statistics_accumulator.h File Reference

Per-window PETSc accumulator storage and pointwise application. More...

#include "variables.h"
#include "statistics_window.h"

Include dependency graph for statistics_accumulator.h:

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

Go to the source code of this file.

Data Structures
struct	PicurvWindowStorage
	Independent accumulator state for one window on one block. More...
struct	PicurvStatisticsPayload
	One checkpointable accumulator vector, resolved by enumeration index. More...
struct	PicurvDerivedField
	One derived output field, resolved by enumeration index. More...

Macros
#define	PICURV_STATISTICS_PAYLOAD_NAME_LENGTH   96
	Maximum stored length of a payload name, including the terminator.
#define	PICURV_STATISTICS_VARIANCE_FLOOR   1.0e-12
	Tolerance within which a negative variance is treated as floating-point noise.

Typedefs
typedef struct PicurvWindowStorage	PicurvWindowStorage
	Independent accumulator state for one window on one block.

Functions
PetscErrorCode	PicurvWindowStoragePayloadCount (const PicurvWindowStorage *storage, PetscInt *count)
	Reports how many checkpointable vectors one window's storage holds.
PetscErrorCode	PicurvWindowStoragePayload (UserCtx *user, const PicurvWindowDefinition *definition, const PicurvWindowStorage *storage, PetscInt index, PicurvStatisticsPayload *payload)
	Resolves one enumerated payload of a window's storage.
PetscErrorCode	PicurvStatisticsComponentDM (UserCtx *user, PetscInt components, DM *dm)
	Resolves the DM carrying a given number of accumulator components.
PetscErrorCode	PicurvProductComponentCount (PetscInt dof, PetscInt *count)
	Reports how many symmetric product components a field's second moment needs.
PetscErrorCode	PicurvCovarianceComponentCount (PetscInt dof_a, PetscInt dof_b, PetscInt *count)
	Reports how many components a covariance between two fields needs.
PetscErrorCode	PicurvWindowStorageCreate (UserCtx *user, const PicurvWindowDefinition *definition, PicurvWindowStorage *storage)
	Allocates the accumulator state one window owns on one block.
PetscErrorCode	PicurvWindowStorageDestroy (PicurvWindowStorage *storage)
	Releases accumulator state previously created for one window.
PetscErrorCode	PicurvWindowDerivedCount (const PicurvWindowDefinition *definition, const PicurvWindowStorage *storage, const char *outputs, PetscInt *count)
	Reports how many derived fields a requested output set produces.
PetscErrorCode	PicurvWindowDerive (UserCtx *user, const PicurvWindowDefinition *definition, const PicurvWindowStorage *storage, const char *outputs, PetscInt index, Vec scalar_target, Vec vector_target, PicurvDerivedField *field)
	Derives one output field from centered accumulator state.
PetscErrorCode	PicurvWindowSpatialMean (UserCtx *user, const PicurvWindowDefinition *definition, const PicurvWindowStorage *storage, Vec field, PetscReal *mean)
	Reports the spatial mean of a derived field over the points a window sampled.
PetscErrorCode	PicurvWindowValidFractionRange (UserCtx *user, const PicurvWindowDefinition *definition, const PicurvWindowStorage *storage, PetscInt sample_count, PetscReal *minimum, PetscReal *maximum)
	Reports the range of per-point valid fraction across a window's domain.
PetscErrorCode	PicurvWindowAccumulate (UserCtx *user, const PicurvWindowDefinition *definition, PicurvWindowStorage *storage, PetscReal weight)
	Applies one accepted completed state to a window's accumulators.

Detailed Description

Per-window PETSc accumulator storage and pointwise application.

Holds the independent state each named window owns, and applies one accepted completed state to it, per 4. Accumulated State and Component Order.

Storage is allocated once by the vector factory and released once at teardown. Application is strictly pointwise: it reads a source field value at an owned point and updates that point's accumulator, so it performs no halo exchange and allocates nothing.

Definition in file statistics_accumulator.h.

---

Data Structure Documentation

PicurvWindowStorage

struct PicurvWindowStorage

Independent accumulator state for one window on one block.

Per-point occupancy is tracked separately from the field moments because the fluid mask can move: a point contributes only to the states in which it was valid, so its own count and weight are what normalize its moments.

Every product is one vector carrying all of its components, not one vector per component. A symmetric second-order tensor is a single object: splitting it would cost six memory streams in the per-step accumulation loop instead of one cache line, and six collective gathers per checkpoint instead of one. Component counts that neither da nor fda provides are carried by a DM mirroring the block decomposition at that degree of freedom.

Definition at line 34 of file statistics_accumulator.h.

Data Fields
PetscInt	field_count	Fields accumulated.
PetscInt	covariance_count	Covariance pairs accumulated.
Vec	count	Per-point accepted sample count.
Vec	weight	Per-point valid weight.
Vec	weight_sq	Per-point squared-weight sum.
Vec *	mean	One per field, matching that field's layout.
Vec *	m2	One per field; NULL when no second moment was requested.
Vec *	cm	One per covariance pair.

PicurvStatisticsPayload

struct PicurvStatisticsPayload

One checkpointable accumulator vector, resolved by enumeration index.

The enumeration order is the persistence contract: the manifest inventory, the checkpoint writer, and the restart reader all walk it identically, so a payload lands in the vector it came from without a separate lookup table.

Definition at line 55 of file statistics_accumulator.h.

Data Fields
char	name[96]	File basename, no extension.
Vec	vec	Borrowed accumulator vector; never owned by the caller.
PetscInt	components	Degrees of freedom the vector carries.
const char *	role	Inventory role: occupancy, mean, second_moment, co_moment.
const char *	layout	Catalog layout name for the inventory entry.

PicurvDerivedField

struct PicurvDerivedField

One derived output field, resolved by enumeration index.

Definition at line 145 of file statistics_accumulator.h.

Data Fields
char	name[96]	Output field name, window qualified.
PetscInt	components	One or three.

Macro Definition Documentation

PICURV_STATISTICS_PAYLOAD_NAME_LENGTH

#define PICURV_STATISTICS_PAYLOAD_NAME_LENGTH   96

Maximum stored length of a payload name, including the terminator.

Definition at line 46 of file statistics_accumulator.h.

PICURV_STATISTICS_VARIANCE_FLOOR

#define PICURV_STATISTICS_VARIANCE_FLOOR   1.0e-12

Tolerance within which a negative variance is treated as floating-point noise.

Definition at line 142 of file statistics_accumulator.h.

Typedef Documentation

PicurvWindowStorage

typedef struct PicurvWindowStorage PicurvWindowStorage

Independent accumulator state for one window on one block.

Per-point occupancy is tracked separately from the field moments because the fluid mask can move: a point contributes only to the states in which it was valid, so its own count and weight are what normalize its moments.

Every product is one vector carrying all of its components, not one vector per component. A symmetric second-order tensor is a single object: splitting it would cost six memory streams in the per-step accumulation loop instead of one cache line, and six collective gathers per checkpoint instead of one. Component counts that neither da nor fda provides are carried by a DM mirroring the block decomposition at that degree of freedom.

Function Documentation

PicurvWindowStoragePayloadCount()

PetscErrorCode PicurvWindowStoragePayloadCount	(	const PicurvWindowStorage *	storage,
		PetscInt *	count
	)

Reports how many checkpointable vectors one window's storage holds.

Parameters

[in]	storage	Storage to measure.
[out]	count	Payload count, including the three occupancy vectors.

Returns

Zero on success, or PETSC_ERR_ARG_NULL for a null argument.

Reports how many checkpointable vectors one window's storage holds.

See also

PicurvWindowStoragePayloadCount()

Definition at line 268 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    PetscCheck(storage != NULL && count != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
               "Storage and count output are required.");
    *count = 3 + storage->field_count + storage->covariance_count;
    /* A field without a requested second moment holds no product vector, so the
     * enumeration skips it rather than emitting an empty payload. */
    for (PetscInt field_index = 0; storage->m2 && field_index < storage->field_count; ++field_index) {
        if (storage->m2[field_index]) *count += 1;
    }
    PetscFunctionReturn(0);
}

Here is the caller graph for this function:

PicurvWindowStoragePayload()

PetscErrorCode PicurvWindowStoragePayload	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		const PicurvWindowStorage *	storage,
		PetscInt	index,
		PicurvStatisticsPayload *	payload
	)

Resolves one enumerated payload of a window's storage.

Names are derived from catalogued field names and fixed component suffixes, so they are stable across runs, rank counts, and configuration reorderings.

Parameters

[in]	user	Block context the storage belongs to.
[in]	definition	Window definition naming the accumulated fields and pairs.
[in]	storage	Storage to enumerate.
[in]	index	Payload index in [0, count).
[out]	payload	Resolved payload; the vector is borrowed, not duplicated.

Returns

Zero on success, or PETSC_ERR_ARG_OUTOFRANGE for an index outside the range.

Resolves one enumerated payload of a window's storage.

See also

PicurvWindowStoragePayload()

Definition at line 286 of file statistics_accumulator.c.

{
    PetscInt total = 0;
    PetscInt cursor = index;
 
    PetscFunctionBeginUser;
    PetscCheck(user != NULL && definition != NULL && storage != NULL && payload != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
               "Context, definition, storage, and payload output are required.");
    PetscCall(PicurvWindowStoragePayloadCount(storage, &total));
    PetscCheck(index >= 0 && index < total, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
               "Payload index %" PetscInt_FMT " is outside [0, %" PetscInt_FMT ").", index, total);
    PetscCall(PetscMemzero(payload, sizeof(*payload)));
 
    /* Occupancy first, then means, then products, then co-moments. The order is the
     * persistence contract: the manifest inventory, the writer, and the reader all
     * walk it identically. */
    if (cursor < 3) {
        static const char *const occupancy_name[3] = {"count", "weight", "weight_sq"};
        Vec occupancy[3];
 
        occupancy[0] = storage->count;
        occupancy[1] = storage->weight;
        occupancy[2] = storage->weight_sq;
        PetscCall(PetscStrncpy(payload->name, occupancy_name[cursor], sizeof(payload->name)));
        payload->vec = occupancy[cursor];
        payload->components = 1;
        payload->role = "occupancy";
        payload->layout = FieldLayoutName(FIELD_LAYOUT_CELL_CENTERED);
        PetscFunctionReturn(0);
    }
    cursor -= 3;
 
    if (cursor < storage->field_count) {
        const FieldDescriptor *descriptor = NULL;
 
        PetscCall(FieldGetDescriptor((FieldId)definition->fields[cursor].field_id, &descriptor));
        PetscCall(PetscSNPrintf(payload->name, sizeof(payload->name), "%s_mean",
                                descriptor->canonical_name));
        payload->vec = storage->mean[cursor];
        payload->components = descriptor->dof;
        payload->role = "mean";
        payload->layout = FieldLayoutName(descriptor->layout);
        PetscFunctionReturn(0);
    }
    cursor -= storage->field_count;
 
    {
        PetscInt product_count = 0;
        PetscInt seen = 0;
 
        for (PetscInt field_index = 0; storage->m2 && field_index < storage->field_count; ++field_index) {
            if (storage->m2[field_index]) ++product_count;
        }
        if (cursor < product_count) {
            for (PetscInt field_index = 0; field_index < storage->field_count; ++field_index) {
                const FieldDescriptor *descriptor = NULL;
 
                if (!storage->m2[field_index]) continue;
                if (seen++ != cursor) continue;
                PetscCall(FieldGetDescriptor((FieldId)definition->fields[field_index].field_id, &descriptor));
                PetscCall(PetscSNPrintf(payload->name, sizeof(payload->name), "%s_m2",
                                        descriptor->canonical_name));
                payload->vec = storage->m2[field_index];
                PetscCall(PicurvProductComponentCount(descriptor->dof, &payload->components));
                payload->role = "second_moment";
                payload->layout = FieldLayoutName(descriptor->layout);
                PetscFunctionReturn(0);
            }
        }
        cursor -= product_count;
    }
 
    {
        const FieldDescriptor *first = NULL;
        const FieldDescriptor *second = NULL;
 
        PetscCheck(cursor >= 0 && cursor < storage->covariance_count, PETSC_COMM_SELF, PETSC_ERR_PLIB,
                   "Payload index %" PetscInt_FMT " fell through the storage enumeration.", index);
        PetscCall(FieldGetDescriptor((FieldId)definition->covariances[cursor].first, &first));
        PetscCall(FieldGetDescriptor((FieldId)definition->covariances[cursor].second, &second));
        PetscCall(PetscSNPrintf(payload->name, sizeof(payload->name), "%s_%s_cm",
                                first->canonical_name, second->canonical_name));
        payload->vec = storage->cm[cursor];
        PetscCall(PicurvCovarianceComponentCount(first->dof, second->dof, &payload->components));
        payload->role = "co_moment";
        payload->layout = FieldLayoutName(first->layout);
    }
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvStatisticsComponentDM()

PetscErrorCode PicurvStatisticsComponentDM	(	UserCtx *	user,
		PetscInt	components,
		DM *	dm
	)

Resolves the DM carrying a given number of accumulator components.

Component counts of one and three reuse the DMs the block already owns; six is carried by the symmetric-tensor DM created alongside them. Every one of these mirrors the block decomposition exactly, so a pointwise loop can read a source field and write an accumulator at the same index.

Parameters

[in]	user	Block context owning the DMs.
[in]	components	Component count to place.
[out]	dm	Resolved DM; borrowed, never destroyed by the caller.

Returns

Zero on success, or PETSC_ERR_ARG_OUTOFRANGE for an unsupported count.

Resolves the DM carrying a given number of accumulator components.

See also

PicurvStatisticsComponentDM()

Definition at line 115 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    PetscCheck(user != NULL && dm != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
               "Context and DM output are required.");
    switch (components) {
    case 1: *dm = user->da;   break;
    case 3: *dm = user->fda;  break;
    case 6: *dm = user->fda6; break;
    default:
        SETERRQ(PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
                "No block DM carries %" PetscInt_FMT " accumulator components.", components);
    }
    PetscCheck(*dm != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE,
               "The block DM for %" PetscInt_FMT " accumulator components was never created.",
               components);
    PetscFunctionReturn(0);
}

Here is the caller graph for this function:

PicurvProductComponentCount()

PetscErrorCode PicurvProductComponentCount	(	PetscInt	dof,
		PetscInt *	count
	)

Reports how many symmetric product components a field's second moment needs.

Parameters

[in]	dof	Degree of freedom of the field.
[out]	count	Component count: one for a scalar, six for a three-vector.

Returns

Zero on success, or PETSC_ERR_ARG_OUTOFRANGE for an unsupported dof.

Reports how many symmetric product components a field's second moment needs.

See also

PicurvProductComponentCount()

Definition at line 83 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    PetscCheck(count != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Count output is required.");
    PetscCheck(dof == 1 || dof == 3, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
               "Self-products are supported for scalar and three-vector fields, got dof %" PetscInt_FMT ".", dof);
    *count = (dof == 1) ? 1 : 6;
    PetscFunctionReturn(0);
}

Here is the caller graph for this function:

PicurvCovarianceComponentCount()

PetscErrorCode PicurvCovarianceComponentCount	(	PetscInt	dof_a,
		PetscInt	dof_b,
		PetscInt *	count
	)

Reports how many components a covariance between two fields needs.

Parameters

[in]	dof_a	Degree of freedom of the first member.
[in]	dof_b	Degree of freedom of the second member.
[out]	count	Component count.

Returns

Zero on success, or PETSC_ERR_ARG_OUTOFRANGE for an unsupported pairing.

Reports how many components a covariance between two fields needs.

See also

PicurvCovarianceComponentCount()

Definition at line 97 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    PetscCheck(count != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Count output is required.");
    /* Scalar-scalar and vector-scalar pairs only; a vector-vector cross product
     * would need a nine-component carrier that nothing allocates. */
    PetscCheck((dof_a == 1 && dof_b == 1) || (dof_a == 3 && dof_b == 1) || (dof_a == 1 && dof_b == 3),
               PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
               "Covariance is supported for scalar-scalar and vector-scalar pairs, got dof %" PetscInt_FMT
               " and %" PetscInt_FMT ".", dof_a, dof_b);
    *count = (dof_a == 3 || dof_b == 3) ? 3 : 1;
    PetscFunctionReturn(0);
}

Here is the caller graph for this function:

PicurvWindowStorageCreate()

PetscErrorCode PicurvWindowStorageCreate	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		PicurvWindowStorage *	storage
	)

Allocates the accumulator state one window owns on one block.

Every vector is duplicated from one the factory already built, so no new DM or layout decision is introduced.

Parameters

[in]	user	Block context supplying the source fields.
[in]	definition	Window definition naming the requested fields and pairs.
[out]	storage	Storage to populate; zeroed on entry.

Returns

Zero on success, or a PETSc error for an unknown field or unsupported layout.

Allocates the accumulator state one window owns on one block.

See also

PicurvWindowStorageCreate()

Definition at line 140 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    PROFILE_FUNCTION_BEGIN;
    PetscCheck(user != NULL && definition != NULL && storage != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Context, definition, and storage are required.");
    PetscCall(PetscMemzero(storage, sizeof(*storage)));
    storage->field_count = definition->field_count;
    storage->covariance_count = definition->covariance_count;
 
    /* Occupancy is per point and scalar regardless of what is accumulated. */
    PetscCall(DMCreateGlobalVector(user->da, &storage->count));
    PetscCall(VecSet(storage->count, 0.0));
    PetscCall(DMCreateGlobalVector(user->da, &storage->weight));
    PetscCall(VecSet(storage->weight, 0.0));
    PetscCall(DMCreateGlobalVector(user->da, &storage->weight_sq));
    PetscCall(VecSet(storage->weight_sq, 0.0));
 
    if (definition->field_count > 0) {
        PetscCall(PetscCalloc1((size_t)definition->field_count, &storage->mean));
        PetscCall(PetscCalloc1((size_t)definition->field_count, &storage->m2));
    }
    for (PetscInt field_index = 0; field_index < definition->field_count; ++field_index) {
        FieldView view;
        PetscInt  components = 0;
        DM        product_dm = NULL;
 
        PetscCall(FieldGetView(user, (FieldId)definition->fields[field_index].field_id, &view));
        /* The mean matches the source field's own layout, so it comes from that
         * field's DM and inherits its decomposition. */
        PetscCall(DMCreateGlobalVector(view.dm, &storage->mean[field_index]));
        PetscCall(VecSet(storage->mean[field_index], 0.0));
        if (!definition->fields[field_index].want_second) continue;
        PetscCall(PicurvProductComponentCount(view.descriptor->dof, &components));
        PetscCall(PicurvStatisticsComponentDM(user, components, &product_dm));
        PetscCall(DMCreateGlobalVector(product_dm, &storage->m2[field_index]));
        PetscCall(VecSet(storage->m2[field_index], 0.0));
    }
 
    if (definition->covariance_count > 0) {
        PetscCall(PetscCalloc1((size_t)definition->covariance_count, &storage->cm));
    }
    for (PetscInt pair_index = 0; pair_index < definition->covariance_count; ++pair_index) {
        FieldView view_a, view_b;
        PetscInt  components = 0;
        DM        pair_dm = NULL;
 
        PetscCall(FieldGetView(user, (FieldId)definition->covariances[pair_index].first, &view_a));
        PetscCall(FieldGetView(user, (FieldId)definition->covariances[pair_index].second, &view_b));
        PetscCall(PicurvCovarianceComponentCount(view_a.descriptor->dof, view_b.descriptor->dof, &components));
        PetscCall(PicurvStatisticsComponentDM(user, components, &pair_dm));
        PetscCall(DMCreateGlobalVector(pair_dm, &storage->cm[pair_index]));
        PetscCall(VecSet(storage->cm[pair_index], 0.0));
    }
 
    /* Report what this window costs and what it covers, once per block at setup.
     * The point count is a function of layout, dimensions, and periodicity alone, so
     * it tells an operator what "per point" will mean before any sample is taken. */
    {
        SpatialTargetPlan plan;
        PetscInt payloads = 0;
        PetscInt points = 0;
 
        PetscCall(PicurvWindowStoragePayloadCount(storage, &payloads));
        PetscCall(SpatialTargetPlanCreate(user, (FieldId)definition->fields[0].field_id,
                                          PICURV_STATISTICS_MASK_FLUID, &plan));
        PetscCall(SpatialTargetPlanGlobalPointCount(&plan, PETSC_COMM_WORLD, &points));
        LOG_ALLOW(LOCAL, LOG_DEBUG,
                  "Statistics window '%s' block %d: %d accumulator vector(s) over %d point(s).\n",
                  definition->name, (int)user->_this, (int)payloads, (int)points);
    }
    PROFILE_FUNCTION_END;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvWindowStorageDestroy()

PetscErrorCode PicurvWindowStorageDestroy

(

PicurvWindowStorage *

storage

)

Releases accumulator state previously created for one window.

Parameters

[in,out]

storage

Storage to release; safe to call on zeroed storage.

Returns

Zero on success.

Releases accumulator state previously created for one window.

See also

PicurvWindowStorageDestroy()

Definition at line 220 of file statistics_accumulator.c.

{
    PetscFunctionBeginUser;
    if (storage == NULL) PetscFunctionReturn(0);
    if (storage->count) PetscCall(VecDestroy(&storage->count));
    if (storage->weight) PetscCall(VecDestroy(&storage->weight));
    if (storage->weight_sq) PetscCall(VecDestroy(&storage->weight_sq));
    for (PetscInt field_index = 0; storage->mean && field_index < storage->field_count; ++field_index) {
        if (storage->mean[field_index]) PetscCall(VecDestroy(&storage->mean[field_index]));
    }
    for (PetscInt field_index = 0; storage->m2 && field_index < storage->field_count; ++field_index) {
        if (storage->m2[field_index]) PetscCall(VecDestroy(&storage->m2[field_index]));
    }
    for (PetscInt pair_index = 0; storage->cm && pair_index < storage->covariance_count; ++pair_index) {
        if (storage->cm[pair_index]) PetscCall(VecDestroy(&storage->cm[pair_index]));
    }
    if (storage->mean) PetscCall(PetscFree(storage->mean));
    if (storage->m2) PetscCall(PetscFree(storage->m2));
    if (storage->cm) PetscCall(PetscFree(storage->cm));
    PetscCall(PetscMemzero(storage, sizeof(*storage)));
    PetscFunctionReturn(0);
}

Here is the caller graph for this function:

PicurvWindowDerivedCount()

PetscErrorCode PicurvWindowDerivedCount	(	const PicurvWindowDefinition *	definition,
		const PicurvWindowStorage *	storage,
		const char *	outputs,
		PetscInt *	count
	)

Reports how many derived fields a requested output set produces.

Parameters

[in]	definition	Window definition naming the accumulated fields and pairs.
[in]	storage	Accumulator state the outputs are derived from.
[in]	outputs	Comma-separated output kinds: mean, reynolds_stress, rms, tke, flux.
[out]	count	Number of derived fields.

Returns

Zero on success, or PETSC_ERR_ARG_WRONG for an unknown output kind.

Reports how many derived fields a requested output set produces.

See also

PicurvWindowDerivedCount()

Definition at line 501 of file statistics_accumulator.c.

{
    PetscBool wanted[DERIVED_KIND_COUNT];
 
    PetscFunctionBeginUser;
    PetscCheck(definition != NULL && storage != NULL && count != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Definition, storage, and count are required.");
    PetscCall(ParseDerivedKinds(outputs, wanted));
    *count = 0;
    for (PetscInt k = 0; k < DERIVED_KIND_COUNT; ++k) {
        PetscInt extent = 0;
 
        if (!wanted[k]) continue;
        PetscCall(DerivedKindExtent(definition, storage, (DerivedKind)k, &extent));
        *count += extent;
    }
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvWindowDerive()

PetscErrorCode PicurvWindowDerive	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		const PicurvWindowStorage *	storage,
		const char *	outputs,
		PetscInt	index,
		Vec	scalar_target,
		Vec	vector_target,
		PicurvDerivedField *	field
	)

Derives one output field from centered accumulator state.

Normalizes in exactly one place: R_ij = C_ij / W, RMS_i = sqrt(R_ii), k = (R_xx + R_yy + R_zz) / 2, and a flux is the co-moment over the same weight. Each uses the moment kernels rather than repeating the division, so the online and offline halves of the pipeline cannot disagree about what centered state means.

A variance that comes out slightly negative through floating-point cancellation is clamped only where a square root would otherwise fail, and only within PICURV_STATISTICS_VARIANCE_FLOOR. Stored state is never modified.

Points the window never sampled are left at zero rather than divided by a zero weight; the valid-fraction range reports how much of the domain that covers.

Parameters

[in]	user	Block context supplying the target domain.
[in]	definition	Window definition.
[in]	storage	Accumulator state to read.
[in]	outputs	Comma-separated output kinds.
[in]	index	Derived field index in [0, count).
[out]	scalar_target	Scalar destination, used when the field has one component.
[out]	vector_target	Vector destination, used when the field has three.
[out]	field	Resolved name and component count of the derived field.

Returns

Zero on success, or a PETSc error.

Derives one output field from centered accumulator state.

See also

PicurvWindowDerive()

Definition at line 576 of file statistics_accumulator.c.

{
    PetscBool wanted[DERIVED_KIND_COUNT];
    DerivedKind kind = DERIVED_MEAN;
    SpatialTargetPlan plan;
    PetscReal ***weight_arr = NULL, ***weight_sq_arr = NULL, ***count_arr = NULL;
    PetscScalar ****source = NULL, ****target = NULL;
    const FieldDescriptor *descriptor = NULL;
    DM source_dm = NULL;
    Vec source_vec = NULL;
    PetscInt offset = 0, slot = 0, member = 0;
    /* Resolved where the branch knows which fields it read; 1.0 leaves the result in
     * solver units, which is also what an undeclared scale falls back to. */
    PetscReal dimensional_factor = 1.0;
    /* The source's component count is not the output's: a scalar RMS reads a
     * six-component tensor, so each needs its own DM. */
    PetscInt source_components = 0;
 
    PetscFunctionBeginUser;
    PROFILE_FUNCTION_BEGIN;
    PetscCheck(user != NULL && definition != NULL && storage != NULL && field != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
               "Context, definition, storage, and output are required.");
    PetscCheck(index >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
               "Derived index %" PetscInt_FMT " is negative.", index);
    PetscCall(PetscMemzero(field, sizeof(*field)));
    /* The output list is parsed once and the extents walked once. Resolving the
     * index is what bounds-checks it, so counting first would repeat both. */
    PetscCall(ParseDerivedKinds(outputs, wanted));
    PetscCall(ResolveDerivedIndex(definition, storage, wanted, index, &kind, &offset));
 
    /* Locate the accumulator slot and component this index selects. */
    switch (kind) {
    case DERIVED_MEAN:
        slot = offset;
        PetscCall(FieldGetDescriptor((FieldId)definition->fields[slot].field_id, &descriptor));
        source_vec = storage->mean[slot];
        field->components = descriptor->dof;
        source_components = descriptor->dof;
        PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_mean",
                                definition->name, descriptor->canonical_name));
        break;
    case DERIVED_TKE:
        for (slot = 0; slot < storage->field_count; ++slot) {
            if (!storage->m2[slot]) continue;
            PetscCall(FieldGetDescriptor((FieldId)definition->fields[slot].field_id, &descriptor));
            if (descriptor->dof != 3) continue;
            if (offset-- == 0) break;
        }
        source_vec = storage->m2[slot];
        field->components = 1;
        PetscCall(PicurvProductComponentCount(descriptor->dof, &source_components));
        PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_tke",
                                definition->name, descriptor->canonical_name));
        break;
    case DERIVED_FLUX:
        slot = offset;
        {
            const FieldDescriptor *first = NULL;
            const FieldDescriptor *second = NULL;
 
            PetscCall(FieldGetDescriptor((FieldId)definition->covariances[slot].first, &first));
            PetscCall(FieldGetDescriptor((FieldId)definition->covariances[slot].second, &second));
            PetscCall(PicurvCovarianceComponentCount(first->dof, second->dof, &field->components));
            source_components = field->components;
            source_vec = storage->cm[slot];
            if (user->simCtx->pps && user->simCtx->pps->dimensionalize) {
                PetscReal first_scale = 1.0, second_scale = 1.0;
 
                /* A co-moment carries the product of its two fields' dimensions. */
                PetscCall(PicurvFieldReferenceScale(user->simCtx, first->canonical_name,
                                                    &first_scale, NULL, 0));
                PetscCall(PicurvFieldReferenceScale(user->simCtx, second->canonical_name,
                                                    &second_scale, NULL, 0));
                dimensional_factor = first_scale * second_scale;
            }
            PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_%s_flux",
                                    definition->name, first->canonical_name, second->canonical_name));
        }
        break;
    default:
        /* Stress and RMS both walk the fields carrying a product; stress emits every
         * symmetric component, RMS only the diagonal. */
        for (slot = 0; slot < storage->field_count; ++slot) {
            PetscInt extent = 0;
 
            if (!storage->m2[slot]) continue;
            PetscCall(FieldGetDescriptor((FieldId)definition->fields[slot].field_id, &descriptor));
            PetscCall(PicurvProductComponentCount(descriptor->dof, &extent));
            if (kind == DERIVED_RMS) extent = descriptor->dof;
            if (offset < extent) { member = offset; break; }
            offset -= extent;
        }
        source_vec = storage->m2[slot];
        field->components = 1;
        PetscCall(PicurvProductComponentCount(descriptor->dof, &source_components));
        if (kind == DERIVED_RMS) {
            PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_rms%s",
                                    definition->name, descriptor->canonical_name,
                                    descriptor->dof == 1 ? "" : kAxisName[member]));
        } else if (descriptor->dof == 1) {
            PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_variance",
                                    definition->name, descriptor->canonical_name));
        } else {
            char label[8];
 
            PetscCall(ProductComponentLabel(member, label, sizeof(label)));
            PetscCall(PetscSNPrintf(field->name, sizeof(field->name), "%s_%s_R_%s",
                                    definition->name, descriptor->canonical_name, label));
        }
        break;
    }
 
    PetscCall(SpatialTargetPlanCreate(user, (FieldId)definition->fields[0].field_id,
                                      PICURV_STATISTICS_MASK_FLUID, &plan));
    PetscCall(PicurvStatisticsComponentDM(user, source_components, &source_dm));
    {
        Vec destination = (field->components == 1) ? scalar_target : vector_target;
        DM destination_dm = NULL;
 
        PetscCheck(destination != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
                   "Derived field '%s' needs a %d-component destination.",
                   field->name, (int)field->components);
        PetscCall(PicurvStatisticsComponentDM(user, field->components, &destination_dm));
        PetscCall(VecZeroEntries(destination));
 
        PetscCall(DMDAVecGetArrayRead(user->da, storage->weight, &weight_arr));
        PetscCall(DMDAVecGetArrayRead(user->da, storage->weight_sq, &weight_sq_arr));
        PetscCall(DMDAVecGetArrayRead(user->da, storage->count, &count_arr));
        PetscCall(DMDAVecGetArrayDOFRead(source_dm, source_vec, &source));
        PetscCall(DMDAVecGetArrayDOF(destination_dm, destination, &target));
 
        for (PetscInt k = plan.start[2]; k < plan.end[2]; ++k) {
            for (PetscInt j = plan.start[1]; j < plan.end[1]; ++j) {
                for (PetscInt i = plan.start[0]; i < plan.end[0]; ++i) {
                    PicurvCoMomentState pair;
 
                    /* A point the window never sampled has no average at all, so it
                     * is left at zero rather than divided by a zero weight. */
                    if (weight_arr[k][j][i] <= 0.0) continue;
                    pair.count = count_arr[k][j][i];
                    pair.weight = weight_arr[k][j][i];
                    pair.weight_sq = weight_sq_arr[k][j][i];
                    pair.mean_x = 0.0;
                    pair.mean_y = 0.0;
 
                    if (kind == DERIVED_MEAN) {
                        for (PetscInt c = 0; c < field->components; ++c) {
                            target[k][j][i][c] = source[k][j][i][c];
                        }
                    } else if (kind == DERIVED_TKE) {
                        /* The trace of the stress tensor, halved: components 0, 3
                         * and 5 are xx, yy and zz in the stored symmetric order. */
                        PetscReal trace = 0.0;
 
                        for (PetscInt c = 0; c < 3; ++c) {
                            pair.cm = source[k][j][i][ProductDiagonalIndex(c)];
                            trace += PicurvCoMomentStateCovariance(&pair);
                        }
                        target[k][j][i][0] = 0.5 * trace;
                    } else if (kind == DERIVED_RMS) {
                        PetscReal deviation = 0.0;
 
                        pair.cm = source[k][j][i][(descriptor->dof == 1)
                                                  ? 0 : ProductDiagonalIndex(member)];
                        PetscCall(SafeStandardDeviation(PicurvCoMomentStateCovariance(&pair),
                                                        field->name, &deviation));
                        target[k][j][i][0] = deviation;
                    } else if (kind == DERIVED_FLUX) {
                        for (PetscInt c = 0; c < field->components; ++c) {
                            pair.cm = source[k][j][i][c];
                            target[k][j][i][c] = PicurvCoMomentStateCovariance(&pair);
                        }
                    } else {
                        pair.cm = source[k][j][i][member];
                        target[k][j][i][0] = PicurvCoMomentStateCovariance(&pair);
                    }
                }
            }
        }
 
        PetscCall(DMDAVecRestoreArrayDOF(destination_dm, destination, &target));
        PetscCall(DMDAVecRestoreArrayDOFRead(source_dm, source_vec, &source));
        PetscCall(DMDAVecRestoreArrayRead(user->da, storage->count, &count_arr));
        PetscCall(DMDAVecRestoreArrayRead(user->da, storage->weight_sq, &weight_sq_arr));
        PetscCall(DMDAVecRestoreArrayRead(user->da, storage->weight, &weight_arr));
 
        /* Applied once, here, so the VTK field and the convergence-history CSV - which
         * both read this staging vector - cannot end up in different unit systems. */
        if (user->simCtx->pps && user->simCtx->pps->dimensionalize) {
            if (kind != DERIVED_FLUX && descriptor) {
                PetscReal base = 1.0;
 
                PetscCall(PicurvFieldReferenceScale(user->simCtx, descriptor->canonical_name,
                                                    &base, NULL, 0));
                dimensional_factor = PetscPowRealInt(base, kDerivedKindScaleExponent[kind]);
            }
            if (PetscAbsReal(dimensional_factor - 1.0) > PETSC_MACHINE_EPSILON) {
                PetscCall(VecScale(destination, dimensional_factor));
                LOG_ALLOW(GLOBAL, LOG_DEBUG, "Scaled derived '%s' by %.4e.\n",
                          kDerivedKindName[kind], (double)dimensional_factor);
            }
        }
    }
    PROFILE_FUNCTION_END;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvWindowSpatialMean()

PetscErrorCode PicurvWindowSpatialMean	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		const PicurvWindowStorage *	storage,
		Vec	field,
		PetscReal *	mean
	)

Reports the spatial mean of a derived field over the points a window sampled.

The average is taken over targeted points that actually accumulated weight, not over the whole vector. A derived field is zero everywhere outside the target domain and at any point the mask never admitted, and those zeros are absences rather than measurements: including them would scale the answer down by the fraction of the vector the window never covered.

Performs a collective reduction, so callers use it for reporting rather than per point.

Parameters

[in]	user	Block context supplying the target domain.
[in]	definition	Window definition naming the accumulated fields.
[in]	storage	Accumulator state supplying per-point occupancy.
[in]	field	Derived field to average, on the cell-centred scalar DM.
[out]	mean	Spatial mean; zero when the window sampled no point.

Returns

Zero on success, or a PETSc error.

Reports the spatial mean of a derived field over the points a window sampled.

See also

PicurvWindowSpatialMean()

Definition at line 793 of file statistics_accumulator.c.

{
    SpatialTargetPlan plan;
    /* Every direction collapses: a window mean is one number for the domain. */
    const PetscBool   whole_domain[3] = {PETSC_TRUE, PETSC_TRUE, PETSC_TRUE};
 
    PetscFunctionBeginUser;
    PROFILE_FUNCTION_BEGIN;
    PetscCheck(user != NULL && definition != NULL && storage != NULL && field != NULL && mean != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL,
               "Context, definition, storage, field, and output are required.");
    *mean = 0.0;
    if (definition->field_count == 0) { PROFILE_FUNCTION_END; PetscFunctionReturn(0); }
 
    PetscCall(SpatialTargetPlanCreate(user, (FieldId)definition->fields[0].field_id,
                                      PICURV_STATISTICS_MASK_FLUID, &plan));
    /* No denominator means each admitted point counts once, so the result is the mean
     * of the field over those points rather than a volume-weighted average. The
     * accumulated weight is passed as the inclusion mask: a point the window never
     * sampled holds a zero that means "never measured", and averaging it in would
     * scale the answer down by the fraction of the domain the window never covered. */
    PetscCall(PicurvSpatialRatioAverage(user, &plan, field, NULL, storage->weight,
                                        whole_domain, PETSC_COMM_WORLD, NULL, mean));
    PROFILE_FUNCTION_END;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvWindowValidFractionRange()

PetscErrorCode PicurvWindowValidFractionRange	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		const PicurvWindowStorage *	storage,
		PetscInt	sample_count,
		PetscReal *	minimum,
		PetscReal *	maximum
	)

Reports the range of per-point valid fraction across a window's domain.

A point contributes only to the states in which the mask accepted it, so with a moving immersed body different points carry different sample counts. The ratio of a point's own count to the window's accepted-sample count is its valid fraction, and the range of that ratio is the window's mask-health indicator: a minimum of one means every point saw every state, and a minimum of zero means some point contributed nothing at all.

Performs a collective reduction, so callers gate it on an already-active reporting path rather than calling it every step.

Parameters

[in]	user	Block context supplying the target domain and mask.
[in]	definition	Window definition naming the accumulated fields.
[in]	storage	Accumulator state to inspect.
[in]	sample_count	Accepted states the window has recorded.
[out]	minimum	Smallest valid fraction; one when no point is targeted.
[out]	maximum	Largest valid fraction; zero when no point is targeted.

Returns

Zero on success, or a PETSc error.

Reports the range of per-point valid fraction across a window's domain.

See also

PicurvWindowValidFractionRange()

Definition at line 828 of file statistics_accumulator.c.

{
    SpatialTargetPlan plan;
    PetscReal ***nvert = NULL;
    PetscReal ***count_arr = NULL;
    PetscReal local_min = PETSC_MAX_REAL;
    PetscReal local_max = 0.0;
    PetscReal reduced_min = 0.0, reduced_max = 0.0;
 
    PetscFunctionBeginUser;
    PROFILE_FUNCTION_BEGIN;
    PetscCheck(user != NULL && definition != NULL && storage != NULL &&
               minimum != NULL && maximum != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Context, definition, storage, and outputs are required.");
    *minimum = 1.0;
    *maximum = 0.0;
    if (definition->field_count == 0 || sample_count <= 0) { PROFILE_FUNCTION_END; PetscFunctionReturn(0); }
 
    PetscCall(SpatialTargetPlanCreate(user, (FieldId)definition->fields[0].field_id,
                                      PICURV_STATISTICS_MASK_FLUID, &plan));
    PetscCall(DMDAVecGetArrayRead(user->da, user->Nvert, &nvert));
    PetscCall(DMDAVecGetArrayRead(user->da, storage->count, &count_arr));
    for (PetscInt k = plan.start[2]; k < plan.end[2]; ++k) {
        for (PetscInt j = plan.start[1]; j < plan.end[1]; ++j) {
            for (PetscInt i = plan.start[0]; i < plan.end[0]; ++i) {
                /* The mask is evaluated at the current state, but a point excluded
                 * now may have contributed earlier, so its stored count is what
                 * decides its fraction rather than its present eligibility. */
                const PetscReal fraction = count_arr[k][j][i] / (PetscReal)sample_count;
 
                local_min = PetscMin(local_min, fraction);
                local_max = PetscMax(local_max, fraction);
            }
        }
    }
    PetscCall(DMDAVecRestoreArrayRead(user->da, storage->count, &count_arr));
    PetscCall(DMDAVecRestoreArrayRead(user->da, user->Nvert, &nvert));
 
    /* A rank owning no targeted point must not drag the minimum down, so its
     * sentinel is neutral under the reduction rather than zero. */
    PetscCallMPI(MPI_Allreduce(&local_min, &reduced_min, 1, MPIU_REAL, MPIU_MIN, PETSC_COMM_WORLD));
    PetscCallMPI(MPI_Allreduce(&local_max, &reduced_max, 1, MPIU_REAL, MPIU_MAX, PETSC_COMM_WORLD));
    *minimum = (reduced_min == PETSC_MAX_REAL) ? 1.0 : reduced_min;
    *maximum = reduced_max;
    PROFILE_FUNCTION_END;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function:

PicurvWindowAccumulate()

PetscErrorCode PicurvWindowAccumulate	(	UserCtx *	user,
		const PicurvWindowDefinition *	definition,
		PicurvWindowStorage *	storage,
		PetscReal	weight
	)

Applies one accepted completed state to a window's accumulators.

Iterates the pointwise target domain, skipping points the mask rejects, and updates each point's occupancy and every requested moment and co-moment through the centered kernels.

Parameters

[in]	user	Block context supplying the source fields.
[in]	definition	Window definition.
[in,out]	storage	Accumulator state to update.
[in]	weight	Weight the window assigned to this state; must be positive.

Returns

Zero on success, or a PETSc error.

Applies one accepted completed state to a window's accumulators.

See also

PicurvWindowAccumulate()

Definition at line 884 of file statistics_accumulator.c.

{
    SpatialTargetPlan plan;
    PetscReal ***nvert = NULL;
    PetscReal ***count_arr = NULL, ***weight_arr = NULL, ***weight_sq_arr = NULL;
 
    PetscFunctionBeginUser;
    PROFILE_FUNCTION_BEGIN;
    PetscCheck(user != NULL && definition != NULL && storage != NULL,
               PETSC_COMM_SELF, PETSC_ERR_ARG_NULL, "Context, definition, and storage are required.");
    PetscCheck(weight > 0.0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE,
               "Accepted states carry a positive weight, got %g.", (double)weight);
    if (definition->field_count == 0) { PROFILE_FUNCTION_END; PetscFunctionReturn(0); }
 
    /* Every requested field shares the pointwise cell-centered domain, so the plan
     * is resolved once rather than per field. */
    PetscCall(SpatialTargetPlanCreate(user, (FieldId)definition->fields[0].field_id,
                                      PICURV_STATISTICS_MASK_FLUID, &plan));
 
    PetscCall(DMDAVecGetArrayRead(user->da, user->Nvert, &nvert));
    PetscCall(DMDAVecGetArray(user->da, storage->count, &count_arr));
    PetscCall(DMDAVecGetArray(user->da, storage->weight, &weight_arr));
    PetscCall(DMDAVecGetArray(user->da, storage->weight_sq, &weight_sq_arr));
 
    /* Pass one advances per-point occupancy, so every field pass afterwards can
     * recover the pre-state weight uniformly as (stored weight - this weight).
     * Occupancy is per point and per accepted state, never per field. */
    for (PetscInt k = plan.start[2]; k < plan.end[2]; ++k) {
        for (PetscInt j = plan.start[1]; j < plan.end[1]; ++j) {
            for (PetscInt i = plan.start[0]; i < plan.end[0]; ++i) {
                if (!SpatialTargetPlanMaskAllows(&plan, nvert[k][j][i])) continue;
                count_arr[k][j][i]     += 1.0;
                weight_arr[k][j][i]    += weight;
                weight_sq_arr[k][j][i] += weight * weight;
            }
        }
    }
 
    /* Pass two updates every cross-field co-moment. It runs before any mean is
     * written back, because a co-moment needs the pre-update mean of *both*
     * members, and pass three overwrites them field by field. */
    for (PetscInt pair = 0; pair < definition->covariance_count; ++pair) {
        FieldView view_a, view_b;
        PetscScalar ****src_a = NULL, ****mean_a = NULL;
        PetscScalar ****src_b = NULL, ****mean_b = NULL;
        PetscScalar ****co_moment = NULL;
        DM pair_dm = NULL;
        PetscInt slot_a = 0, slot_b = 0, dof_a = 0, dof_b = 0, components = 0;
 
        PetscCall(FindFieldSlot(definition, definition->covariances[pair].first, &slot_a));
        PetscCall(FindFieldSlot(definition, definition->covariances[pair].second, &slot_b));
        PetscCall(FieldGetView(user, (FieldId)definition->covariances[pair].first, &view_a));
        PetscCall(FieldGetView(user, (FieldId)definition->covariances[pair].second, &view_b));
        dof_a = view_a.descriptor->dof;
        dof_b = view_b.descriptor->dof;
        PetscCall(PicurvCovarianceComponentCount(dof_a, dof_b, &components));
        PetscCall(PicurvStatisticsComponentDM(user, components, &pair_dm));
 
        /* A component-indexed view serves every degree of freedom, so the loop
         * below needs no scalar-versus-vector branching. */
        PetscCall(DMDAVecGetArrayDOFRead(view_a.dm, view_a.global_vec, &src_a));
        PetscCall(DMDAVecGetArrayDOFRead(view_a.dm, storage->mean[slot_a], &mean_a));
        PetscCall(DMDAVecGetArrayDOFRead(view_b.dm, view_b.global_vec, &src_b));
        PetscCall(DMDAVecGetArrayDOFRead(view_b.dm, storage->mean[slot_b], &mean_b));
        PetscCall(DMDAVecGetArrayDOF(pair_dm, storage->cm[pair], &co_moment));
 
        for (PetscInt k = plan.start[2]; k < plan.end[2]; ++k) {
            for (PetscInt j = plan.start[1]; j < plan.end[1]; ++j) {
                for (PetscInt i = plan.start[0]; i < plan.end[0]; ++i) {
                    if (!SpatialTargetPlanMaskAllows(&plan, nvert[k][j][i])) continue;
                    for (PetscInt c = 0; c < components; ++c) {
                        /* A scalar member contributes its single value against every
                         * component of a vector member, which is what makes a
                         * vector-scalar covariance a three-component object. */
                        const PetscInt component_a = (dof_a == 3) ? c : 0;
                        const PetscInt component_b = (dof_b == 3) ? c : 0;
                        PicurvCoMomentState state;
 
                        state.count = count_arr[k][j][i] - 1.0;
                        state.weight = weight_arr[k][j][i] - weight;
                        state.weight_sq = weight_sq_arr[k][j][i] - weight * weight;
                        state.mean_x = mean_a[k][j][i][component_a];
                        state.mean_y = mean_b[k][j][i][component_b];
                        state.cm = co_moment[k][j][i][c];
                        PetscCall(PicurvCoMomentStateUpdate(&state,
                                                            src_a[k][j][i][component_a],
                                                            src_b[k][j][i][component_b], weight));
                        co_moment[k][j][i][c] = state.cm;
                    }
                }
            }
        }
 
        PetscCall(DMDAVecRestoreArrayDOF(pair_dm, storage->cm[pair], &co_moment));
        PetscCall(DMDAVecRestoreArrayDOFRead(view_b.dm, storage->mean[slot_b], &mean_b));
        PetscCall(DMDAVecRestoreArrayDOFRead(view_b.dm, view_b.global_vec, &src_b));
        PetscCall(DMDAVecRestoreArrayDOFRead(view_a.dm, storage->mean[slot_a], &mean_a));
        PetscCall(DMDAVecRestoreArrayDOFRead(view_a.dm, view_a.global_vec, &src_a));
    }
 
    /* Pass three updates each requested field's mean and, when asked, its centered
     * self-product. A three-vector's self-product is the six symmetric co-moments
     * between component pairs, not three per-component variances, so the co-moment
     * kernel drives every product component including the diagonal. */
    for (PetscInt field_index = 0; field_index < definition->field_count; ++field_index) {
        FieldView view;
        PetscScalar ****src = NULL, ****mean = NULL, ****product = NULL;
        DM product_dm = NULL;
        PetscInt dof = 0, components = 0;
        const PetscBool second = definition->fields[field_index].want_second;
 
        PetscCall(FieldGetView(user, (FieldId)definition->fields[field_index].field_id, &view));
        dof = view.descriptor->dof;
        PetscCall(DMDAVecGetArrayDOFRead(view.dm, view.global_vec, &src));
        PetscCall(DMDAVecGetArrayDOF(view.dm, storage->mean[field_index], &mean));
        if (second) {
            PetscCall(PicurvProductComponentCount(dof, &components));
            PetscCall(PicurvStatisticsComponentDM(user, components, &product_dm));
            PetscCall(DMDAVecGetArrayDOF(product_dm, storage->m2[field_index], &product));
        }
 
        for (PetscInt k = plan.start[2]; k < plan.end[2]; ++k) {
            for (PetscInt j = plan.start[1]; j < plan.end[1]; ++j) {
                for (PetscInt i = plan.start[0]; i < plan.end[0]; ++i) {
                    PetscReal prior_weight = 0.0;
                    PetscReal prior_weight_sq = 0.0;
                    PetscReal prior_count = 0.0;
 
                    if (!SpatialTargetPlanMaskAllows(&plan, nvert[k][j][i])) continue;
 
                    prior_weight    = weight_arr[k][j][i] - weight;
                    prior_weight_sq = weight_sq_arr[k][j][i] - weight * weight;
                    prior_count     = count_arr[k][j][i] - 1.0;
 
                    /* Products are updated before the means are written back,
                     * because the co-moment kernel needs the pre-update means. */
                    for (PetscInt c = 0; c < components; ++c) {
                        const PetscInt a = (dof == 1) ? 0 : kProductFirst[c];
                        const PetscInt b = (dof == 1) ? 0 : kProductSecond[c];
                        PicurvCoMomentState state;
 
                        state.count = prior_count;
                        state.weight = prior_weight;
                        state.weight_sq = prior_weight_sq;
                        state.mean_x = mean[k][j][i][a];
                        state.mean_y = mean[k][j][i][b];
                        state.cm = product[k][j][i][c];
                        PetscCall(PicurvCoMomentStateUpdate(&state, src[k][j][i][a],
                                                            src[k][j][i][b], weight));
                        product[k][j][i][c] = state.cm;
                    }
 
                    for (PetscInt c = 0; c < dof; ++c) {
                        PicurvMomentState moment;
 
                        moment.count = prior_count;
                        moment.weight = prior_weight;
                        moment.weight_sq = prior_weight_sq;
                        moment.mean = mean[k][j][i][c];
                        moment.m2 = 0.0;
                        PetscCall(PicurvMomentStateUpdate(&moment, src[k][j][i][c], weight));
                        mean[k][j][i][c] = moment.mean;
                    }
                }
            }
        }
 
        if (second) PetscCall(DMDAVecRestoreArrayDOF(product_dm, storage->m2[field_index], &product));
        PetscCall(DMDAVecRestoreArrayDOF(view.dm, storage->mean[field_index], &mean));
        PetscCall(DMDAVecRestoreArrayDOFRead(view.dm, view.global_vec, &src));
    }
 
    PetscCall(DMDAVecRestoreArray(user->da, storage->weight_sq, &weight_sq_arr));
    PetscCall(DMDAVecRestoreArray(user->da, storage->weight, &weight_arr));
    PetscCall(DMDAVecRestoreArray(user->da, storage->count, &count_arr));
    PetscCall(DMDAVecRestoreArrayRead(user->da, user->Nvert, &nvert));
    PROFILE_FUNCTION_END;
    PetscFunctionReturn(0);
}

Here is the call graph for this function:

Here is the caller graph for this function: