policy.hpp File Reference

Compile-time dispatch policy tags for the kernel module. More...

#include "core/policy.hpp"

Go to the source code of this file.

Classes
struct	num::kernel::seq_t
	Sequential execution policy tag. Guarantees no OMP parallel regions; safe to call inside an existing parallel region without causing nested-parallelism overhead. More...
struct	num::kernel::par_t
	Parallel execution policy tag. Activates OMP parallel-for / reduction constructs when NUMERICS_HAS_OMP. Falls through to seq_t behaviour when OMP is not available. More...

Namespaces
namespace	num
namespace	num::kernel

Typedefs
using	num::kernel::default_policy = seq_t
	Default policy: par_t if OMP is available, seq_t otherwise. Selected at configure time; zero runtime overhead.

Variables
constexpr seq_t	num::kernel::kseq {}
constexpr par_t	num::kernel::kpar {}
constexpr default_policy	num::kernel::kdefault {}

Detailed Description

Compile-time dispatch policy tags for the kernel module.

The kernel layer has one relevant dimension of variation: parallelism. BLAS dispatch is already handled inside raw:: for ops that have cblas equivalents. Vectorization (SIMD) is handled by the compiler or by raw::. The only thing the caller needs to choose is: run sequentially or in parallel (OMP).

Two tags: seq_t – always sequential; safe to call from inside an OMP region par_t – OpenMP parallel; falls through to seq_t if OMP not available

Default policy (kdefault / default_policy): par_t when NUMERICS_HAS_OMP is defined seq_t otherwise

Usage: num::kernel::array::axpby(a, x, b, y); // default policy num::kernel::array::axpby(a, x, b, y, num::kernel::kseq); // force seq num::kernel::array::axpby(a, x, b, y, num::kernel::kpar); // force par

For templates that cascade policy through multiple kernel calls: template<typename Policy = num::kernel::default_policy>> void my_kernel(..., Policy p = {}) { num::kernel::reduce::l1_norm(x, p); num::kernel::array::axpby(a, x, b, y, p); }

Definition in file policy.hpp.