Shape-grouped cross-section manager for batch computation. More...

#include <XSectBatch.hpp>

Collaboration diagram for openswmm::XSectGroups:

Public Member Functions
void	build (const SimulationContext &ctx)
	Build shape groups from SimulationContext link data.

void	attachTransectTables (const SimulationContext &ctx)
	Attach transect tables to the IRREGULAR shape group.

void	build (const XSectParams *params, int n_links)
	Build shape groups from an array of XSectParams.

void	computeAreas (const double depths, double areas, int n_links) const
	Compute area for every link, reading depth from `depths[link]`.

void	computeHydRad (const double depths, double hydrad, int n_links) const
	Compute hydraulic radius for every link.

void	computeWidths (const double depths, double widths, int n_links) const
	Compute top width for every link.

void	computeAreaAndHydRad (const double depths, double areas, double *hydrad, int n_links) const
	Fused area + hydraulic radius computation (single gather/scatter).

void	computeAreaHydRadTriple (const double d1, const double d2, const double dm, double a1, double a2, double am, double hrad1, double hrad_mid, int n_links) const
	Fused triple: d1→(a1,hrad1), d2→a2, dm→(am,hrad_mid) in one pass over shape groups. Replaces three separate compute calls in STEP B / STEP D of computeLinkGeometry.

void	computeWidthsTriple (const double d1, const double d2, const double dm, double w1, double w2, double wm, int n_links) const
	Fused triple: d1→w1, d2→w2, dm→wm in one pass over shape groups.

void	computeAreaHydRadTripleTeam (const double d1, const double d2, const double dm, double a1, double a2, double am, double hrad1, double hrad_mid, int n_links, int tid, int nthreads) const
	Team-callable triple kernels: thread `tid` of `nthreads` processes only its static slice of every shape group.

void	computeWidthsTripleTeam (const double d1, const double d2, const double dm, double w1, double w2, double wm, int n_links, int tid, int nthreads) const

void	setBypassMask (const std::uint8_t *bypassed_by_link) const
	Restrict the triple kernels to non-bypassed links for the current Picard iteration.

void	computeSectionFactors (const double areas, double sfact, int n_links) const
	Compute section factor for every link (from area, not depth).

void	computeDepthsFromArea (const double areas, double depths, int n_links) const
	Compute depth from area for every link (inverse).

int	numGroups () const
	Number of non-empty shape groups.

const ShapeGroup &	group (int i) const
	Access a specific shape group.

const ShapeGroup *	findGroup (XSectShape shape) const
	Find the group for a given shape (returns nullptr if no links have that shape).

Detailed Description

Shape-grouped cross-section manager for batch computation.

Call build() once after the model is loaded/built. This sorts links by shape type and builds contiguous SoA parameter blocks. Then use computeAreas(), computeHydRad(), computeWidths() in the routing hot loop — each iterates over shape groups and calls the shape-specific vectorised kernel.

Results are written directly into the caller's global arrays (indexed by link) using the scatter index link_idx.

Member Function Documentation

◆ attachTransectTables()

void openswmm::XSectGroups::attachTransectTables ( const SimulationContext & ctx )

Attach transect tables to the IRREGULAR shape group.

Must be called after build() when IRREGULAR shapes exist. Populates per-link area/hrad/width table pointers from the precomputed transect tables in the context.

Parameters

ctx	SimulationContext with transect_tables populated.

◆ build() [1/2]

void openswmm::XSectGroups::build ( const SimulationContext & ctx )

Build shape groups from SimulationContext link data.

Scans all links, groups them by xsect shape, and copies the geometry parameters into contiguous SoA arrays. Only shapes that have at least one link get a group.

Parameters

ctx	SimulationContext (must have links populated).

◆ build() [2/2]

void openswmm::XSectGroups::build	(	const XSectParams *	params,
		int	n_links )

Build shape groups from an array of XSectParams.

Alternative to build(ctx) — useful for testing.

Parameters

params	Array of per-link XSectParams.
n_links	Number of links.

◆ computeAreaAndHydRad()

void openswmm::XSectGroups::computeAreaAndHydRad	(	const double *	depths,
		double *	areas,
		double *	hydrad,
		int	n_links ) const

Fused area + hydraulic radius computation (single gather/scatter).

Parameters

depths	[in] Global depth array.
areas	[out] Global area array.
hydrad	[out] Global hydraulic radius array.
n_links	Total number of links.

◆ computeAreaHydRadTriple()

void openswmm::XSectGroups::computeAreaHydRadTriple	(	const double *	d1,
		const double *	d2,
		const double *	dm,
		double *	a1,
		double *	a2,
		double *	am,
		double *	hrad1,
		double *	hrad_mid,
		int	n_links ) const

Fused triple: d1→(a1,hrad1), d2→a2, dm→(am,hrad_mid) in one pass over shape groups. Replaces three separate compute calls in STEP B / STEP D of computeLinkGeometry.

Here is the call graph for this function:

◆ computeAreaHydRadTripleTeam()

void openswmm::XSectGroups::computeAreaHydRadTripleTeam	(	const double *	d1,
		const double *	d2,
		const double *	dm,
		double *	a1,
		double *	a2,
		double *	am,
		double *	hrad1,
		double *	hrad_mid,
		int	n_links,
		int	tid,
		int	nthreads ) const

Team-callable triple kernels: thread tid of nthreads processes only its static slice of every shape group.

Call from EVERY thread of an OpenMP team (typically the persistent DW Picard team) with that thread's id/team size; the caller owns the barrier that orders the outputs for consumers. Slices are disjoint (single-producer per element, including the shared group-local scratch buffers, which are sliced on cache-line boundaries), and per-element results are position-independent — so results are bit-identical to the serial triple at any thread count. Honors the bypass mask exactly like the serial forms. (tid=0, nthreads=1) IS the serial form.

Here is the caller graph for this function:

◆ computeAreas()

void openswmm::XSectGroups::computeAreas	(	const double *	depths,
		double *	areas,
		int	n_links ) const

Compute area for every link, reading depth from depths[link].

Parameters

depths	[in] Global depth array (indexed by link).
areas	[out] Global area array (indexed by link).
n_links	Total number of links.

◆ computeDepthsFromArea()

void openswmm::XSectGroups::computeDepthsFromArea	(	const double *	areas,
		double *	depths,
		int	n_links ) const

Compute depth from area for every link (inverse).

Parameters

areas	[in] Global area array.
depths	[out] Global depth array.
n_links	Total number of links.

Here is the call graph for this function:

◆ computeHydRad()

void openswmm::XSectGroups::computeHydRad	(	const double *	depths,
		double *	hydrad,
		int	n_links ) const

Compute hydraulic radius for every link.

Parameters

depths	[in] Global depth array.
hydrad	[out] Global hydraulic radius array.
n_links	Total number of links.

◆ computeSectionFactors()

void openswmm::XSectGroups::computeSectionFactors	(	const double *	areas,
		double *	sfact,
		int	n_links ) const

Compute section factor for every link (from area, not depth).

Parameters

areas	[in] Global area array.
sfact	[out] Global section factor array.
n_links	Total number of links.

Here is the call graph for this function:

◆ computeWidths()

void openswmm::XSectGroups::computeWidths	(	const double *	depths,
		double *	widths,
		int	n_links ) const

Compute top width for every link.

Parameters

depths	[in] Global depth array.
widths	[out] Global top width array.
n_links	Total number of links.

◆ computeWidthsTriple()

void openswmm::XSectGroups::computeWidthsTriple	(	const double *	d1,
		const double *	d2,
		const double *	dm,
		double *	w1,
		double *	w2,
		double *	wm,
		int	n_links ) const

Fused triple: d1→w1, d2→w2, dm→wm in one pass over shape groups.

Here is the call graph for this function:

◆ computeWidthsTripleTeam()

void openswmm::XSectGroups::computeWidthsTripleTeam	(	const double *	d1,
		const double *	d2,
		const double *	dm,
		double *	w1,
		double *	w2,
		double *	wm,
		int	n_links,
		int	tid,
		int	nthreads ) const

Here is the caller graph for this function:

◆ findGroup()

const ShapeGroup * openswmm::XSectGroups::findGroup ( XSectShape shape ) const

Find the group for a given shape (returns nullptr if no links have that shape).

◆ group()

const ShapeGroup & openswmm::XSectGroups::group ( int i ) const

inline

Access a specific shape group.

◆ numGroups()

int openswmm::XSectGroups::numGroups ( ) const

inline

Number of non-empty shape groups.

◆ setBypassMask()

void openswmm::XSectGroups::setBypassMask ( const std::uint8_t * bypassed_by_link ) const

Restrict the triple kernels to non-bypassed links for the current Picard iteration.

Legacy findLinkFlows skips a "bypassed" link (both end nodes converged) entirely, leaving its geometry at the previously computed values. The batch triple kernels instead recompute every link from its (unchanged) stored depths — producing bit-identical values, i.e. pure waste that grows with the bypass fraction (Bellinge averages ~9 Picard iterations per step, so most kernel invocations run with a mostly-converged network). This packs each shape group down to its active links so the kernels and gather/scatter touch only those. Bypassed links keep their stored outputs, exactly as legacy.

Pass nullptr to clear (all links active, zero overhead). The mask only affects computeWidthsTriple and computeAreaHydRadTriple; all other compute* methods are init/reporting-path and stay unmasked.

Parameters

bypassed_by_link Per-link flags indexed by GLOBAL link index (1 = bypassed/skip), or nullptr for all-active.

Note: const: the mask is per-iteration scratch (mutable members), set through the same const XSectGroups* the solver computes with.

The documentation for this class was generated from the following files:

/home/runner/work/openswmm.engine/openswmm.engine/src/engine/hydraulics/XSectBatch.hpp
/home/runner/work/openswmm.engine/openswmm.engine/src/engine/hydraulics/XSectBatch.cpp

Public Member Functions

Detailed Description

Member Function Documentation

◆ attachTransectTables()

◆ build() [1/2]

◆ build() [2/2]

◆ computeAreaAndHydRad()

◆ computeAreaHydRadTriple()

◆ computeAreaHydRadTripleTeam()

◆ computeAreas()

◆ computeDepthsFromArea()

◆ computeHydRad()

◆ computeSectionFactors()

◆ computeWidths()

◆ computeWidthsTriple()

◆ computeWidthsTripleTeam()

◆ findGroup()

◆ group()

◆ numGroups()

◆ setBypassMask()