FvClosureKernels.hpp File Reference

The explicit-FV cross-section closure as a single-source POD kernel: one monotone cubic-Hermite table per section, with T ≡ dA/dh and I₁ ≡ ∫A by construction, and a Newton inverse on the same cubic. More...

#include <cmath>
#include <cstdint>
#include <type_traits>

Include dependency graph for FvClosureKernels.hpp:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	openswmm::fv::FvClosure
	One section's closure. Fixed-size POD: inline arrays, no pointers. More...
struct	openswmm::fv::kernels::ClosureEval
	A, T and I₁ at one depth, from one panel locate. More...

Namespaces
namespace	openswmm
namespace	openswmm::fv
namespace	openswmm::fv::kernels

Macros
#define	OPENSWMM_KERNEL_FN   inline

Enumerations
enum	: uint8_t {   openswmm::fv::kClosureTabulated = 0 ,   openswmm::fv::kClosurePolynomial = 1 }
	The closure's representation class. More...

Functions
OPENSWMM_KERNEL_FN void	openswmm::fv::kernels::closureLocate (const FvClosure &c, double h, int &i, double &t) noexcept
	Locate a uniform depth panel and its clamped local coordinate.
OPENSWMM_KERNEL_FN void	openswmm::fv::kernels::closureHermite (const FvClosure &c, int i, double t, double &a, double &dadh, double &int_dh) noexcept
	Evaluate one cubic area panel, its derivative and its partial integral.
OPENSWMM_KERNEL_FN ClosureEval	openswmm::fv::kernels::closureEval (const FvClosure &c, double h) noexcept
	Evaluate area, top width and cumulative hydrostatic integral together.
OPENSWMM_KERNEL_FN ClosureEval	openswmm::fv::kernels::closureEval (const FvClosure &c, double h, double &radius) noexcept
	Evaluate closure area/width/integral and friction radius at one depth.
OPENSWMM_KERNEL_FN double	openswmm::fv::kernels::closureArea (const FvClosure &c, double h) noexcept
	Evaluate the closure flow area.
OPENSWMM_KERNEL_FN double	openswmm::fv::kernels::closureWidth (const FvClosure &c, double h) noexcept
	Evaluate the closure area derivative used as top width.
OPENSWMM_KERNEL_FN double	openswmm::fv::kernels::closureI1 (const FvClosure &c, double h) noexcept
	Evaluate the closure hydrostatic first moment.
OPENSWMM_KERNEL_FN double	openswmm::fv::kernels::closureHydRad (const FvClosure &c, double h) noexcept
	Evaluate the friction hydraulic radius.
OPENSWMM_KERNEL_FN double	openswmm::fv::kernels::closureDepthOfArea (const FvClosure &c, double a) noexcept
	Invert aggregate closure area using the same forward geometry.

Variables
constexpr int	openswmm::fv::kClosurePanels = 128
	Panels of the closure table, uniform in depth on [0, y_full].

Detailed Description

The explicit-FV cross-section closure as a single-source POD kernel: one monotone cubic-Hermite table per section, with T ≡ dA/dh and I₁ ≡ ∫A by construction, and a Newton inverse on the same cubic.

Plan FV1D_CLOSURE_KERNEL_PERF_PLAN_2026-09-11 §2a/§2a‴. The table is built once at mesh build (NetworkMeshBuilder::buildClosure) by sampling the EXACT section geometry (SectionGeometry.hpp) at N+1 uniform depths on [0, y_full], with the tapered Preissmann slot folded in; above y_full the closure is the analytic slot line, as it always was. Every function here is a plain inline over scalars and a const FvClosure& — no pointers, no allocation, no exceptions, no dispatch on shape — so the same bodies compile for the CPU solver and, annotated, for the device backend.

Two classes share the struct:

• kTabulated: the Hermite table (closed shapes, table-defined shapes, transects). A per panel is a cubic in the local coordinate t ∈ [0,1]; T is its derivative; I₁ its integral plus the running sum; R is interpolated linearly.
• kPolynomial: open sections whose area is a polynomial of degree ≤ 2 (RECT_OPEN, TRAPEZOIDAL, TRIANGULAR). Closed forms in the hot loop — cheaper than any table gather — with an exact quadratic-formula inverse (plan §2a″ cost weighting).

Why a Hermite table with the exact width as node slope: the legacy A and W tables are independent tabulations, so Newton on them was unsafe and the solver paid Brent's ~6-8 evaluations and a dependent divide chain per inversion (185 ns on a circle, of which only ~28 ns were closure calls). Here the derivative IS the closure's own, monotone by the Fritsch–Carlson limiter, so a safeguarded Newton usually converges in a few steps; the bounded inverse returns its last iterate without a separate convergence status.

Author

Caleb Buahin caleb.nosp@m..bua.nosp@m.hin@g.nosp@m.mail.nosp@m..com

Copyright

Copyright (c) 2026 Caleb Buahin. All rights reserved.

License

Apache-2.0

Macro Definition Documentation

OPENSWMM_KERNEL_FN

#define OPENSWMM_KERNEL_FN   inline