Using Hbr

Using Hbr


As of OpenSubdiv 3.0, all Hbr dependencies have been removed from the core APIs (Sdc, Vtr, Far, Osd). The legacy source code of Hbr is provided purely for regression and legacy purposes. If your code is currently depending on Hbr functionality, we recommend migrating to the newer APIs as we cannot guarantee that this code will be maintained in future releases. For more information see the 3.0 release notes

Vertex Template API

The Hbr API abstracts the vertex class through templating. Client-code is expected to provide a vertex class that implements the requisite interpolation functionality.

Here is an example of a simple vertex class that accounts for 3D position, but does not support arbitrary variables or varying interpolation.

struct Vertex {

    Vertex() { }

    Vertex( int /*i*/ ) { }

    Vertex( const Vertex & src ) { _pos[0]=src._pos[0]; _pos[1]=src._pos[1]; _pos[2]=src._pos[2]; }

   ~Vertex( ) { }

    void AddWithWeight(const Vertex& src, float weight, void * =0 ) {

    void AddVaryingWithWeight(const Vertex& , float, void * =0 ) { }

    void Clear( void * =0 ) { _pos[0]=_pos[1]=_pos[2]=0.0f; }

    void SetPosition(float x, float y, float z) { _pos[0]=x; _pos[1]=y; _pos[2]=z; }

    void ApplyVertexEdit(const OpenSubdiv::HbrVertexEdit<Vertex> & edit) {
        const float *src = edit.GetEdit();
        switch(edit.GetOperation()) {
          case OpenSubdiv::HbrHierarchicalEdit<Vertex>::Set:
            _pos[0] = src[0];
            _pos[1] = src[1];
            _pos[2] = src[2];
          case OpenSubdiv::HbrHierarchicalEdit<Vertex>::Add:
            _pos[0] += src[0];
            _pos[1] += src[1];
            _pos[2] += src[2];
          case OpenSubdiv::HbrHierarchicalEdit<Vertex>::Subtract:
            _pos[0] -= src[0];
            _pos[1] -= src[1];
            _pos[2] -= src[2];

    void ApplyMovingVertexEdit(const OpenSubdiv::HbrMovingVertexEdit<Vertex> &) { }

    // custom functions & data not required by Hbr -------------------------

    Vertex( float x, float y, float z ) { _pos[0]=x; _pos[1]=y; _pos[2]=z; }

    const float * GetPos() const { return _pos; }

    float _pos[3];

In some cases, if only topological analysis is required, the class can be left un-implemented. Far and Osd for instance store vertex data in serialized interleaved vectors. Here is the Osd::Vertex class for reference:

class Vertex {
    Vertex() {}

    Vertex(int /* index */) {}

    Vertex(Vertex const & /* src */) {}

    void AddWithWeight(Vertex const & /* i */, float /* weight */, void * = 0) {}

    void AddVaryingWithWeight(const Vertex & /* i */, float /* weight */, void * = 0) {}

    void Clear(void * = 0) {}

    void ApplyVertexEdit(FarVertexEdit const &) { }

Creating a Mesh

The following tutorial walks through the steps of instantiating a simple Hbr mesh.

The code found in regression/common/shape_utils.h can also be used as an example. While this implementation covers many of Hbr's features, it does not provide coverage for the complete Renderman specification though.

Instantiating an HbrMesh

First we need to instantiate a mesh object.

Hbr supports 3 subdivision schemes:
  • Catmull-Clark (catmark)
  • Loop
  • Bilinear

The scheme is selected by passing an specialized instance of HbrSubdivision<T>, HbrCatmarkSubdivision<T> in this case. The scheme can be shared across multiple mesh objects, so we only need a single instance.

static OpenSubdiv::HbrCatmarkSubdivision<Vertex> _scheme;

OpenSubdiv::HbrMesh<Vertex> * mesh = new OpenSubdiv::HbrMesh<Vertex>( _scheme );

Creating Vertices

Adding vertices to the mesh is accomplished using the HbrMesh::NewVertex() method.

Because Hbr uses a dedicated vertex allocator to help alleviate the performance impact of intensive fragmented memory allocations. This optimization results in the following design pattern:

Vertex vtx;
for(int i=0;i<numVerts; i++ ) {
    Vertex * v = mesh->NewVertex( i, vtx);

    // v->SetPosition();

We instantiate a single "default" vertex object named 'vtx' on the stack. We then recover the pointer to the actual vertex created in the mesh from the NewVertex() method. Once we have recovered that pointer, we can set the data for our vertex by using any of the custom accessors.

Creating Faces

Once all the vertices have been registered in the mesh, we can start adding the faces with HbrMesh::NewFace(). Assuming we had an obj style reader, we need to know the number of vertices in the face and the indices of these vertices.

for (int f=0; f<numFaces; ++f) {

    int nverts = obj->GetNumVertices(f);

    const int * faceverts = obj->GetFaceVerts(f);

    mesh->NewFace(nv, fv, 0);

However, Hbr is not able to handle non-manifold geometry. In order to avoid tripping asserts or causing memory violations, let's rewrite the previous loop with some some prototype code to check the validity of the topology.

for (int f=0; f<numFaces; ++f) {

    int nv = obj->GetNumVertices(f);

    const int * fv = obj->GetFaceVerts(f);

    // triangles only for Loop subdivision !
    if ((scheme==kLoop) and (nv!=3)) {
        printf("Trying to create a Loop subd with non-triangle face\n");

    // now check the half-edges connectivity
    for(int j=0;j<nv;j++) {
        OpenSubdiv::HbrVertex<T> * origin      = mesh->GetVertex( fv[j] );
        OpenSubdiv::HbrVertex<T> * destination = mesh->GetVertex( fv[(j+1)%nv] );
        OpenSubdiv::HbrHalfedge<T> * opposite  = destination->GetEdge(origin);

        if(origin==NULL || destination==NULL) {
            printf(" An edge was specified that connected a nonexistent vertex\n");

        if(origin == destination) {
            printf(" An edge was specified that connected a vertex to itself\n");

        if(opposite && opposite->GetOpposite() ) {
            printf(" A non-manifold edge incident to more than 2 faces was found\n");

        if(origin->GetEdge(destination)) {
            printf(" An edge connecting two vertices was specified more than once."
                   " It's likely that an incident face was flipped\n");

    mesh->NewFace(nv, fv, 0);

Wrapping Things Up

Once we have vertices and faces set in our mesh, we still need to wrap things up by calling HbrMesh::Finish():


Finish iterates over the mesh to apply the boundary interpolation rules and checks for singular vertices. At this point, there is one final topology check remaining to validate the mesh:


if (mesh->GetNumDisconnectedVertices()) {
    printf("The specified subdivmesh contains disconnected surface components.\n");

    // abort or iterate over the mesh to remove the offending vertices

Boundary Interpolation Rules

The rule-set can be selected using the following accessors:

Vertex and varying data:

mesh->SetInterpolateBoundaryMethod( OpenSubdiv::HbrMesh<Vertex>::k_InterpolateBoundaryEdgeOnly );

Face-varying data:

mesh->SetFVarInterpolateBoundaryMethod( OpenSubdiv::HbrMesh<Vertex>::k_InterpolateBoundaryEdgeOnly );

Additional information on boundary interpolation rules can be found here


The boundary interpolation rules must be set before the call to HbrMesh::Finish(), which sets the sharpness values to boundary edges and vertices based on these rules.

Adding Creases

Hbr supports a sharpness attribute on both edges and vertices.

Sharpness is set using the SetSharpness(float) accessors.

Vertex Creases

Given an index, vertices are very easy to access in the mesh.

int idx;     // vertex index
float sharp; // the edge sharpness

OpenSubdiv::HbrVertex<Vertex> * v = mesh->GetVertex( idx );

if(v) {
    v->SetSharpness( std::max(0.0f, sharp) );
} else
   printf("cannot find vertex for corner tag (%d)\n", idx );

Edge Creases

Usually, edge creases are described with a vertex indices pair. Here is some sample code to locate the matching half-edge and set a crease sharpness.

int v0, v1;  // vertex indices
float sharp; // the edge sharpness

OpenSubdiv::HbrVertex<Vertex> * v = mesh->GetVertex( v0 ),
                              * w = mesh->GetVertex( v1 );

OpenSubdiv::HbrHalfedge<Vertex> * e = 0;

if( v && w ) {

    if((e = v->GetEdge(w)) == 0)
        e = w->GetEdge(v);

    if(e) {
        e->SetSharpness( std::max(0.0f, sharp) );
    } else
       printf("cannot find edge for crease tag (%d,%d)\n", v0, v1 );


Hbr faces support a "hole" tag.

int idx; // the face index

OpenSubdiv::HbrFace<Vertex> * f = mesh->GetFace( idx );
if(f) {
} else
   printf("cannot find face for hole tag (%d)\n", idx );


The hole tag is hierarchical : sub-faces can also be marked as holes.

See: Hierarchical Edits

Hierarchical Edits

Hbr supports the following types of hierarchical edits:

Type Function
Corner edits Modify vertex sharpness
Crease edits Modify edge sharpness
FaceEdit Modify custom "face data"
FVarEdit Modify face-varying data
VertexEdit Modify vertex and varying data
HoleEdit Set "hole" tag

Modifications are one of the following 3 operations:


Here is a simple example that creates a hierarchical vertex edit.

// path = 655, 2, 3, 0
int faceid = 655,
    nsubfaces = 2,
    subfaces[2] = { 2, 3 },
    vertexid = 0;

int offset = 0,       // offset to the vertex or varying data
    numElems = 3;     // number of elements to apply the modifier to (x,y,z = 3)

bool isP = false;     // shortcut to identify modifications to the vertex position "P"

OpenSubdiv::HbrHierarchicalEdit<Vertex>::Operation op =

float values[3] = { 1.0f, 0.5f, 0.0f }; // edit values

OpenSubdiv::HbrVertexEdit<T> * edit =
     new OpenSubdiv::HbrVertexEdit<T>(faceid,

Face-varying Data

Here is a walk-through of how to store face-varying data for a (u,v) pair. Unlike vertex and varying data which is accessed through the templated vertex API, face-varying data is directly aggregated as vectors of float data.

Instantiating the HbrMesh

The HbrMesh needs to retain some knowledge about the face-varying data that it carries in order to refine it correctly.

int fvarwidth = 2; // total width of the fvar data

static int indices[1] = { 0 }, // 1 offset set to 0
           widths[1] = { 2 };  // 2 floats in a (u,v) pair

int const   fvarcount   = fvarwidth > 0 ? 1 : 0,
          * fvarindices = fvarwidth > 0 ? indices : NULL,
          * fvarwidths  = fvarwidth > 0 ? widths : NULL;

mesh = new OpenSubdiv::HbrMesh<T>( &_scheme,
                                   fvarwidth );

Setting the Face-Varying Data

After the topology has been created, Hbr is ready to accept face-varying data. Here is some sample code:

for (int i=0, idx=0; i<numFaces; ++i ) {

    OpenSubdiv::HbrFace<Vertex> * f = mesh->GetFace(i);

    int nv = f->GetNumVertices(); // note: this is not the fastest way

    OpenSubdiv::HbrHalfedge<Vertex> * e = f->GetFirstEdge();

    for (int j=0; j<nv; ++j, e=e->GetNext()) {

        OpenSubdiv::HbrFVarData<Vertex> & fvt = e->GetOrgVertex()->GetFVarData(f);

        float const * fvdata = GetFaceVaryingData( i, j );

        if (not fvt.IsInitialized()) {

            // if no fvar daa exists yet on the vertex
            fvt.SetAllData(2, fvdata);

        } else if (not fvt.CompareAll(2, fvdata)) {

            // if there already is fvar data and there is a boundary add the new data
            OpenSubdiv::HbrFVarData<T> & nfvt = e->GetOrgVertex()->NewFVarData(f);
            nfvt.SetAllData(2, fvdata);


Retrieving the Face-Varying Data

The HbrFVarData structures are expanded during the refinement process, with every sub-face being assigned a set of interpolated face-varying data. This data can be accessed in 2 ways :

From a face, passing a vertex index:

// OpenSubdiv::HbrFace<Vertex> * f

OpenSubdiv::HbrFVarData const &fv = f.GetFVarData(vindex);

const float * data = fv.GetData()

From a vertex, passing a pointer to an incident face:

// OpenSubdiv::HbrFace<Vertex> * f

OpenSubdiv::HbrFVarData const &fv = myVertex.GetFVarData(f);

const float * data = fv.GetData()

Valence Operators

When manipulating meshes, it is often necessary to iterate over neighboring faces or vertices. Rather than gather lists of pointers and return them, Hbr exposes an operator pattern that guarantees consistent mesh traversals.

The following example shows how to use an operator to count the number of neighboring vertices (use HbrVertex::GetValence() for proper valence counts)

//OpenSubdiv::HbrVertex<Vertex> * v;

class MyOperator : public OpenSubdiv::HbrVertexOperator<Vertex> {

    int count;

    MyOperator() : count(0) { }

    virtual void operator() (OpenSubdiv::HbrVertex<Vertex> &v) {

MyOperator op;

v->ApplyOperatorSurroundingVertices( op );

Managing Singular Vertices

Certain topological configurations would force vertices to share multiple half-edge cycles. Because Hbr is a half-edge representation, these "singular" vertices have to be duplicated as part of the HbrMesh::Finish() phase of the instantiation.

These duplicated vertices can cause problems for client-code that tries to populate buffers of vertex or varying data. The following sample code shows how to match the vertex data to singular vertex splits:

// Populating an OsdCpuVertexBuffer with vertex data (positions,...)
float const * vtxData = inMeshFn.getRawPoints(&returnStatus);

OpenSubdiv::OsdCpuVertexBuffer *vertexBuffer =
    OpenSubdiv::OsdCpuVertexBuffer::Create(numVertexElements, numFarVerts);

vertexBuffer->UpdateData(vtxData, 0, numVertices );

// Duplicate the vertex data into the split singular vertices
std::vector<std::pair<int, int> > const splits = hbrMesh->GetSplitVertices();
for (int i=0; i<(int)splits.size(); ++i) {
    vertexBuffer->UpdateData(vtxData+splits[i].second*numVertexElements, splits[i].first, 1);