

• Split diagonal and offdiagonal contributions,
• Compress repeated offdiagonal values,
• Avoid unnecessary memory loads/stores in repeated mapping calls,
• Reuse optimized kernels in spectral transformation and Clenshaw steps.

Since mapping is applied many times during polynomial filtering, even moderate per-call memory savings accumulate into meaningful end-to-end speedup.

The next section covers additional memory-access reductions with custom and rescaled mappings. See Reducing Memory Access.

This page was generated using Literate.jl.