Change RenderableModelEntityItem::computeShapeInfo to support new HFM format

This commit is contained in:
sabrina-shanman 2019-11-04 13:59:36 -08:00
parent 88b6472ced
commit 33bb0e2b19

View file

@ -380,26 +380,35 @@ void RenderableModelEntityItem::computeShapeInfo(ShapeInfo& shapeInfo) {
ShapeInfo::PointCollection& pointCollection = shapeInfo.getPointCollection();
pointCollection.clear();
uint32_t i = 0;
size_t numParts = 0;
for (const HFMMesh& mesh : collisionGeometry.meshes) {
numParts += mesh.triangleListMesh.parts.size();
}
pointCollection.reserve(numParts);
// the way OBJ files get read, each section under a "g" line is its own meshPart. We only expect
// to find one actual "mesh" (with one or more meshParts in it), but we loop over the meshes, just in case.
foreach (const HFMMesh& mesh, collisionGeometry.meshes) {
for (const HFMMesh& mesh : collisionGeometry.meshes) {
const hfm::TriangleListMesh& triangleListMesh = mesh.triangleListMesh;
// each meshPart is a convex hull
foreach (const HFMMeshPart &meshPart, mesh.parts) {
pointCollection.emplace_back();
ShapeInfo::PointList& pointsInPart = pointCollection[i];
for (const glm::ivec2& part : triangleListMesh.parts) {
// run through all the triangles and (uniquely) add each point to the hull
uint32_t numIndices = (uint32_t)meshPart.triangleIndices.size();
pointCollection.emplace_back();
ShapeInfo::PointList& pointsInPart = pointCollection.back();
uint32_t numIndices = (uint32_t)part.y;
// TODO: assert rather than workaround after we start sanitizing HFMMesh higher up
//assert(numIndices % TRIANGLE_STRIDE == 0);
numIndices -= numIndices % TRIANGLE_STRIDE; // WORKAROUND lack of sanity checking in FBXSerializer
for (uint32_t j = 0; j < numIndices; j += TRIANGLE_STRIDE) {
glm::vec3 p0 = mesh.vertices[meshPart.triangleIndices[j]];
glm::vec3 p1 = mesh.vertices[meshPart.triangleIndices[j + 1]];
glm::vec3 p2 = mesh.vertices[meshPart.triangleIndices[j + 2]];
uint32_t indexStart = (uint32_t)part.x;
uint32_t indexEnd = indexStart + numIndices;
for (uint32_t j = indexStart; j < indexEnd; j += TRIANGLE_STRIDE) {
// NOTE: It seems odd to skip vertices when initializing a btConvexHullShape, but let's keep the behavior similar to the old behavior for now
glm::vec3 p0 = triangleListMesh.vertices[triangleListMesh.indices[j]];
glm::vec3 p1 = triangleListMesh.vertices[triangleListMesh.indices[j + 1]];
glm::vec3 p2 = triangleListMesh.vertices[triangleListMesh.indices[j + 2]];
if (std::find(pointsInPart.cbegin(), pointsInPart.cend(), p0) == pointsInPart.cend()) {
pointsInPart.push_back(p0);
}
@ -411,37 +420,11 @@ void RenderableModelEntityItem::computeShapeInfo(ShapeInfo& shapeInfo) {
}
}
// run through all the quads and (uniquely) add each point to the hull
numIndices = (uint32_t)meshPart.quadIndices.size();
// TODO: assert rather than workaround after we start sanitizing HFMMesh higher up
//assert(numIndices % QUAD_STRIDE == 0);
numIndices -= numIndices % QUAD_STRIDE; // WORKAROUND lack of sanity checking in FBXSerializer
for (uint32_t j = 0; j < numIndices; j += QUAD_STRIDE) {
glm::vec3 p0 = mesh.vertices[meshPart.quadIndices[j]];
glm::vec3 p1 = mesh.vertices[meshPart.quadIndices[j + 1]];
glm::vec3 p2 = mesh.vertices[meshPart.quadIndices[j + 2]];
glm::vec3 p3 = mesh.vertices[meshPart.quadIndices[j + 3]];
if (std::find(pointsInPart.cbegin(), pointsInPart.cend(), p0) == pointsInPart.cend()) {
pointsInPart.push_back(p0);
}
if (std::find(pointsInPart.cbegin(), pointsInPart.cend(), p1) == pointsInPart.cend()) {
pointsInPart.push_back(p1);
}
if (std::find(pointsInPart.cbegin(), pointsInPart.cend(), p2) == pointsInPart.cend()) {
pointsInPart.push_back(p2);
}
if (std::find(pointsInPart.cbegin(), pointsInPart.cend(), p3) == pointsInPart.cend()) {
pointsInPart.push_back(p3);
}
}
if (pointsInPart.size() == 0) {
qCDebug(entitiesrenderer) << "Warning -- meshPart has no faces";
pointCollection.pop_back();
continue;
}
++i;
}
}
@ -474,61 +457,83 @@ void RenderableModelEntityItem::computeShapeInfo(ShapeInfo& shapeInfo) {
QVector<glm::mat4> localTransforms;
const HFMModel& hfmModel = model->getHFMModel();
uint32_t numHFMShapes = (uint32_t)hfmModel.shapes.size();
uint32_t numHFMMeshes = (uint32_t)hfmModel.meshes.size();
int totalNumVertices = 0;
localTransforms.reserve(numHFMShapes);
glm::vec3 dimensions = getScaledDimensions();
glm::mat4 invRegistraionOffset = glm::translate(dimensions * (getRegistrationPoint() - ENTITY_ITEM_DEFAULT_REGISTRATION_POINT));
for (uint32_t s = 0; s < numHFMShapes; s++) {
const HFMShape& shape = hfmModel.shapes[s];
// for (uint32_t i = 0; i < numHFMMeshes; i++) {
const HFMMesh& mesh = hfmModel.meshes.at(shape.mesh);
const HFMMeshPart& part = mesh.parts.at(shape.meshPart);
/* if (shape.skinDeformer != hfm::UNDEFINED_KEY) {
const HFMCluster& cluster = hfmModel.skinDeformers[shape.skinDeformer].clusters.at(0);
auto jointMatrix = model->getRig().getJointTransform(cluster.jointIndex);
// we backtranslate by the registration offset so we can apply that offset to the shapeInfo later
localTransforms.push_back(invRegistraionOffset * jointMatrix * cluster.inverseBindMatrix);
} else {*/
if (shape.joint != hfm::UNDEFINED_KEY) {
auto jointMatrix = model->getRig().getJointTransform(shape.joint);
// we backtranslate by the registration offset so we can apply that offset to the shapeInfo later
localTransforms.push_back(invRegistraionOffset * jointMatrix/* * cluster.inverseBindMatrix*/);
if (shape.skinDeformer != hfm::UNDEFINED_KEY) {
const auto& skinDeformer = hfmModel.skinDeformers[shape.skinDeformer];
glm::mat4 inverseBindMatrix;
if (!skinDeformer.clusters.empty()) {
const auto& cluster = skinDeformer.clusters.back();
inverseBindMatrix = cluster.inverseBindMatrix;
}
localTransforms.push_back(invRegistraionOffset * jointMatrix * inverseBindMatrix);
} else {
localTransforms.push_back(invRegistraionOffset * jointMatrix);
}
} else {
localTransforms.push_back(invRegistraionOffset);
}
/* if (i < hfmModel.skinDeformers.size() && hfmModel.skinDeformers[i].clusters.size() > 0) {
const HFMCluster& cluster = hfmModel.skinDeformers[i].clusters.at(0);
auto jointMatrix = model->getRig().getJointTransform(cluster.jointIndex);
// we backtranslate by the registration offset so we can apply that offset to the shapeInfo later
localTransforms.push_back(invRegistraionOffset * jointMatrix * cluster.inverseBindMatrix);
} else {
localTransforms.push_back(invRegistraionOffset);
}*/
totalNumVertices += mesh.vertices.size();
}
const int32_t MAX_VERTICES_PER_STATIC_MESH = 1e6;
if (totalNumVertices > MAX_VERTICES_PER_STATIC_MESH) {
qWarning() << "model" << getModelURL() << "has too many vertices" << totalNumVertices << "and will collide as a box.";
ShapeInfo::TriangleIndices& triangleIndices = shapeInfo.getTriangleIndices();
triangleIndices.clear();
Extents extents;
int32_t shapeCount = 0;
int32_t instanceIndex = 0;
// NOTE: Each pointCollection corresponds to a mesh. Therefore, we should have one pointCollection per mesh instance
// A mesh instance is a unique combination of mesh/transform. For every mesh instance, there are as many shapes as there are parts for that mesh.
// We assume the shapes are grouped by mesh instance, and the group contains one of each mesh part.
uint32_t numInstances = 0;
std::vector<std::vector<std::vector<uint32_t>>> shapesPerInstancePerMesh;
shapesPerInstancePerMesh.resize(hfmModel.meshes.size());
for (uint32_t shapeIndex = 0; shapeIndex < hfmModel.shapes.size();) {
const auto& shape = hfmModel.shapes[shapeIndex];
uint32_t meshIndex = shape.mesh;
const auto& mesh = hfmModel.meshes[meshIndex];
uint32_t numMeshParts = (uint32_t)mesh.parts.size();
assert(numMeshParts != 0);
auto& shapesPerInstance = shapesPerInstancePerMesh[meshIndex];
shapesPerInstance.emplace_back();
auto& shapes = shapesPerInstance.back();
shapes.resize(numMeshParts);
std::iota(shapes.begin(), shapes.end(), shapeIndex);
shapeIndex += numMeshParts;
++numInstances;
}
const uint32_t MAX_ALLOWED_MESH_COUNT = 1000;
if (numInstances > MAX_ALLOWED_MESH_COUNT) {
// too many will cause the deadlock timer to throw...
qWarning() << "model" << getModelURL() << "has too many collision meshes" << numInstances << "and will collide as a box.";
shapeInfo.setParams(SHAPE_TYPE_BOX, 0.5f * dimensions);
return;
}
std::vector<std::shared_ptr<const graphics::Mesh>> meshes;
if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
auto& hfmMeshes = _collisionGeometryResource->getHFMModel().meshes;
meshes.reserve(hfmMeshes.size());
for (auto& hfmMesh : hfmMeshes) {
meshes.push_back(hfmMesh._mesh);
size_t totalNumVertices = 0;
for (const auto& shapesPerInstance : shapesPerInstancePerMesh) {
for (const auto& instanceShapes : shapesPerInstance) {
const uint32_t firstShapeIndex = instanceShapes.front();
const auto& firstShape = hfmModel.shapes[firstShapeIndex];
const auto& mesh = hfmModel.meshes[firstShape.mesh];
const auto& triangleListMesh = mesh.triangleListMesh;
// Added once per instance per mesh
totalNumVertices += triangleListMesh.vertices.size();
}
} else {
meshes = model->getNetworkModel()->getMeshes();
}
int32_t numMeshes = (int32_t)(meshes.size());
const int MAX_ALLOWED_MESH_COUNT = 1000;
if (numMeshes > MAX_ALLOWED_MESH_COUNT) {
// too many will cause the deadlock timer to throw...
const size_t MAX_VERTICES_PER_STATIC_MESH = 1e6;
if (totalNumVertices > MAX_VERTICES_PER_STATIC_MESH) {
qWarning() << "model" << getModelURL() << "has too many vertices" << totalNumVertices << "and will collide as a box.";
shapeInfo.setParams(SHAPE_TYPE_BOX, 0.5f * dimensions);
return;
}
@ -536,169 +541,118 @@ void RenderableModelEntityItem::computeShapeInfo(ShapeInfo& shapeInfo) {
ShapeInfo::PointCollection& pointCollection = shapeInfo.getPointCollection();
pointCollection.clear();
if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
pointCollection.resize(numMeshes);
pointCollection.resize(numInstances);
} else {
pointCollection.resize(1);
}
ShapeInfo::TriangleIndices& triangleIndices = shapeInfo.getTriangleIndices();
triangleIndices.clear();
for (uint32_t meshIndex = 0; meshIndex < hfmModel.meshes.size(); ++meshIndex) {
const auto& mesh = hfmModel.meshes[meshIndex];
const auto& triangleListMesh = mesh.triangleListMesh;
const auto& vertices = triangleListMesh.vertices;
const auto& indices = triangleListMesh.indices;
const std::vector<glm::ivec2>& parts = triangleListMesh.parts;
Extents extents;
int32_t meshCount = 0;
int32_t pointListIndex = 0;
for (auto& mesh : meshes) {
if (!mesh) {
continue;
}
const gpu::BufferView& vertices = mesh->getVertexBuffer();
const gpu::BufferView& indices = mesh->getIndexBuffer();
const gpu::BufferView& parts = mesh->getPartBuffer();
const auto& shapesPerInstance = shapesPerInstancePerMesh[meshIndex];
for (const std::vector<uint32_t>& instanceShapes : shapesPerInstance) {
ShapeInfo::PointList& points = pointCollection[instanceIndex];
ShapeInfo::PointList& points = pointCollection[pointListIndex];
// reserve room
int32_t sizeToReserve = (int32_t)(vertices.size());
if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
// a list of points for each instance
instanceIndex++;
} else {
// only one list of points
sizeToReserve += (int32_t)((gpu::Size)points.size());
}
points.reserve(sizeToReserve);
// get mesh instance transform
const uint32_t meshIndexOffset = (uint32_t)points.size();
const uint32_t instanceShapeIndexForTransform = instanceShapes.front();
const auto& instanceShapeForTransform = hfmModel.shapes[instanceShapeIndexForTransform];
glm::mat4 localTransform;
if (instanceShapeForTransform.joint != hfm::UNDEFINED_KEY) {
auto jointMatrix = model->getRig().getJointTransform(instanceShapeForTransform.joint);
// we backtranslate by the registration offset so we can apply that offset to the shapeInfo later
if (instanceShapeForTransform.skinDeformer != hfm::UNDEFINED_KEY) {
const auto& skinDeformer = hfmModel.skinDeformers[instanceShapeForTransform.skinDeformer];
glm::mat4 inverseBindMatrix;
if (!skinDeformer.clusters.empty()) {
const auto& cluster = skinDeformer.clusters.back();
inverseBindMatrix = cluster.inverseBindMatrix;
}
localTransform = invRegistraionOffset * jointMatrix * inverseBindMatrix;
} else {
localTransform = invRegistraionOffset * jointMatrix;
}
} else {
localTransform = invRegistraionOffset;
}
// reserve room
int32_t sizeToReserve = (int32_t)(vertices.getNumElements());
if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
// a list of points for each mesh
pointListIndex++;
} else {
// only one list of points
sizeToReserve += (int32_t)((gpu::Size)points.size());
}
points.reserve(sizeToReserve);
// copy points
auto vertexItr = vertices.cbegin();
while (vertexItr != vertices.cend()) {
glm::vec3 point = extractTranslation(localTransform * glm::translate(*vertexItr));
points.push_back(point);
++vertexItr;
}
for (const auto& instanceShapeIndex : instanceShapes) {
const auto& instanceShape = hfmModel.shapes[instanceShapeIndex];
extents.addExtents(instanceShape.transformedExtents);
}
// copy points
uint32_t meshIndexOffset = (uint32_t)points.size();
const glm::mat4& localTransform = localTransforms[meshCount];
gpu::BufferView::Iterator<const glm::vec3> vertexItr = vertices.cbegin<const glm::vec3>();
while (vertexItr != vertices.cend<const glm::vec3>()) {
glm::vec3 point = extractTranslation(localTransform * glm::translate(*vertexItr));
points.push_back(point);
extents.addPoint(point);
++vertexItr;
}
if (type == SHAPE_TYPE_STATIC_MESH) {
// copy into triangleIndices
triangleIndices.reserve((int32_t)((gpu::Size)(triangleIndices.size()) + indices.getNumElements()));
gpu::BufferView::Iterator<const graphics::Mesh::Part> partItr = parts.cbegin<const graphics::Mesh::Part>();
while (partItr != parts.cend<const graphics::Mesh::Part>()) {
auto numIndices = partItr->_numIndices;
if (partItr->_topology == graphics::Mesh::TRIANGLES) {
if (type == SHAPE_TYPE_STATIC_MESH) {
// copy into triangleIndices
triangleIndices.reserve((int32_t)((gpu::Size)(triangleIndices.size()) + indices.size()));
auto partItr = parts.cbegin();
while (partItr != parts.cend()) {
auto numIndices = partItr->y;
// TODO: assert rather than workaround after we start sanitizing HFMMesh higher up
//assert(numIndices % TRIANGLE_STRIDE == 0);
numIndices -= numIndices % TRIANGLE_STRIDE; // WORKAROUND lack of sanity checking in FBXSerializer
auto indexItr = indices.cbegin<const gpu::BufferView::Index>() + partItr->_startIndex;
auto indexItr = indices.cbegin() + partItr->x;
auto indexEnd = indexItr + numIndices;
while (indexItr != indexEnd) {
triangleIndices.push_back(*indexItr + meshIndexOffset);
++indexItr;
}
} else if (partItr->_topology == graphics::Mesh::TRIANGLE_STRIP) {
// TODO: resurrect assert after we start sanitizing HFMMesh higher up
//assert(numIndices > 2);
uint32_t approxNumIndices = TRIANGLE_STRIDE * numIndices;
if (approxNumIndices > (uint32_t)(triangleIndices.capacity() - triangleIndices.size())) {
// we underestimated the final size of triangleIndices so we pre-emptively expand it
triangleIndices.reserve(triangleIndices.size() + approxNumIndices);
}
auto indexItr = indices.cbegin<const gpu::BufferView::Index>() + partItr->_startIndex;
auto indexEnd = indexItr + (numIndices - 2);
// first triangle uses the first three indices
triangleIndices.push_back(*(indexItr++) + meshIndexOffset);
triangleIndices.push_back(*(indexItr++) + meshIndexOffset);
triangleIndices.push_back(*(indexItr++) + meshIndexOffset);
// the rest use previous and next index
uint32_t triangleCount = 1;
while (indexItr != indexEnd) {
if ((*indexItr) != graphics::Mesh::PRIMITIVE_RESTART_INDEX) {
if (triangleCount % 2 == 0) {
// even triangles use first two indices in order
triangleIndices.push_back(*(indexItr - 2) + meshIndexOffset);
triangleIndices.push_back(*(indexItr - 1) + meshIndexOffset);
} else {
// odd triangles swap order of first two indices
triangleIndices.push_back(*(indexItr - 1) + meshIndexOffset);
triangleIndices.push_back(*(indexItr - 2) + meshIndexOffset);
}
triangleIndices.push_back(*indexItr + meshIndexOffset);
++triangleCount;
}
++indexItr;
}
++partItr;
}
++partItr;
}
} else if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
// for each mesh copy unique part indices, separated by special bogus (flag) index values
gpu::BufferView::Iterator<const graphics::Mesh::Part> partItr = parts.cbegin<const graphics::Mesh::Part>();
while (partItr != parts.cend<const graphics::Mesh::Part>()) {
// collect unique list of indices for this part
std::set<int32_t> uniqueIndices;
auto numIndices = partItr->_numIndices;
if (partItr->_topology == graphics::Mesh::TRIANGLES) {
} else if (type == SHAPE_TYPE_SIMPLE_COMPOUND) {
// for each mesh copy unique part indices, separated by special bogus (flag) index values
auto partItr = parts.cbegin();
while (partItr != parts.cend()) {
// collect unique list of indices for this part
std::set<int32_t> uniqueIndices;
auto numIndices = partItr->y;
// TODO: assert rather than workaround after we start sanitizing HFMMesh higher up
//assert(numIndices% TRIANGLE_STRIDE == 0);
numIndices -= numIndices % TRIANGLE_STRIDE; // WORKAROUND lack of sanity checking in FBXSerializer
auto indexItr = indices.cbegin<const gpu::BufferView::Index>() + partItr->_startIndex;
auto indexItr = indices.cbegin() + partItr->x;
auto indexEnd = indexItr + numIndices;
while (indexItr != indexEnd) {
uniqueIndices.insert(*indexItr);
++indexItr;
}
} else if (partItr->_topology == graphics::Mesh::TRIANGLE_STRIP) {
// TODO: resurrect assert after we start sanitizing HFMMesh higher up
//assert(numIndices > TRIANGLE_STRIDE - 1);
auto indexItr = indices.cbegin<const gpu::BufferView::Index>() + partItr->_startIndex;
auto indexEnd = indexItr + (numIndices - 2);
// first triangle uses the first three indices
uniqueIndices.insert(*(indexItr++));
uniqueIndices.insert(*(indexItr++));
uniqueIndices.insert(*(indexItr++));
// the rest use previous and next index
uint32_t triangleCount = 1;
while (indexItr != indexEnd) {
if ((*indexItr) != graphics::Mesh::PRIMITIVE_RESTART_INDEX) {
if (triangleCount % 2 == 0) {
// EVEN triangles use first two indices in order
uniqueIndices.insert(*(indexItr - 2));
uniqueIndices.insert(*(indexItr - 1));
} else {
// ODD triangles swap order of first two indices
uniqueIndices.insert(*(indexItr - 1));
uniqueIndices.insert(*(indexItr - 2));
}
uniqueIndices.insert(*indexItr);
++triangleCount;
}
++indexItr;
// store uniqueIndices in triangleIndices
triangleIndices.reserve(triangleIndices.size() + (int32_t)uniqueIndices.size());
for (auto index : uniqueIndices) {
triangleIndices.push_back(index);
}
}
// flag end of part
triangleIndices.push_back(END_OF_MESH_PART);
// store uniqueIndices in triangleIndices
triangleIndices.reserve(triangleIndices.size() + (int32_t)uniqueIndices.size());
for (auto index : uniqueIndices) {
triangleIndices.push_back(index);
++partItr;
}
// flag end of part
triangleIndices.push_back(END_OF_MESH_PART);
++partItr;
// flag end of mesh
triangleIndices.push_back(END_OF_MESH);
}
// flag end of mesh
triangleIndices.push_back(END_OF_MESH);
}
++meshCount;
++shapeCount;
}
// scale and shift