Master include for the numerics kernel module. More...

#include "kernel/raw.hpp"
#include "kernel/policy.hpp"
#include "kernel/array.hpp"
#include "kernel/reduce.hpp"
#include "kernel/dense.hpp"
#include "kernel/subspace.hpp"

Detailed Description

Master include for the numerics kernel module.

The kernel module is the performance substrate that the rest of numerics sits on top of. It is organized in tiers and sub-namespaces:

Tier 1 – raw.hpp (namespace num::kernel::raw) Raw-pointer, always-inline inner loops with restrict and ivdep. BLAS dispatch built in where cblas equivalents exist. No OMP. Use when writing a fused custom kernel that composes multiple operations into a single memory pass.

Tier 2 – policy.hpp + array.hpp + reduce.hpp + dense.hpp

policy.hpp (num::kernel) seq_t, par_t, default_policy, kseq, kpar, kdefault

array.hpp (num::kernel::array) axpby, axpbyz – fused elementwise, dispatched map, zip_map, reduce – template, always-inline

reduce.hpp (num::kernel::reduce) l1_norm, linf_norm, sum – scalar reductions, dispatched

dense.hpp (num::kernel::dense) ger – rank-1 update, dispatched trsv_lower, trsv_upper – triangular solves (serial; BLAS via raw::)

Dispatch uses compile-time policy tags (seq_t / par_t) rather than a runtime enum. The default (kdefault / default_policy) is par_t when NUMERICS_HAS_OMP is defined, seq_t otherwise. Zero runtime overhead.

Quick reference: num::kernel::raw::axpy(y, x, alpha, n) – raw pointer num::kernel::array::axpby(a, x, b, y) – type-safe, default policy num::kernel::array::axpby(a, x, b, y, kseq) – force sequential num::kernel::reduce::l1_norm(x, kpar) – force parallel num::kernel::dense::trsv_lower(L, b, x) – triangular solve

Definition in file kernel.hpp.