mirror of
https://github.com/PixarAnimationStudios/OpenSubdiv
synced 2024-12-23 08:20:06 +00:00
a1c7be7c8e
This set of commits includes the addition of a new evaluation interface that treats a subdivision mesh more like a piecewise parametric surface primitive. The new interface was placed in namespace "Bfr" for "Base Face Representation" as all concepts and classes relate to a single face of the base mesh.
1294 lines
46 KiB
C++
1294 lines
46 KiB
C++
//
|
|
// Copyright 2021 Pixar
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "Apache License")
|
|
// with the following modification; you may not use this file except in
|
|
// compliance with the Apache License and the following modification to it:
|
|
// Section 6. Trademarks. is deleted and replaced with:
|
|
//
|
|
// 6. Trademarks. This License does not grant permission to use the trade
|
|
// names, trademarks, service marks, or product names of the Licensor
|
|
// and its affiliates, except as required to comply with Section 4(c) of
|
|
// the License and to reproduce the content of the NOTICE file.
|
|
//
|
|
// You may obtain a copy of the Apache License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the Apache License with the above modification is
|
|
// distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
|
|
// KIND, either express or implied. See the Apache License for the specific
|
|
// language governing permissions and limitations under the Apache License.
|
|
//
|
|
|
|
#include "../bfr/hash.h"
|
|
#include "../bfr/limits.h"
|
|
#include "../bfr/surface.h"
|
|
#include "../bfr/surfaceFactory.h"
|
|
#include "../bfr/surfaceFactoryCache.h"
|
|
#include "../bfr/faceTopology.h"
|
|
#include "../bfr/faceSurface.h"
|
|
#include "../bfr/regularPatchBuilder.h"
|
|
#include "../bfr/irregularPatchBuilder.h"
|
|
#include "../bfr/patchTree.h"
|
|
|
|
#include <map>
|
|
#include <cstdio>
|
|
|
|
namespace OpenSubdiv {
|
|
namespace OPENSUBDIV_VERSION {
|
|
|
|
namespace Bfr {
|
|
//
|
|
// DEBUG - static variables to keep track of constructed Surfaces
|
|
// - note that these global variables have extremely limited use:
|
|
// - they are initialized once per process
|
|
// - reported and reset on destruction of a SurfaceFactory
|
|
//
|
|
//#define _BFR_DEBUG_TOP_TYPE_STATS
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
static int __numLinearPatches = 0;
|
|
static int __numExpRegularPatches = 0;
|
|
static int __numRegularPatches = 0;
|
|
static int __numIrregularPatches = 0;
|
|
static int __numIrregularUncached = 0;
|
|
static int __numIrregularInCache = 0;
|
|
#endif
|
|
|
|
//
|
|
// Definition of the private/nested SurfaceSet class:
|
|
//
|
|
// This class (essentially a struct) encapsulates a clients specification
|
|
// of a set of multiple surfaces and their intended interpolation types
|
|
// (vertex, varying, and face-varying). The multiple public creation
|
|
// methods to request common subsets of surfaces all populate an instance
|
|
// of SurfaceSet for internal use.
|
|
//
|
|
class SurfaceFactory::SurfaceSet {
|
|
public:
|
|
SurfaceSet() : numSurfs(0), numFVarSurfs(0),
|
|
vtxSurf(0), varSurf(0),
|
|
fvarSurfs(0), fvarSurfPtrs(0), fvarIDs(0) { }
|
|
|
|
public:
|
|
// Assignment to member variable is intended to be explicit:
|
|
int numSurfs;
|
|
int numFVarSurfs;
|
|
|
|
typedef internal::SurfaceData SurfaceType;
|
|
|
|
SurfaceType * vtxSurf;
|
|
SurfaceType * varSurf;
|
|
SurfaceType * fvarSurfs;
|
|
SurfaceType ** fvarSurfPtrs;
|
|
|
|
FVarID const * fvarIDs;
|
|
|
|
void InitializeSurfaces() const {
|
|
if (vtxSurf) vtxSurf->reinitialize();
|
|
if (varSurf) varSurf->reinitialize();
|
|
for (int i = 0; i < numFVarSurfs; ++i) {
|
|
GetFVarSurface(i)->reinitialize();
|
|
}
|
|
}
|
|
|
|
public:
|
|
// Access to member variables is preferred through these methods,
|
|
// which may require a little more logic than expected:
|
|
int GetNumSurfaces() const { return numSurfs; }
|
|
|
|
bool HasVertexSurface() const { return (vtxSurf != 0); }
|
|
SurfaceType * GetVertexSurface() const { return vtxSurf; }
|
|
|
|
bool HasVaryingSurface() const { return (varSurf != 0); }
|
|
SurfaceType * GetVaryingSurface() const { return varSurf; }
|
|
|
|
// More than one FVar surface may be present, and each may have
|
|
// a unique ID:
|
|
bool HasFVarSurfaces() const { return numFVarSurfs > 0; }
|
|
int GetNumFVarSurfaces() const { return numFVarSurfs; }
|
|
|
|
FVarID GetFVarSurfaceID(int i) const {
|
|
return fvarIDs ? fvarIDs[i] : FVarID(i);
|
|
}
|
|
|
|
SurfaceType * GetFVarSurface(int i) const {
|
|
// Note that FVar Surfaces may be specified either as an
|
|
// array of Surfaces or an array of Surface pointers:
|
|
return fvarSurfs ? (fvarSurfs + i) : fvarSurfPtrs[i];
|
|
}
|
|
};
|
|
|
|
|
|
//
|
|
// Main constructor and supporting initialization methods:
|
|
//
|
|
SurfaceFactory::SurfaceFactory(Sdc::SchemeType subdivScheme,
|
|
Sdc::Options const & subdivOptions,
|
|
Options const & factoryOptions) :
|
|
_topologyCache(0) {
|
|
|
|
// Order of operations not important here:
|
|
setSubdivisionOptions(subdivScheme, subdivOptions);
|
|
setFactoryOptions(factoryOptions);
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::setSubdivisionOptions(Sdc::SchemeType subdivScheme,
|
|
Sdc::Options const & subdivOptions) {
|
|
|
|
// Assign the main member variables before others derived from them:
|
|
_subdivScheme = subdivScheme;
|
|
_subdivOptions = subdivOptions;
|
|
|
|
// Initialize members dependent on subdivision topology:
|
|
_regFaceSize = Sdc::SchemeTypeTraits::GetRegularFaceSize(_subdivScheme);
|
|
|
|
_linearScheme =
|
|
(Sdc::SchemeTypeTraits::GetLocalNeighborhoodSize(_subdivScheme) == 0);
|
|
|
|
_linearFVarInterp = _linearScheme ||
|
|
(_subdivOptions.GetFVarLinearInterpolation() ==
|
|
Sdc::Options::FVAR_LINEAR_ALL);
|
|
|
|
// Initialize members related to the "face has limit" test:
|
|
_rejectSmoothBoundariesForLimit = !_linearScheme &&
|
|
(_subdivOptions.GetVtxBoundaryInterpolation() ==
|
|
Sdc::Options::VTX_BOUNDARY_NONE);
|
|
|
|
_rejectIrregularFacesForLimit = !_linearScheme && (_regFaceSize == 3);
|
|
|
|
_testNeighborhoodForLimit = _rejectSmoothBoundariesForLimit ||
|
|
_rejectIrregularFacesForLimit;
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::setFactoryOptions(Options const & factoryOptions) {
|
|
|
|
// Assign the main member variable before others derived from them:
|
|
_factoryOptions = factoryOptions;
|
|
|
|
if (_factoryOptions.IsCachingEnabled()) {
|
|
if (_factoryOptions.GetExternalCache()) {
|
|
_topologyCache = _factoryOptions.GetExternalCache();
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::setInternalCache(SurfaceFactoryCache * cache) {
|
|
|
|
// Remember caching must be on and an external cache takes precedence
|
|
if (_factoryOptions.IsCachingEnabled()) {
|
|
if (_factoryOptions.GetExternalCache() == 0) {
|
|
_topologyCache = cache;
|
|
}
|
|
}
|
|
}
|
|
|
|
SurfaceFactory::~SurfaceFactory() {
|
|
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
// DEBUG - report and reset inventory:
|
|
printf("SurfaceFactory destructor:\n");
|
|
printf(" __numLinearPatches = %6d\n", __numLinearPatches);
|
|
printf(" __numExpRegularPatches = %6d\n", __numExpRegularPatches);
|
|
printf(" __numRegularPatches = %6d\n", __numRegularPatches);
|
|
printf(" __numIrregularPatches = %6d\n", __numIrregularPatches);
|
|
if (!_factoryOptions.DisableTopologyCache()) {
|
|
printf("\n");
|
|
printf(" __numIrregularUncached = %6d\n", __numIrregularUncached);
|
|
printf(" __numIrregularInCache = %6d\n", __numIrregularInCache);
|
|
}
|
|
__numLinearPatches = 0;
|
|
__numExpRegularPatches = 0;
|
|
__numRegularPatches = 0;
|
|
__numIrregularPatches = 0;
|
|
__numIrregularUncached = 0;
|
|
__numIrregularInCache = 0;
|
|
#endif
|
|
}
|
|
|
|
//
|
|
// Notes on presence/absence of a limit surface...
|
|
//
|
|
// Unfortunately it is not trivial to detect when a face does not have
|
|
// an associated limit surface. There are a few cases when a face will
|
|
// not have a limit surface -- divided into simple and complex cases:
|
|
//
|
|
// - simple:
|
|
// - the face is a hole
|
|
// - the face is degenerate (< 3 edges)
|
|
// - complex:
|
|
// - boundary interpolation option "none" is assigned:
|
|
// - in which case some, not all, boundary faces have no limit
|
|
// - Loop subdivision is applied to non-triangles
|
|
//
|
|
// The simple cases are, as the name suggests, simple. But the complex
|
|
// cases require a greater inspection of the topological neighborhood of
|
|
// the face.
|
|
//
|
|
// With boundary faces when "boundary none" is set (not very often) it is
|
|
// not enough to test if a face is a boundary -- if a boundary face has all
|
|
// of its incident boundary edges (i.e. all boundary edges incident to all
|
|
// of its face-vertices) then the boundary face has a limit surface. This
|
|
// requires a complete topological description of each corner of the face.
|
|
//
|
|
// Similarly, the case of Loop subdivision in the presence of non-triangles
|
|
// required determining if any corner of the face has an incident face
|
|
// that is not a triangle.
|
|
//
|
|
// The method here inspects a corner at a time and tries to reject a face
|
|
// without a limit surface as soon as possible. But most cases are going to
|
|
// require inspection of all corners -- and that same inspection is likely
|
|
// to be applied later when constructing the limit.
|
|
//
|
|
inline bool
|
|
SurfaceFactory::faceHasLimitSimple(Index faceIndex, int faceSize) const {
|
|
|
|
return (faceSize >= 3) && (faceSize <= Limits::MaxFaceSize()) &&
|
|
!isFaceHole(faceIndex);
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::faceHasLimitNeighborhood(FaceTopology const & topology) const {
|
|
|
|
assert(_testNeighborhoodForLimit);
|
|
|
|
MultiVertexTag tag = topology.GetTag();
|
|
|
|
if ((_rejectSmoothBoundariesForLimit && tag.HasNonSharpBoundary()) ||
|
|
(_rejectIrregularFacesForLimit && tag.HasIrregularFaceSizes())) {
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::faceHasLimitNeighborhood(Index faceIndex) const {
|
|
|
|
assert(_testNeighborhoodForLimit);
|
|
|
|
//
|
|
// The FaceTopology was not available, and rather than construct it
|
|
// in its entirety, determine a corner at a time and return if any
|
|
// corner warrants it:
|
|
//
|
|
typedef Vtr::internal::StackBuffer<Index,32,true> CornerIndexBuffer;
|
|
|
|
CornerIndexBuffer cFaceVertIndices;
|
|
|
|
FaceVertex faceVtx;
|
|
VertexDescriptor & vtxDesc = faceVtx.GetVertexDescriptor();
|
|
|
|
int faceSize = getFaceSize(faceIndex);
|
|
for (int i = 0; i < faceSize; ++i) {
|
|
// Have the subclass load VertexDescriptor and finalize:
|
|
faceVtx.Initialize(faceSize, _regFaceSize);
|
|
|
|
int faceInRing = populateFaceVertexDescriptor(faceIndex, i, &vtxDesc);
|
|
if (faceInRing < 0) return false;
|
|
|
|
faceVtx.Finalize(faceInRing);
|
|
|
|
// Inspect the tag to reject cases with no limit surface:
|
|
VertexTag faceVtxTag = faceVtx.GetTag();
|
|
|
|
if (_rejectSmoothBoundariesForLimit) {
|
|
if (faceVtxTag.IsUnOrdered()) {
|
|
// Need to load face-vertices, connect faces and inspect...
|
|
cFaceVertIndices.SetSize(faceVtx.GetNumFaceVertices());
|
|
|
|
if (getFaceVertexIncidentFaceVertexIndices(
|
|
faceIndex, i, cFaceVertIndices) < 0) return false;
|
|
|
|
faceVtx.ConnectUnOrderedFaces(cFaceVertIndices);
|
|
}
|
|
if (faceVtxTag.HasNonSharpBoundary()) return false;
|
|
}
|
|
if (_rejectIrregularFacesForLimit) {
|
|
if (faceVtxTag.HasIrregularFaceSizes()) return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::FaceHasLimitSurface(Index faceIndex) const {
|
|
|
|
if (!faceHasLimitSimple(faceIndex, getFaceSize(faceIndex))) {
|
|
return false;
|
|
}
|
|
if (_testNeighborhoodForLimit) {
|
|
if (!isFaceNeighborhoodRegular(faceIndex, 0, 0)) {
|
|
return faceHasLimitNeighborhood(faceIndex);
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
Parameterization
|
|
SurfaceFactory::GetFaceParameterization(Index faceIndex) const {
|
|
|
|
return Parameterization(_subdivScheme, getFaceSize(faceIndex));
|
|
}
|
|
|
|
|
|
//
|
|
// Namespace with internal utilities to compute keys for unique surface
|
|
// topologies to help cache their limit surface representations:
|
|
//
|
|
namespace {
|
|
// Need to redefine since SurfaceFactoryCache::Key is protected:
|
|
typedef std::uint64_t KeyIntType;
|
|
|
|
//
|
|
// Note that the data used in determining a topology key is not
|
|
// purely topological. While most data determines a unique limit
|
|
// surface, a few parameters determine the approximation to it
|
|
// (e.g. the various adaptive refinement levels) or dictate other
|
|
// properties of its representation (e.g. double precision).
|
|
//
|
|
// It may be worth separating these -- writing a method to deal
|
|
// with the pure topology first, then combining it with details
|
|
// of the representation.
|
|
//
|
|
|
|
#ifdef _OPENSUBDIV_BFR_INCLUDE_PACKED_TOPOLOGY_KEY_
|
|
//
|
|
// This alternate function for computing the cache key was applied
|
|
// to common topologies with low-valence and simply packs integer
|
|
// bitfields with the topology of the face and all of its corners.
|
|
//
|
|
// It is typically 3x faster than the hashing method, but is suited
|
|
// for general use. Given the relatively low cost of computing the
|
|
// cache key (compared to building an irregular patch) use of this
|
|
// faster function is not generally significant, but could become
|
|
// so in future when/if other costs are reduced.
|
|
//
|
|
// When the cost of building new patches is very low due to a high
|
|
// percentage of cache hits, the cost of computing the cache keys
|
|
// (relative to constructing the entire Surface) can rise to over
|
|
// 10% in extreme cases.
|
|
//
|
|
KeyIntType
|
|
packTopologyKey(FaceSurface const & surface,
|
|
IrregularPatchBuilder::Options options) {
|
|
|
|
//
|
|
// Keep the bitfield struct local in scope unless needed elsewhere:
|
|
//
|
|
struct KeyBits {
|
|
// Be sure to clear to avoid uninitialized bits:
|
|
KeyBits() { std::memset(this, 0, sizeof(*this)); }
|
|
|
|
// Bits for general options:
|
|
KeyIntType subdScheme : 2;
|
|
KeyIntType sharpLevel : 4;
|
|
KeyIntType smoothLevel : 4;
|
|
KeyIntType usesDouble : 1;
|
|
KeyIntType unused : 1; // future use
|
|
// Bits for the corners:
|
|
KeyIntType v0Valence : 6;
|
|
KeyIntType v1Valence : 6;
|
|
KeyIntType v2Valence : 6;
|
|
KeyIntType v3Valence : 6;
|
|
KeyIntType v0IsSharp : 1;
|
|
KeyIntType v1IsSharp : 1;
|
|
KeyIntType v2IsSharp : 1;
|
|
KeyIntType v3IsSharp : 1;
|
|
KeyIntType v0IsBoundary : 1;
|
|
KeyIntType v1IsBoundary : 1;
|
|
KeyIntType v2IsBoundary : 1;
|
|
KeyIntType v3IsBoundary : 1;
|
|
KeyIntType v0FaceInBoundary : 5;
|
|
KeyIntType v1FaceInBoundary : 5;
|
|
KeyIntType v2FaceInBoundary : 5;
|
|
KeyIntType v3FaceInBoundary : 5;
|
|
|
|
//
|
|
// Static methods to determine if bitfields can be used:
|
|
//
|
|
static bool InteriorValenceFits(int n) { return n < (1 << 6); }
|
|
static bool BoundaryValenceFits(int n) { return n < (1 << 5); }
|
|
|
|
// A place holder if future Sdc::Options inhibit use of bitfields:
|
|
static bool OptionsInhibitUsage(Sdc::Options) { return false; }
|
|
};
|
|
assert(sizeof(KeyBits) == sizeof(KeyIntType));
|
|
|
|
//
|
|
// Quickly test if the topology can be packed into bitfields, or
|
|
// if hashing must be used. Bitfields cannot be used when the
|
|
// following features are present:
|
|
//
|
|
// - any sharp edges of any kind (semi-sharp or inf-sharp)
|
|
// - any semi-sharp vertices (inf-sharp is 1-bit per corner)
|
|
// - any incident irregular faces
|
|
//
|
|
// These can quickly be determined by inspecting the topology tags.
|
|
// Two other situations are:
|
|
//
|
|
// - any vertex with valence too high (more than ~6 bits)
|
|
// - any Sdc::Option that cannot be encoded (in theory only)
|
|
//
|
|
// The former must inspect the valence of each face-vertex, while
|
|
// the latter requires inspecting the Sdc::Options -- and this is
|
|
// called out more as a future possibility...
|
|
//
|
|
// In theory, if certain Sdc::Options impact the limit surface,
|
|
// they might need to be encoded, or might not be able to be fully
|
|
// encoded in future. This is not currently the case in practice:
|
|
// boundary interpolation options are essentially unused as the
|
|
// boundary conditions are explicitly applied; the creasing method
|
|
// can be ignored here because creases cannot be packed; and the
|
|
// Catmark triangle subdivision option can be ignored because the
|
|
// presence of any irregular faces cannot be packed.
|
|
//
|
|
|
|
// Immediate rejection of bitfields:
|
|
MultiVertexTag combinedTag = surface.GetTag();
|
|
if (combinedTag.HasSharpEdges() ||
|
|
combinedTag.HasSemiSharpVertices() ||
|
|
combinedTag.HasIrregularFaceSizes()) {
|
|
return false;
|
|
}
|
|
|
|
// Conditional rejection of bitfields for high valence:
|
|
FaceVertexSubset const * subsets = surface.GetSubsets();
|
|
|
|
for (int i = 0; i < surface.GetFaceSize(); ++i) {
|
|
int valence = subsets[i]._numFacesTotal;
|
|
if (subsets[i].IsBoundary()) {
|
|
if (!KeyBits::BoundaryValenceFits(valence)) return false;
|
|
} else {
|
|
if (!KeyBits::InteriorValenceFits(valence)) return false;
|
|
}
|
|
}
|
|
|
|
// Conditional rejection of bitfields for specific Sdc::Options:
|
|
if (KeyBits::OptionsInhibitUsage(surface.GetSdcOptionsInEffect())) {
|
|
return false;
|
|
}
|
|
|
|
//
|
|
// Pack the topology of each FaceVertexSubset into bitfields:
|
|
//
|
|
KeyBits keyBits;
|
|
|
|
keyBits.subdScheme = surface.GetSdcScheme();
|
|
keyBits.sharpLevel = options.sharpLevel;
|
|
keyBits.smoothLevel = options.smoothLevel;
|
|
keyBits.usesDouble = options.doublePrecision;
|
|
|
|
keyBits.v0Valence = subsets[0]._numFacesTotal;
|
|
keyBits.v0IsSharp = subsets[0].IsSharp();
|
|
keyBits.v0IsBoundary = subsets[0].IsBoundary();
|
|
keyBits.v0FaceInBoundary = subsets[0]._numFacesBefore;
|
|
|
|
keyBits.v1Valence = subsets[1]._numFacesTotal;
|
|
keyBits.v1IsSharp = subsets[1].IsSharp();
|
|
keyBits.v1IsBoundary = subsets[1].IsBoundary();
|
|
keyBits.v1FaceInBoundary = subsets[1]._numFacesBefore;
|
|
|
|
keyBits.v2Valence = subsets[2]._numFacesTotal;
|
|
keyBits.v2IsSharp = subsets[2].IsSharp();
|
|
keyBits.v2IsBoundary = subsets[2].IsBoundary();
|
|
keyBits.v2FaceInBoundary = subsets[2]._numFacesBefore;
|
|
|
|
if (surface.GetFaceSize() == 4) {
|
|
keyBits.v3Valence = subsets[3]._numFacesTotal;
|
|
keyBits.v3IsSharp = subsets[3].IsSharp();
|
|
keyBits.v3IsBoundary = subsets[3].IsBoundary();
|
|
keyBits.v3FaceInBoundary = subsets[3]._numFacesBefore;
|
|
}
|
|
|
|
KeyIntType keyValue = 0;
|
|
std::memcpy(&keyValue, &keyBits, sizeof(KeyIntType));
|
|
return keyValue;
|
|
}
|
|
#endif
|
|
|
|
//
|
|
// Function to assign the topology of any FaceSurface to the desired
|
|
// integer using a hashing function that considers all topological
|
|
// features (incident face sizes, crease and corner sharpness, etc.):
|
|
//
|
|
KeyIntType
|
|
hashTopologyKey(FaceSurface const & surface,
|
|
IrregularPatchBuilder::Options options) {
|
|
|
|
//
|
|
// Structs for "headers" for the entire surface and each corner,
|
|
// to be assigned and copied into a larger buffer to be hashed:
|
|
//
|
|
typedef unsigned char uchar;
|
|
|
|
struct SurfaceHeader {
|
|
// Be sure to clear to avoid uninitialized bits:
|
|
SurfaceHeader() { std::memset(this, 0, sizeof(*this)); }
|
|
|
|
short faceSize;
|
|
uchar subdScheme;
|
|
uchar subdCreasing;
|
|
uchar subdTriSmooth;
|
|
uchar sharpLevel;
|
|
uchar smoothLevel;
|
|
uchar usesDouble;
|
|
};
|
|
struct CornerHeader {
|
|
// Be sure to clear to avoid uninitialized bits:
|
|
CornerHeader() { std::memset(this, 0, sizeof(*this)); }
|
|
|
|
short numFaces;
|
|
short faceInBoundary;
|
|
|
|
uchar isBoundary : 1;
|
|
uchar isInfSharp : 1;
|
|
uchar isSemiSharp : 1;
|
|
uchar hasFaceSizes : 1;
|
|
uchar hasSharpEdges : 1;
|
|
};
|
|
|
|
//
|
|
// Consider using a "delimiter" between corners in the buffer of
|
|
// data to be hashed, i.e. a value with a distinct bit pattern
|
|
// such as -1, to help prevent aliasing (may not be needed):
|
|
//
|
|
int delimiter = -1;
|
|
bool useCornerDelimiter = false;
|
|
|
|
//
|
|
// Local buffers for accumulating and hashing:
|
|
//
|
|
Vtr::internal::StackBuffer<float,16,true> floatBuffer;
|
|
Vtr::internal::StackBuffer<short,16,true> shortBuffer;
|
|
|
|
Vtr::internal::StackBuffer<char,256,true> hashBuffer;
|
|
|
|
//
|
|
// Determine size of the main buffer to hash ahead of time:
|
|
//
|
|
// Note there is some redundancy in the use of the uncommon
|
|
// faces sizes and sharp edges around each corner due to the
|
|
// way the corners' incident faces overlap. For typical cases
|
|
// the extra data used is not large. Only in extreme cases is
|
|
// it likely to be an issue -- but then the added processing
|
|
// and construction costs associated with such cases (e.g.
|
|
// high valence vertices, heavy use of creasing) will make
|
|
// make the overhead here insignificant.
|
|
//
|
|
size_t hashBufferSize = sizeof(SurfaceHeader);
|
|
|
|
int faceSize = surface.GetFaceSize();
|
|
for (int i = 0; i < faceSize; ++i) {
|
|
FaceVertexSubset const & cSub = surface.GetCornerSubset(i);
|
|
|
|
int N = cSub.GetNumFaces();
|
|
|
|
hashBufferSize += sizeof(CornerHeader);
|
|
|
|
hashBufferSize += cSub._tag.IsSemiSharp() ?
|
|
sizeof(float) : 0;
|
|
hashBufferSize += cSub._tag.HasUnCommonFaceSizes() ?
|
|
(sizeof(short) * N) : 0;
|
|
hashBufferSize += cSub._tag.HasSharpEdges() ?
|
|
(sizeof(float) * (N - cSub.IsBoundary())) : 0;
|
|
|
|
hashBufferSize += useCornerDelimiter ? sizeof(delimiter) : 0;
|
|
}
|
|
hashBuffer.SetSize((int)hashBufferSize);
|
|
|
|
//
|
|
// Start populating the buffer with the surface header:
|
|
//
|
|
Sdc::Options subdOptions = surface.GetSdcOptionsInEffect();
|
|
|
|
SurfaceHeader sHeader;
|
|
sHeader.faceSize = (short) faceSize;
|
|
sHeader.subdScheme = (uchar) surface.GetSdcScheme();
|
|
sHeader.subdCreasing = (uchar) subdOptions.GetCreasingMethod();
|
|
sHeader.subdTriSmooth = (uchar) subdOptions.GetTriangleSubdivision();
|
|
sHeader.sharpLevel = (uchar) options.sharpLevel;
|
|
sHeader.smoothLevel = (uchar) options.smoothLevel;
|
|
sHeader.usesDouble = (uchar) options.doublePrecision;
|
|
|
|
std::memcpy(hashBuffer, &sHeader, sizeof(sHeader));
|
|
|
|
//
|
|
// Populate the buffer for each corner of the surface:
|
|
//
|
|
char * bufferPtr = hashBuffer + sizeof(sHeader);
|
|
|
|
for (int corner = 0; corner < faceSize; ++corner) {
|
|
FaceVertex const & cTop = surface.GetCornerTopology(corner);
|
|
FaceVertexSubset const & cSub = surface.GetCornerSubset(corner);
|
|
|
|
// Assign the corner header:
|
|
CornerHeader cHeader;
|
|
cHeader.numFaces = (short) cSub.GetNumFaces();
|
|
cHeader.faceInBoundary = cSub._numFacesBefore;
|
|
cHeader.isBoundary = cSub.IsBoundary();
|
|
cHeader.isInfSharp = cSub.IsSharp();
|
|
cHeader.isSemiSharp = cSub._tag.IsSemiSharp();
|
|
cHeader.hasFaceSizes = cSub._tag.HasUnCommonFaceSizes();
|
|
cHeader.hasSharpEdges = cSub._tag.HasSharpEdges();
|
|
|
|
std::memcpy(bufferPtr, &cHeader, sizeof(cHeader));
|
|
bufferPtr += sizeof(cHeader);
|
|
|
|
if (cHeader.isSemiSharp) {
|
|
float sharpness = (cSub._localSharpness > 0.0f)
|
|
? cSub._localSharpness
|
|
: cTop.GetVertexSharpness();
|
|
std::memcpy(bufferPtr, &sharpness, sizeof(sharpness));
|
|
bufferPtr += sizeof(sharpness);
|
|
}
|
|
if (cHeader.hasFaceSizes) {
|
|
int n = cSub.GetNumFaces();
|
|
shortBuffer.SetSize(n);
|
|
for (int i = 0, f = cTop.GetFaceFirst(cSub); i < n;
|
|
f = cTop.GetFaceNext(f), ++i) {
|
|
shortBuffer[i] = (short) cTop.GetFaceSize(f);
|
|
}
|
|
std::memcpy(bufferPtr, shortBuffer, n * sizeof(short));
|
|
bufferPtr += n * sizeof(short);
|
|
}
|
|
if (cHeader.hasSharpEdges) {
|
|
int n = cSub.GetNumFaces() - cSub.IsBoundary();
|
|
floatBuffer.SetSize(n);
|
|
for (int i = 0, f = cTop.GetFaceFirst(cSub); i < n;
|
|
f = cTop.GetFaceNext(f), ++i) {
|
|
floatBuffer[i] = cTop.GetFaceEdgeSharpness(f, 1);
|
|
}
|
|
std::memcpy(bufferPtr, floatBuffer, n * sizeof(float));
|
|
bufferPtr += n * sizeof(float);
|
|
}
|
|
|
|
if (useCornerDelimiter) {
|
|
std::memcpy(bufferPtr, &delimiter, sizeof(delimiter));
|
|
bufferPtr += sizeof(delimiter);
|
|
}
|
|
}
|
|
assert((bufferPtr - hashBuffer) == (int)hashBufferSize);
|
|
|
|
return internal::Hash64(hashBuffer, hashBufferSize);
|
|
}
|
|
}
|
|
|
|
//
|
|
// Methods supporting construction of linear, regular and irregular patches:
|
|
//
|
|
void
|
|
SurfaceFactory::assignLinearSurface(SurfaceType * surfacePtr,
|
|
Index faceIndex, FVarID const * fvarPtrOrVtx) const {
|
|
|
|
SurfaceType & surface = *surfacePtr;
|
|
|
|
// Initialize instance members from the associated irregular patch:
|
|
int faceSize = getFaceSize(faceIndex);
|
|
|
|
surface.setParam(Parameterization(_subdivScheme, faceSize));
|
|
|
|
surface.setRegular(faceSize == _regFaceSize);
|
|
surface.setLinear(true);
|
|
|
|
surface.setRegPatchMask(0);
|
|
if (_regFaceSize == 4) {
|
|
surface.setRegPatchType(Far::PatchDescriptor::QUADS);
|
|
} else {
|
|
surface.setRegPatchType(Far::PatchDescriptor::TRIANGLES);
|
|
}
|
|
|
|
//
|
|
// Finally, gather patch control points from the appropriate indices:
|
|
//
|
|
Index * surfaceCVs = surface.resizeCVs(faceSize);
|
|
|
|
int count = 0;
|
|
if (fvarPtrOrVtx == 0) {
|
|
count = getFaceVertexIndices(faceIndex, surfaceCVs);
|
|
} else {
|
|
count = getFaceFVarValueIndices(faceIndex, *fvarPtrOrVtx, surfaceCVs);
|
|
}
|
|
// If subclass fails to get indices, Surface will remain invalid
|
|
if (count < faceSize) return;
|
|
|
|
surface.setValid(true);
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numLinearPatches ++;
|
|
#endif
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::assignRegularSurface(SurfaceType * surfacePtr,
|
|
Index const patchPoints[]) const {
|
|
|
|
SurfaceType & surface = *surfacePtr;
|
|
|
|
//
|
|
// Assign the parameterization and discriminants first:
|
|
//
|
|
surface.setParam(Parameterization(_subdivScheme, _regFaceSize));
|
|
|
|
surface.setRegular(true);
|
|
surface.setLinear(false);
|
|
|
|
//
|
|
// Assemble the regular patch:
|
|
//
|
|
surface.setRegPatchType(RegularPatchBuilder::GetPatchType(_regFaceSize));
|
|
surface.setRegPatchMask(RegularPatchBuilder::GetBoundaryMask(_regFaceSize,
|
|
patchPoints));
|
|
|
|
//
|
|
// Copy the patch control points from the given indices:
|
|
//
|
|
int patchSize = RegularPatchBuilder::GetPatchSize(_regFaceSize);
|
|
|
|
Index const * pSrc = patchPoints;
|
|
Index * pDst = surface.resizeCVs(patchSize);
|
|
|
|
// Remember to replace negative indices in boundary patches:
|
|
if (surface.getRegPatchMask() == 0) {
|
|
std::memcpy(pDst, pSrc, patchSize * sizeof(Index));
|
|
} else {
|
|
// Consider delegating this task to the RegularPatchBuilder:
|
|
Index pPhantom = pSrc[5];
|
|
assert(pPhantom >= 0);
|
|
for (int i = 0; i < patchSize; ++i) {
|
|
pDst[i] = (pSrc[i] < 0) ? pPhantom : pSrc[i];
|
|
}
|
|
}
|
|
|
|
surface.setValid(true);
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numExpRegularPatches ++;
|
|
#endif
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::assignRegularSurface(SurfaceType * surfacePtr,
|
|
FaceSurface const & descriptor) const {
|
|
|
|
SurfaceType & surface = *surfacePtr;
|
|
|
|
//
|
|
// Assign the parameterization and discriminants first:
|
|
//
|
|
surface.setParam(Parameterization(_subdivScheme, _regFaceSize));
|
|
|
|
surface.setRegular(true);
|
|
surface.setLinear(false);
|
|
|
|
//
|
|
// Assemble the regular patch:
|
|
//
|
|
RegularPatchBuilder builder(descriptor);
|
|
|
|
surface.setRegPatchType(builder.GetPatchType());
|
|
surface.setRegPatchMask(builder.GetPatchParamBoundaryMask());
|
|
|
|
//
|
|
// Gather the patch control points from the given indices:
|
|
//
|
|
builder.GatherControlVertexIndices(
|
|
surface.resizeCVs(builder.GetNumControlVertices()));
|
|
|
|
surface.setValid(true);
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numRegularPatches ++;
|
|
#endif
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::assignIrregularSurface(SurfaceType * surfacePtr,
|
|
FaceSurface const & descriptor) const {
|
|
|
|
//
|
|
// A builder for the irregular patch is required regardless of
|
|
// whether a new instance is constructed:
|
|
//
|
|
IrregularPatchBuilder::Options buildOptions;
|
|
|
|
buildOptions.sharpLevel = _factoryOptions.GetApproxLevelSharp();
|
|
buildOptions.smoothLevel = _factoryOptions.GetApproxLevelSmooth();
|
|
buildOptions.doublePrecision = surfacePtr->isDouble();
|
|
|
|
IrregularPatchBuilder builder(descriptor, buildOptions);
|
|
|
|
//
|
|
// Construct a new irregular patch or identify one from the cache:
|
|
//
|
|
internal::IrregularPatchSharedPtr patch(0);
|
|
|
|
if (_topologyCache == 0) {
|
|
patch = builder.Build();
|
|
} else {
|
|
//
|
|
// Compute the cache key for the topology of this face, search the
|
|
// cache for an existing patch and build/add one if not found:
|
|
//
|
|
// Be sure to use the return result of Add() when adding as it may
|
|
// be the case that another thread added a patch with the same key
|
|
// while this one was being built. Using the instance assigned to
|
|
// the cache intentionally releases the one built here.
|
|
//
|
|
SurfaceFactoryCache::KeyType key =
|
|
hashTopologyKey(descriptor, buildOptions);
|
|
|
|
patch = _topologyCache->Find(key);
|
|
if (patch == 0) {
|
|
patch = _topologyCache->Add(key, builder.Build());
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numIrregularInCache ++;
|
|
#endif
|
|
}
|
|
}
|
|
|
|
//
|
|
// Assign the Surface parameterization, discriminants and patch:
|
|
//
|
|
SurfaceType & surface = *surfacePtr;
|
|
|
|
surface.setParam(Parameterization(_subdivScheme, descriptor.GetFaceSize()));
|
|
|
|
surface.setRegular(false);
|
|
surface.setLinear(false);
|
|
|
|
surface.setIrregPatchPtr(patch);
|
|
|
|
// Gather the patch control points from the given indices:
|
|
builder.GatherControlVertexIndices(
|
|
surface.resizeCVs(patch->GetNumControlPoints()));
|
|
|
|
surface.setValid(true);
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numIrregularPatches ++;
|
|
__numIrregularUncached += surface.ownsIrregPatch();
|
|
#endif
|
|
}
|
|
|
|
void
|
|
SurfaceFactory::copyNonLinearSurface(
|
|
SurfaceType * surfaceDstPtr,
|
|
SurfaceType const & surfaceSrc,
|
|
FaceSurface const & descriptor) const {
|
|
|
|
SurfaceType & surfaceDst = *surfaceDstPtr;
|
|
|
|
// Should be creating a linear patch directly rather than copying:
|
|
assert(!surfaceSrc.isLinear());
|
|
|
|
//
|
|
// Assign the topological fields of the patch first:
|
|
//
|
|
surfaceDst.setParam(surfaceSrc.getParam());
|
|
|
|
surfaceDst.setLinear(surfaceSrc.isLinear());
|
|
surfaceDst.setRegular(surfaceSrc.isRegular());
|
|
|
|
surfaceDst.resizeCVs(surfaceSrc.getNumCVs());
|
|
|
|
//
|
|
// Assign regular/irregular fields and gather control points:
|
|
//
|
|
if (surfaceDst.isRegular()) {
|
|
surfaceDst.setRegPatchType(surfaceSrc.getRegPatchType());
|
|
surfaceDst.setRegPatchMask(surfaceSrc.getRegPatchMask());
|
|
|
|
RegularPatchBuilder builder(descriptor);
|
|
assert(builder.GetNumControlVertices() == surfaceDst.getNumCVs());
|
|
|
|
builder.GatherControlVertexIndices(surfaceDst.getCVIndices());
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numRegularPatches ++;
|
|
#endif
|
|
} else {
|
|
surfaceDst.setIrregPatchPtr(surfaceSrc.getIrregPatchPtr());
|
|
|
|
IrregularPatchBuilder builder(descriptor);
|
|
assert(builder.GetNumControlVertices() == surfaceDst.getNumCVs());
|
|
|
|
builder.GatherControlVertexIndices(surfaceDst.getCVIndices());
|
|
#ifdef _BFR_DEBUG_TOP_TYPE_STATS
|
|
__numIrregularPatches ++;
|
|
#endif
|
|
}
|
|
|
|
surfaceDst.setValid(true);
|
|
}
|
|
|
|
|
|
//
|
|
// Methods to deal with topology assembly and inspection:
|
|
//
|
|
// Note the difference between the "init" and "gather" methods: "init"
|
|
// fully resolves and initializes the topology by gathering face indices
|
|
// locally and dealing with unordered faces if present, while the "gather"
|
|
// method simply gathers the corner information -- allowing indices to be
|
|
// provided for further use if needed.
|
|
//
|
|
bool
|
|
SurfaceFactory::initFaceNeighborhoodTopology(Index faceIndex,
|
|
FaceTopology * faceTopologyPtr) const {
|
|
|
|
FaceTopology & topology = *faceTopologyPtr;
|
|
|
|
if (!gatherFaceNeighborhoodTopology(faceIndex, &topology)) {
|
|
return false;
|
|
}
|
|
if (!topology.HasUnOrderedCorners()) {
|
|
return true;
|
|
}
|
|
|
|
// Gather the indices to determine topology between unordered faces:
|
|
typedef Vtr::internal::StackBuffer<Index,72,true> IndexBuffer;
|
|
|
|
IndexBuffer indices(topology._numFaceVertsTotal);
|
|
if (gatherFaceNeighborhoodIndices(faceIndex, topology, 0, indices) < 0) {
|
|
return false;
|
|
}
|
|
topology.ResolveUnOrderedCorners(indices);
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::gatherFaceNeighborhoodTopology(Index faceIndex,
|
|
FaceTopology * faceTopologyPtr) const {
|
|
|
|
FaceTopology & faceTopology = *faceTopologyPtr;
|
|
|
|
int N = getFaceSize(faceIndex);
|
|
|
|
faceTopology.Initialize(N);
|
|
|
|
for (int i = 0; i < N; ++i) {
|
|
FaceVertex & faceVtx = faceTopology.GetTopology(i);
|
|
VertexDescriptor & vtxDesc = faceVtx.GetVertexDescriptor();
|
|
|
|
faceVtx.Initialize(N, _regFaceSize);
|
|
|
|
// Subclass returning negative here indicates unsupported features
|
|
// or some other kind of failure:
|
|
int faceInRing = populateFaceVertexDescriptor(faceIndex, i, &vtxDesc);
|
|
if (faceInRing < 0) return false;
|
|
|
|
faceVtx.Finalize(faceInRing);
|
|
}
|
|
|
|
faceTopology.Finalize();
|
|
|
|
return true;
|
|
}
|
|
|
|
int
|
|
SurfaceFactory::gatherFaceNeighborhoodIndices(Index faceIndex,
|
|
FaceTopology const & faceTopology,
|
|
FVarID const * fvarPtrOrVtx,
|
|
Index controlIndices[]) const {
|
|
|
|
int faceSize = faceTopology.GetFaceSize();
|
|
|
|
Index * indices = controlIndices;
|
|
int nIndices = 0;
|
|
|
|
for (int i = 0; i < faceSize; ++i) {
|
|
int numFaceVerts = (fvarPtrOrVtx == 0) ?
|
|
getFaceVertexIncidentFaceVertexIndices(faceIndex, i,
|
|
indices) :
|
|
getFaceVertexIncidentFaceFVarValueIndices(faceIndex, i,
|
|
*fvarPtrOrVtx, indices);
|
|
|
|
if (numFaceVerts != faceTopology.GetNumFaceVertices(i)) {
|
|
return -1;
|
|
}
|
|
|
|
indices += numFaceVerts;
|
|
nIndices += numFaceVerts;
|
|
}
|
|
return nIndices;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::isFaceNeighborhoodRegular(Index faceIndex,
|
|
FVarID const * fvarPtrOrVtx,
|
|
Index indices[]) const {
|
|
return (fvarPtrOrVtx == 0) ?
|
|
getFaceNeighborhoodVertexIndicesIfRegular(faceIndex, indices) :
|
|
getFaceNeighborhoodFVarValueIndicesIfRegular(faceIndex, *fvarPtrOrVtx,
|
|
indices);
|
|
}
|
|
|
|
//
|
|
// Main internal methods to populate set of limit Surfaces:
|
|
//
|
|
bool
|
|
SurfaceFactory::populateAllSurfaces(Index faceIndex,
|
|
SurfaceSet * surfaceSetPtr) const {
|
|
|
|
SurfaceSet & surfaces = *surfaceSetPtr;
|
|
|
|
// Abort if no Surfaces are specified to populate:
|
|
if (surfaces.GetNumSurfaces() == 0) {
|
|
return false;
|
|
}
|
|
|
|
//
|
|
// Be sure to re-initialize all Surfaces up-front, rather than
|
|
// deferring it to the assignment of each. A failure of any one
|
|
// surface may leave others unvisited -- leaving it unchanged
|
|
// from previous use.
|
|
//
|
|
surfaces.InitializeSurfaces();
|
|
|
|
// Quickly reject faces with no limit (typically holes) -- some cases
|
|
// require full topological inspection and will be rejected later:
|
|
if (!faceHasLimitSimple(faceIndex, getFaceSize(faceIndex))) {
|
|
return false;
|
|
}
|
|
|
|
// Determine if we have any non-linear cases to deal with -- which
|
|
// require gathering and inspection of the full neighborhood around
|
|
// the given face:
|
|
int numFVarSurfaces = surfaces.GetNumFVarSurfaces();
|
|
|
|
bool hasNonLinearSurfaces =
|
|
(surfaces.HasVertexSurface() && !_linearScheme) ||
|
|
(numFVarSurfaces && !_linearFVarInterp);
|
|
|
|
bool hasLinearSurfaces =
|
|
surfaces.HasVaryingSurface() ||
|
|
(surfaces.HasVertexSurface() && _linearScheme) ||
|
|
(numFVarSurfaces && _linearFVarInterp);
|
|
|
|
if (hasNonLinearSurfaces || _testNeighborhoodForLimit) {
|
|
if (!populateNonLinearSurfaces(faceIndex, &surfaces)) {
|
|
return false;
|
|
}
|
|
}
|
|
if (hasLinearSurfaces) {
|
|
if (!populateLinearSurfaces(faceIndex, &surfaces)) {
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::populateLinearSurfaces(Index faceIndex,
|
|
SurfaceSet * surfaceSetPtr) const {
|
|
|
|
SurfaceSet & surfaces = *surfaceSetPtr;
|
|
|
|
if (surfaces.HasVaryingSurface()) {
|
|
assignLinearSurface(surfaces.GetVaryingSurface(), faceIndex, 0);
|
|
}
|
|
|
|
if (_linearScheme && surfaces.HasVertexSurface()) {
|
|
assignLinearSurface(surfaces.GetVertexSurface(), faceIndex, 0);
|
|
}
|
|
|
|
if (_linearFVarInterp) {
|
|
int numFVarSurfaces = surfaces.GetNumFVarSurfaces();
|
|
for (int i = 0; i < numFVarSurfaces; ++i) {
|
|
FVarID fvarID = surfaces.GetFVarSurfaceID(i);
|
|
assignLinearSurface(surfaces.GetFVarSurface(i), faceIndex, &fvarID);
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
SurfaceFactory::populateNonLinearSurfaces(Index faceIndex,
|
|
SurfaceSet * surfaceSetPtr) const {
|
|
|
|
SurfaceSet & surfaces = *surfaceSetPtr;
|
|
|
|
typedef Vtr::internal::StackBuffer<Index,72,true> IndexBuffer;
|
|
|
|
bool vtxIsNonLinear = surfaces.HasVertexSurface() && !_linearScheme;
|
|
bool fvarIsNonLinear = surfaces.HasFVarSurfaces() && !_linearFVarInterp;
|
|
bool anyNonLinear = vtxIsNonLinear || fvarIsNonLinear;
|
|
|
|
//
|
|
// First need to determine the vertex topology of the face and take
|
|
// appropriate action based on inputs. It may be the case that the
|
|
// topology is only used to determine if the non-linear face has a
|
|
// limit surface and no non-linear surfaces are generated here (and
|
|
// linear varying or face-varying surfaces are determined elsewhere).
|
|
//
|
|
// So determine the topology and deal with any required tests for
|
|
// the presence of limit surface.
|
|
//
|
|
// If the face is "explicitly regular", i.e. the subclass can provide
|
|
// an immediate regular patch representation, the more tedious work
|
|
// to assemble the more general topological representation is avoided.
|
|
//
|
|
// Note that while the vertex surface may be explicitly regular, if
|
|
// the face-varying topology does not match, i.e. there is a UV seam
|
|
// present around the face, the more general topological representation
|
|
// will be necessary to deal with a potentially irregular face-varying
|
|
// surface.
|
|
//
|
|
FaceTopology faceTopology(_subdivScheme, _subdivOptions);
|
|
IndexBuffer vtxIndices(16);
|
|
FaceSurface vtxSurfDesc;
|
|
|
|
bool vtxIsExplicitlyRegular =
|
|
isFaceNeighborhoodRegular(faceIndex, 0, vtxIndices);
|
|
if (vtxIsExplicitlyRegular) {
|
|
if (_testNeighborhoodForLimit && !anyNonLinear) {
|
|
return true;
|
|
}
|
|
} else {
|
|
//
|
|
// Three steps are required to get full topological description:
|
|
// - gathering the full description of the neighborhood
|
|
// - gathering vertex indices for the neighborhood
|
|
// - using the indices to resolve any unordered topology
|
|
// Gathering indices for the vertex surface and/or to resolve
|
|
// unordered topology is conditional.
|
|
//
|
|
if (!gatherFaceNeighborhoodTopology(faceIndex, &faceTopology)) {
|
|
return false;
|
|
}
|
|
if (vtxIsNonLinear || faceTopology.HasUnOrderedCorners()) {
|
|
vtxIndices.SetSize(faceTopology._numFaceVertsTotal);
|
|
if (gatherFaceNeighborhoodIndices(faceIndex, faceTopology, 0,
|
|
vtxIndices) < 0) {
|
|
return false;
|
|
}
|
|
if (faceTopology.HasUnOrderedCorners()) {
|
|
faceTopology.ResolveUnOrderedCorners(vtxIndices);
|
|
}
|
|
}
|
|
if (_testNeighborhoodForLimit) {
|
|
if (!faceHasLimitNeighborhood(faceTopology)) {
|
|
return false;
|
|
} else if (!anyNonLinear) {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
// Initialize the vertex surface descriptor for use creating both
|
|
// the vertex Surface and any non-linear FVar Surfaces:
|
|
vtxSurfDesc.Initialize(faceTopology, vtxIndices);
|
|
}
|
|
|
|
//
|
|
// Construct the Surface for vertex topology first, as face-varying
|
|
// surfaces that match topology may make use of it:
|
|
//
|
|
bool vtxSurfIsValid = false;
|
|
if (vtxIsNonLinear) {
|
|
SurfaceType & vtxSurf = *surfaces.GetVertexSurface();
|
|
|
|
if (vtxIsExplicitlyRegular) {
|
|
assignRegularSurface(&vtxSurf, vtxIndices);
|
|
} else if (vtxSurfDesc.IsRegular()) {
|
|
assignRegularSurface(&vtxSurf, vtxSurfDesc);
|
|
} else {
|
|
assignIrregularSurface(&vtxSurf, vtxSurfDesc);
|
|
}
|
|
vtxSurfIsValid = vtxSurf.isValid();
|
|
}
|
|
|
|
//
|
|
// Construct the Surface for the given face-varying topologies --
|
|
// all of which are potentially distinct.
|
|
//
|
|
// If the vertex topology is explicitly regular, the face-varying
|
|
// surface can only make use of it if it shares the same topology
|
|
// and the subclass provides corresponding control points.
|
|
//
|
|
// In all other cases the full topological description and the full
|
|
// description of the vertex surface must be provided. The set of
|
|
// face-varying indices must then be gathered and used to create a
|
|
// face-varying surface descriptor, which uses the indices to find
|
|
// the relevant face-varying subsets for each corner.
|
|
//
|
|
if (fvarIsNonLinear) {
|
|
// We can re-use the vertex index buffer for face-varying indices:
|
|
IndexBuffer & fvIndices = vtxIndices;
|
|
|
|
int numFVarSurfaces = surfaces.GetNumFVarSurfaces();
|
|
for (int i = 0; i < numFVarSurfaces; ++i) {
|
|
SurfaceType & fvarSurf = *surfaces.GetFVarSurface(i);
|
|
FVarID fvarID = surfaces.GetFVarSurfaceID(i);
|
|
|
|
// First check if trivially regular, quickly assign and continue:
|
|
bool fvarIsExplicitlyRegular = vtxIsExplicitlyRegular &&
|
|
isFaceNeighborhoodRegular(faceIndex, &fvarID, fvIndices);
|
|
|
|
if (fvarIsExplicitlyRegular) {
|
|
assignRegularSurface(&fvarSurf, fvIndices);
|
|
continue;
|
|
}
|
|
|
|
// Make sure topology, indices and vertex surface are initialized
|
|
// (will not be if vertex surface was explicitly regular):
|
|
if (!vtxSurfDesc.IsInitialized()) {
|
|
if (!initFaceNeighborhoodTopology(faceIndex, &faceTopology)) {
|
|
return false;
|
|
}
|
|
vtxSurfDesc.Initialize(faceTopology, 0);
|
|
}
|
|
fvIndices.SetSize(faceTopology._numFaceVertsTotal);
|
|
|
|
// Gather FVar indices and initialize FVar surface descriptor:
|
|
if (gatherFaceNeighborhoodIndices(faceIndex, faceTopology,
|
|
&fvarID, fvIndices) < 0) {
|
|
return false;
|
|
}
|
|
|
|
FaceSurface fvarSurfDesc(vtxSurfDesc, fvIndices);
|
|
|
|
// Detect matching or other topology and dispatch accordingly:
|
|
if (fvarSurfDesc.FVarTopologyMatchesVertex() && vtxSurfIsValid) {
|
|
copyNonLinearSurface(&fvarSurf, *surfaces.GetVertexSurface(),
|
|
fvarSurfDesc);
|
|
} else if (fvarSurfDesc.IsRegular()) {
|
|
assignRegularSurface(&fvarSurf, fvarSurfDesc);
|
|
} else {
|
|
assignIrregularSurface(&fvarSurf, fvarSurfDesc);
|
|
}
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
//
|
|
// Main internal method to initialize instances of Surface:
|
|
//
|
|
bool
|
|
SurfaceFactory::initSurfaces(Index faceIndex,
|
|
internal::SurfaceData * vtxSurface,
|
|
internal::SurfaceData * varSurface,
|
|
internal::SurfaceData * fvarSurfaces,
|
|
int fvarCount,
|
|
FVarID const fvarIDs[]) const {
|
|
|
|
// Note the SurfaceData for the first FVar Surface is assumed below
|
|
// to be the head of an array of SurfaceData, which will not be true
|
|
// if additional members are added to Surface<REAL> in future:
|
|
assert(sizeof(internal::SurfaceData) == sizeof(Surface<float>));
|
|
|
|
SurfaceSet surfaces;
|
|
|
|
surfaces.vtxSurf = vtxSurface;
|
|
surfaces.varSurf = varSurface;
|
|
surfaces.fvarSurfs = fvarSurfaces;
|
|
surfaces.fvarIDs = fvarIDs;
|
|
surfaces.numFVarSurfs = fvarCount;
|
|
surfaces.numSurfs = fvarCount + (vtxSurface != 0) + (varSurface != 0);
|
|
|
|
return populateAllSurfaces(faceIndex, &surfaces);
|
|
}
|
|
|
|
} // end namespace Bfr
|
|
|
|
} // end namespace OPENSUBDIV_VERSION
|
|
} // end namespace OpenSubdiv
|