Merge pull request #13845 from SamGondelman/perf

Move blendshape packing to worker thread

interface/src/Menu.cpp

@@ -451,6 +451,9 @@ Menu::Menu() {
         });
     }
 
+    addCheckableActionToQMenuAndActionHash(renderOptionsMenu, MenuOption::ComputeBlendshapes, 0, true,
+        DependencyManager::get<ModelBlender>().data(), SLOT(setComputeBlendshapes(bool)));
+
     // Developer > Assets >>>
     // Menu item is not currently needed but code should be kept in case it proves useful again at some stage.
 //#define WANT_ASSET_MIGRATION

interface/src/Menu.h

@@ -221,6 +221,7 @@ namespace MenuOption {
     const QString NotificationSounds = "play_notification_sounds";
     const QString NotificationSoundsSnapshot = "play_notification_sounds_snapshot";
     const QString NotificationSoundsTablet = "play_notification_sounds_tablet";
+    const QString ComputeBlendshapes = "Compute Blendshapes";
 }
 
 #endif // hifi_Menu_h

interface/src/avatar/AvatarManager.cpp

@@ -359,21 +359,21 @@ void AvatarManager::simulateAvatarFades(float deltaTime) {
     QReadLocker locker(&_hashLock);
     QVector<AvatarSharedPointer>::iterator avatarItr = _avatarsToFade.begin();
     const render::ScenePointer& scene = qApp->getMain3DScene();
+    render::Transaction transaction;
     while (avatarItr != _avatarsToFade.end()) {
         auto avatar = std::static_pointer_cast<Avatar>(*avatarItr);
         avatar->updateFadingStatus(scene);
         if (!avatar->isFading()) {
             // fading to zero is such a rare event we push a unique transaction for each
             if (avatar->isInScene()) {
-                render::Transaction transaction;
                 avatar->removeFromScene(*avatarItr, scene, transaction);
-                scene->enqueueTransaction(transaction);
             }
             avatarItr = _avatarsToFade.erase(avatarItr);
         } else {
             ++avatarItr;
         }
     }
+    scene->enqueueTransaction(transaction);
 }
 
 AvatarSharedPointer AvatarManager::newSharedAvatar() {

libraries/fbx/src/FBX.h

@@ -28,6 +28,20 @@
 #include <graphics/Geometry.h>
 #include <graphics/Material.h>
 
+#if defined(Q_OS_ANDROID)
+#define FBX_PACK_NORMALS 0
+#else
+#define FBX_PACK_NORMALS 1
+#endif
+
+#if FBX_PACK_NORMALS
+using NormalType = glm::uint32;
+#define FBX_NORMAL_ELEMENT gpu::Element::VEC4F_NORMALIZED_XYZ10W2
+#else
+using NormalType = glm::vec3;
+#define FBX_NORMAL_ELEMENT gpu::Element::VEC3F_XYZ
+#endif
+
 // See comment in FBXReader::parseFBX().
 static const int FBX_HEADER_BYTES_BEFORE_VERSION = 23;
 static const QByteArray FBX_BINARY_PROLOG("Kaydara FBX Binary  ");
@@ -226,6 +240,7 @@ public:
     QVector<glm::vec3> vertices;
     QVector<glm::vec3> normals;
     QVector<glm::vec3> tangents;
+    mutable QVector<NormalType> normalsAndTangents; // Populated later if needed for blendshapes
     QVector<glm::vec3> colors;
     QVector<glm::vec2> texCoords;
     QVector<glm::vec2> texCoords1;

libraries/fbx/src/FBXReader.h

@@ -34,20 +34,6 @@
 class QIODevice;
 class FBXNode;
 
-#if defined(Q_OS_ANDROID)
-#define FBX_PACK_NORMALS 0
-#else
-#define FBX_PACK_NORMALS 1
-#endif
-
-#if FBX_PACK_NORMALS
-using NormalType = glm::uint32;
-#define FBX_NORMAL_ELEMENT gpu::Element::VEC4F_NORMALIZED_XYZ10W2
-#else
-using NormalType = glm::vec3;
-#define FBX_NORMAL_ELEMENT gpu::Element::VEC3F_XYZ
-#endif
-
 /// Reads FBX geometry from the supplied model and mapping data.
 /// \exception QString if an error occurs in parsing
 FBXGeometry* readFBX(const QByteArray& model, const QVariantHash& mapping, const QString& url = "", bool loadLightmaps = true, float lightmapLevel = 1.0f);

libraries/graphics/src/graphics/BufferViewHelpers.cpp

@@ -20,9 +20,6 @@
 #include <AABox.h>
 #include <Extents.h>
 
-#include <glm/gtc/packing.hpp>
-#include <glm/detail/type_vec.hpp>
-
 namespace glm {
     using hvec2 = glm::tvec2<glm::detail::hdata>;
     using hvec4 = glm::tvec4<glm::detail::hdata>;
@@ -62,32 +59,6 @@ namespace {
     }
 }
 
-void packNormalAndTangent(glm::vec3 normal, glm::vec3 tangent, glm::uint32& packedNormal, glm::uint32& packedTangent) {
-    auto absNormal = glm::abs(normal);
-    auto absTangent = glm::abs(tangent);
-    normal /= glm::max(1e-6f, glm::max(glm::max(absNormal.x, absNormal.y), absNormal.z));
-    tangent /= glm::max(1e-6f, glm::max(glm::max(absTangent.x, absTangent.y), absTangent.z));
-    normal = glm::clamp(normal, -1.0f, 1.0f);
-    tangent = glm::clamp(tangent, -1.0f, 1.0f);
-    normal *= 511.0f;
-    tangent *= 511.0f;
-    normal = glm::round(normal);
-    tangent = glm::round(tangent);
-
-    glm::detail::i10i10i10i2 normalStruct;
-    glm::detail::i10i10i10i2 tangentStruct;
-    normalStruct.data.x = int(normal.x);
-    normalStruct.data.y = int(normal.y);
-    normalStruct.data.z = int(normal.z);
-    normalStruct.data.w = 0;
-    tangentStruct.data.x = int(tangent.x);
-    tangentStruct.data.y = int(tangent.y);
-    tangentStruct.data.z = int(tangent.z);
-    tangentStruct.data.w = 0;
-    packedNormal = normalStruct.pack;
-    packedTangent = tangentStruct.pack;
-}
-
 template <typename T>
 glm::uint32 forEachGlmVec(const gpu::BufferView& view, std::function<bool(glm::uint32 index, const T& value)> func) {
     QVector<glm::uint32> result;

libraries/graphics/src/graphics/BufferViewHelpers.h

@@ -9,6 +9,8 @@
 #include <QtCore>
 #include <memory>
 #include <glm/glm.hpp>
+#include <glm/gtc/packing.hpp>
+#include <glm/detail/type_vec.hpp>
 
 #include "GpuHelpers.h"
 
@@ -44,7 +46,31 @@ namespace buffer_helpers {
     gpu::BufferView clone(const gpu::BufferView& input);
     gpu::BufferView resized(const gpu::BufferView& input, glm::uint32 numElements);
 
-    void packNormalAndTangent(glm::vec3 normal, glm::vec3 tangent, glm::uint32& packedNormal, glm::uint32& packedTangent);
+    inline void packNormalAndTangent(glm::vec3 normal, glm::vec3 tangent, glm::uint32& packedNormal, glm::uint32& packedTangent) {
+        auto absNormal = glm::abs(normal);
+        auto absTangent = glm::abs(tangent);
+        normal /= glm::max(1e-6f, glm::max(glm::max(absNormal.x, absNormal.y), absNormal.z));
+        tangent /= glm::max(1e-6f, glm::max(glm::max(absTangent.x, absTangent.y), absTangent.z));
+        normal = glm::clamp(normal, -1.0f, 1.0f);
+        tangent = glm::clamp(tangent, -1.0f, 1.0f);
+        normal *= 511.0f;
+        tangent *= 511.0f;
+        normal = glm::round(normal);
+        tangent = glm::round(tangent);
+
+        glm::detail::i10i10i10i2 normalStruct;
+        glm::detail::i10i10i10i2 tangentStruct;
+        normalStruct.data.x = int(normal.x);
+        normalStruct.data.y = int(normal.y);
+        normalStruct.data.z = int(normal.z);
+        normalStruct.data.w = 0;
+        tangentStruct.data.x = int(tangent.x);
+        tangentStruct.data.y = int(tangent.y);
+        tangentStruct.data.z = int(tangent.z);
+        tangentStruct.data.w = 0;
+        packedNormal = normalStruct.pack;
+        packedTangent = tangentStruct.pack;
+    }
 
     namespace mesh {
         glm::uint32 forEachVertex(const graphics::MeshPointer& mesh, std::function<bool(glm::uint32 index, const QVariantMap& attributes)> func);

libraries/render-utils/src/CauterizedModel.cpp

@@ -76,6 +76,7 @@ void CauterizedModel::createRenderItemSet() {
         // Run through all of the meshes, and place them into their segregated, but unsorted buckets
         int shapeID = 0;
         uint32_t numMeshes = (uint32_t)meshes.size();
+        const FBXGeometry& fbxGeometry = getFBXGeometry();
         for (uint32_t i = 0; i < numMeshes; i++) {
             const auto& mesh = meshes.at(i);
             if (!mesh) {
@@ -85,6 +86,10 @@ void CauterizedModel::createRenderItemSet() {
             // Create the render payloads
             int numParts = (int)mesh->getNumParts();
             for (int partIndex = 0; partIndex < numParts; partIndex++) {
+                if (!fbxGeometry.meshes[i].blendshapes.empty() && !_blendedVertexBuffers[i]) {
+                    _blendedVertexBuffers[i] = std::make_shared<gpu::Buffer>();
+                    _blendedVertexBuffers[i]->resize(fbxGeometry.meshes[i].vertices.size() * (sizeof(glm::vec3) + 2 * sizeof(NormalType)));
+                }
                 auto ptr = std::make_shared<CauterizedMeshPartPayload>(shared_from_this(), i, partIndex, shapeID, transform, offset);
                 _modelMeshRenderItems << std::static_pointer_cast<ModelMeshPartPayload>(ptr);
                 auto material = getGeometry()->getShapeMaterial(shapeID);
@@ -170,9 +175,10 @@ void CauterizedModel::updateClusterMatrices() {
     }
 
     // post the blender if we're not currently waiting for one to finish
-    if (geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
+    auto modelBlender = DependencyManager::get<ModelBlender>();
+    if (modelBlender->shouldComputeBlendshapes() && geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
         _blendedBlendshapeCoefficients = _blendshapeCoefficients;
-        DependencyManager::get<ModelBlender>()->noteRequiresBlend(getThisPointer());
+        modelBlender->noteRequiresBlend(getThisPointer());
     }
 }
 

libraries/render-utils/src/MeshPartPayload.cpp

@@ -208,7 +208,9 @@ ModelMeshPartPayload::ModelMeshPartPayload(ModelPointer model, int meshIndex, in
 
     bool useDualQuaternionSkinning = model->getUseDualQuaternionSkinning();
 
-    _blendedVertexBuffer = model->_blendedVertexBuffers[_meshIndex];
+    if (!model->getFBXGeometry().meshes[meshIndex].blendshapes.isEmpty()) {
+        _blendedVertexBuffer = model->_blendedVertexBuffers[meshIndex];
+    }
     auto& modelMesh = model->getGeometry()->getMeshes().at(_meshIndex);
     const Model::MeshState& state = model->getMeshState(_meshIndex);
 

libraries/render-utils/src/Model.cpp

@@ -42,8 +42,9 @@
 using namespace std;
 
 int nakedModelPointerTypeId = qRegisterMetaType<ModelPointer>();
-int weakGeometryResourceBridgePointerTypeId = qRegisterMetaType<Geometry::WeakPointer >();
+int weakGeometryResourceBridgePointerTypeId = qRegisterMetaType<Geometry::WeakPointer>();
-int vec3VectorTypeId = qRegisterMetaType<QVector<glm::vec3> >();
+int vec3VectorTypeId = qRegisterMetaType<QVector<glm::vec3>>();
+int normalTypeVecTypeId = qRegisterMetaType<QVector<NormalType>>("QVector<NormalType>");
 float Model::FAKE_DIMENSION_PLACEHOLDER = -1.0f;
 #define HTTP_INVALID_COM "http://invalid.com"
 
@@ -301,41 +302,52 @@ bool Model::updateGeometry() {
         assert(_meshStates.empty());
 
         const FBXGeometry& fbxGeometry = getFBXGeometry();
+        int i = 0;
         foreach (const FBXMesh& mesh, fbxGeometry.meshes) {
             MeshState state;
             state.clusterDualQuaternions.resize(mesh.clusters.size());
             state.clusterMatrices.resize(mesh.clusters.size());
             _meshStates.push_back(state);
 
-            // Note: we add empty buffers for meshes that lack blendshapes so we can access the buffers by index
-            // later in ModelMeshPayload, however the vast majority of meshes will not have them.
-            // TODO? make _blendedVertexBuffers a map instead of vector and only add for meshes with blendshapes?
-            auto buffer = std::make_shared<gpu::Buffer>();
             if (!mesh.blendshapes.isEmpty()) {
-                std::vector<NormalType> normalsAndTangents;
+                if (!_blendedVertexBuffers[i]) {
-                normalsAndTangents.reserve(mesh.normals.size() + mesh.tangents.size());
+                    _blendedVertexBuffers[i] = std::make_shared<gpu::Buffer>();
-
-                for (auto normalIt = mesh.normals.begin(), tangentIt = mesh.tangents.begin();
-                     normalIt != mesh.normals.end();
-                     ++normalIt, ++tangentIt) {
-#if FBX_PACK_NORMALS
-                    glm::uint32 finalNormal;
-                    glm::uint32 finalTangent;
-                    buffer_helpers::packNormalAndTangent(*normalIt, *tangentIt, finalNormal, finalTangent);
-#else
-                    const auto finalNormal = *normalIt;
-                    const auto finalTangent = *tangentIt;
-#endif
-                    normalsAndTangents.push_back(finalNormal);
-                    normalsAndTangents.push_back(finalTangent);
                 }
+                const auto& buffer = _blendedVertexBuffers[i];
+                QVector<NormalType> normalsAndTangents;
+                normalsAndTangents.resize(2 * mesh.normals.size());
 
-                buffer->resize(mesh.vertices.size() * (sizeof(glm::vec3) + 2 * sizeof(NormalType)));
+                // Interleave normals and tangents
-                buffer->setSubData(0, mesh.vertices.size() * sizeof(glm::vec3), (const gpu::Byte*) mesh.vertices.constData());
+                // Parallel version for performance
-                buffer->setSubData(mesh.vertices.size() * sizeof(glm::vec3),
+                tbb::parallel_for(tbb::blocked_range<int>(0, mesh.normals.size()), [&](const tbb::blocked_range<int>& range) {
-                                   mesh.normals.size() * 2 * sizeof(NormalType), (const gpu::Byte*) normalsAndTangents.data());
+                    auto normalsRange = std::make_pair(mesh.normals.begin() + range.begin(), mesh.normals.begin() + range.end());
+                    auto tangentsRange = std::make_pair(mesh.tangents.begin() + range.begin(), mesh.tangents.begin() + range.end());
+                    auto normalsAndTangentsIt = normalsAndTangents.begin() + 2 * range.begin();
+
+                    for (auto normalIt = normalsRange.first, tangentIt = tangentsRange.first;
+                         normalIt != normalsRange.second;
+                         ++normalIt, ++tangentIt) {
+#if FBX_PACK_NORMALS
+                        glm::uint32 finalNormal;
+                        glm::uint32 finalTangent;
+                        buffer_helpers::packNormalAndTangent(*normalIt, *tangentIt, finalNormal, finalTangent);
+#else
+                        const auto& finalNormal = *normalIt;
+                        const auto& finalTangent = *tangentIt;
+#endif
+                        *normalsAndTangentsIt = finalNormal;
+                        ++normalsAndTangentsIt;
+                        *normalsAndTangentsIt = finalTangent;
+                        ++normalsAndTangentsIt;
+                    }
+                });
+                const auto verticesSize = mesh.vertices.size() * sizeof(glm::vec3);
+                buffer->resize(mesh.vertices.size() * sizeof(glm::vec3) + normalsAndTangents.size() * sizeof(NormalType));
+                buffer->setSubData(0, verticesSize, (const gpu::Byte*) mesh.vertices.constData());
+                buffer->setSubData(verticesSize, 2 * mesh.normals.size() * sizeof(NormalType), (const gpu::Byte*) normalsAndTangents.data());
+                mesh.normalsAndTangents = normalsAndTangents;
             }
-            _blendedVertexBuffers.push_back(buffer);
+            i++;
         }
         needFullUpdate = true;
         emit rigReady();
@@ -1245,20 +1257,21 @@ Blender::Blender(ModelPointer model, int blendNumber, const Geometry::WeakPointe
 
 void Blender::run() {
     DETAILED_PROFILE_RANGE_EX(simulation_animation, __FUNCTION__, 0xFFFF0000, 0, { { "url", _model->getURL().toString() } });
-    QVector<glm::vec3> vertices, normals, tangents;
+    QVector<glm::vec3> vertices;
+    QVector<NormalType> normalsAndTangents;
     if (_model) {
         int offset = 0;
+        int normalsAndTangentsOffset = 0;
         foreach (const FBXMesh& mesh, _meshes) {
             if (mesh.blendshapes.isEmpty()) {
                 continue;
             }
             vertices += mesh.vertices;
-            normals += mesh.normals;
+            normalsAndTangents += mesh.normalsAndTangents;
-            tangents += mesh.tangents;
             glm::vec3* meshVertices = vertices.data() + offset;
-            glm::vec3* meshNormals = normals.data() + offset;
+            NormalType* meshNormalsAndTangents = normalsAndTangents.data() + normalsAndTangentsOffset;
-            glm::vec3* meshTangents = tangents.data() + offset;
             offset += mesh.vertices.size();
+            normalsAndTangentsOffset += mesh.normalsAndTangents.size();
             const float NORMAL_COEFFICIENT_SCALE = 0.01f;
             for (int i = 0, n = qMin(_blendshapeCoefficients.size(), mesh.blendshapes.size()); i < n; i++) {
                 float vertexCoefficient = _blendshapeCoefficients.at(i);
@@ -1268,22 +1281,39 @@ void Blender::run() {
                 }
                 float normalCoefficient = vertexCoefficient * NORMAL_COEFFICIENT_SCALE;
                 const FBXBlendshape& blendshape = mesh.blendshapes.at(i);
-                for (int j = 0; j < blendshape.indices.size(); j++) {
+                tbb::parallel_for(tbb::blocked_range<int>(0, blendshape.indices.size()), [&](const tbb::blocked_range<int>& range) {
-                    int index = blendshape.indices.at(j);
+                    for (auto j = range.begin(); j < range.end(); j++) {
-                    meshVertices[index] += blendshape.vertices.at(j) * vertexCoefficient;
+                        int index = blendshape.indices.at(j);
-                    meshNormals[index] += blendshape.normals.at(j) * normalCoefficient;
+                        meshVertices[index] += blendshape.vertices.at(j) * vertexCoefficient;
-                    if (blendshape.tangents.size() > j) {
+
-                        meshTangents[index] += blendshape.tangents.at(j) * normalCoefficient;
+                        glm::vec3 normal = mesh.normals.at(index) + blendshape.normals.at(j) * normalCoefficient;
+                        glm::vec3 tangent;
+                        if (index < mesh.tangents.size()) {
+                            tangent = mesh.tangents.at(index);
+                            if ((int)j < blendshape.tangents.size()) {
+                                tangent += blendshape.tangents.at(j) * normalCoefficient;
+                            }
+                        }
+#if FBX_PACK_NORMALS
+                        glm::uint32 finalNormal;
+                        glm::uint32 finalTangent;
+                        buffer_helpers::packNormalAndTangent(normal, tangent, finalNormal, finalTangent);
+#else
+                        const auto& finalNormal = normal;
+                        const auto& finalTangent = tangent;
+#endif
+                        meshNormalsAndTangents[2 * index] = finalNormal;
+                        meshNormalsAndTangents[2 * index + 1] = finalTangent;
                     }
-                }
+                });
             }
         }
     }
-    // post the result to the geometry cache, which will dispatch to the model if still alive
+    // post the result to the ModelBlender, which will dispatch to the model if still alive
     QMetaObject::invokeMethod(DependencyManager::get<ModelBlender>().data(), "setBlendedVertices",
-        Q_ARG(ModelPointer, _model), Q_ARG(int, _blendNumber),
+                              Q_ARG(ModelPointer, _model), Q_ARG(int, _blendNumber),
-        Q_ARG(const Geometry::WeakPointer&, _geometry), Q_ARG(const QVector<glm::vec3>&, vertices),
+                              Q_ARG(const Geometry::WeakPointer&, _geometry), Q_ARG(const QVector<glm::vec3>&, vertices),
-        Q_ARG(const QVector<glm::vec3>&, normals), Q_ARG(const QVector<glm::vec3>&, tangents));
+                              Q_ARG(const QVector<NormalType>&, normalsAndTangents));
 }
 
 void Model::setScaleToFit(bool scaleToFit, const glm::vec3& dimensions, bool forceRescale) {
@@ -1443,9 +1473,10 @@ void Model::updateClusterMatrices() {
     }
 
     // post the blender if we're not currently waiting for one to finish
-    if (geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
+    auto modelBlender = DependencyManager::get<ModelBlender>();
+    if (modelBlender->shouldComputeBlendshapes() && geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
         _blendedBlendshapeCoefficients = _blendshapeCoefficients;
-        DependencyManager::get<ModelBlender>()->noteRequiresBlend(getThisPointer());
+        modelBlender->noteRequiresBlend(getThisPointer());
     }
 }
 
@@ -1462,83 +1493,31 @@ bool Model::maybeStartBlender() {
 }
 
 void Model::setBlendedVertices(int blendNumber, const Geometry::WeakPointer& geometry,
-        const QVector<glm::vec3>& vertices, const QVector<glm::vec3>& normals, const QVector<glm::vec3>& tangents) {
+        const QVector<glm::vec3>& vertices, const QVector<NormalType>& normalsAndTangents) {
     auto geometryRef = geometry.lock();
-    if (!geometryRef || _renderGeometry != geometryRef || _blendedVertexBuffers.empty() || blendNumber < _appliedBlendNumber) {
+    if (!geometryRef || _renderGeometry != geometryRef || blendNumber < _appliedBlendNumber) {
         return;
     }
     _appliedBlendNumber = blendNumber;
     const FBXGeometry& fbxGeometry = getFBXGeometry();
     int index = 0;
-    std::vector<NormalType> normalsAndTangents;
+    int normalAndTangentIndex = 0;
     for (int i = 0; i < fbxGeometry.meshes.size(); i++) {
         const FBXMesh& mesh = fbxGeometry.meshes.at(i);
         if (mesh.blendshapes.isEmpty()) {
             continue;
         }
 
-        gpu::BufferPointer& buffer = _blendedVertexBuffers[i];
         const auto vertexCount = mesh.vertices.size();
         const auto verticesSize = vertexCount * sizeof(glm::vec3);
-        const auto offset = index * sizeof(glm::vec3);
+        const auto& buffer = _blendedVertexBuffers[i];
-
+        assert(buffer);
-        normalsAndTangents.clear();
+        buffer->resize(mesh.vertices.size() * sizeof(glm::vec3) + mesh.normalsAndTangents.size() * sizeof(NormalType));
-        normalsAndTangents.resize(normals.size()+tangents.size());
+        buffer->setSubData(0, verticesSize, (gpu::Byte*) vertices.constData() + index * sizeof(glm::vec3));
-//        assert(normalsAndTangents.size() == 2 * vertexCount);
+        buffer->setSubData(verticesSize, 2 * mesh.normals.size() * sizeof(NormalType), (const gpu::Byte*) normalsAndTangents.data() + normalAndTangentIndex * sizeof(NormalType));
-
-        // Interleave normals and tangents
-#if 0
-        // Sequential version for debugging
-        auto normalsRange = std::make_pair(normals.begin() + index, normals.begin() + index + vertexCount);
-        auto tangentsRange = std::make_pair(tangents.begin() + index, tangents.begin() + index + vertexCount);
-        auto normalsAndTangentsIt = normalsAndTangents.begin();
-
-        for (auto normalIt = normalsRange.first, tangentIt = tangentsRange.first;
-             normalIt != normalsRange.second;
-             ++normalIt, ++tangentIt) {
-#if FBX_PACK_NORMALS
-            glm::uint32 finalNormal;
-            glm::uint32 finalTangent;
-            buffer_helpers::packNormalAndTangent(*normalIt, *tangentIt, finalNormal, finalTangent);
-#else
-            const auto finalNormal = *normalIt;
-            const auto finalTangent = *tangentIt;
-#endif
-            *normalsAndTangentsIt = finalNormal;
-            ++normalsAndTangentsIt;
-            *normalsAndTangentsIt = finalTangent;
-            ++normalsAndTangentsIt;
-        }
-#else
-        // Parallel version for performance
-        tbb::parallel_for(tbb::blocked_range<size_t>(index, index+vertexCount), [&](const tbb::blocked_range<size_t>& range) {
-            auto normalsRange = std::make_pair(normals.begin() + range.begin(), normals.begin() + range.end());
-            auto tangentsRange = std::make_pair(tangents.begin() + range.begin(), tangents.begin() + range.end());
-            auto normalsAndTangentsIt = normalsAndTangents.begin() + (range.begin()-index)*2;
-
-            for (auto normalIt = normalsRange.first, tangentIt = tangentsRange.first;
-                 normalIt != normalsRange.second;
-                 ++normalIt, ++tangentIt) {
-#if FBX_PACK_NORMALS
-                glm::uint32 finalNormal;
-                glm::uint32 finalTangent;
-                buffer_helpers::packNormalAndTangent(*normalIt, *tangentIt, finalNormal, finalTangent);
-#else
-                const auto finalNormal = *normalIt;
-                const auto finalTangent = *tangentIt;
-#endif
-                *normalsAndTangentsIt = finalNormal;
-                ++normalsAndTangentsIt;
-                *normalsAndTangentsIt = finalTangent;
-                ++normalsAndTangentsIt;
-            }
-        });
-#endif
-
-        buffer->setSubData(0, verticesSize, (gpu::Byte*) vertices.constData() + offset);
-        buffer->setSubData(verticesSize, 2 * vertexCount * sizeof(NormalType), (const gpu::Byte*) normalsAndTangents.data());
 
         index += vertexCount;
+        normalAndTangentIndex += 2 * mesh.normals.size();
     }
 }
 
@@ -1604,6 +1583,7 @@ void Model::createRenderItemSet() {
     // Run through all of the meshes, and place them into their segregated, but unsorted buckets
     int shapeID = 0;
     uint32_t numMeshes = (uint32_t)meshes.size();
+    auto fbxGeometry = getFBXGeometry();
     for (uint32_t i = 0; i < numMeshes; i++) {
         const auto& mesh = meshes.at(i);
         if (!mesh) {
@@ -1613,6 +1593,9 @@ void Model::createRenderItemSet() {
         // Create the render payloads
         int numParts = (int)mesh->getNumParts();
         for (int partIndex = 0; partIndex < numParts; partIndex++) {
+            if (fbxGeometry.meshes[i].blendshapes.empty() && !_blendedVertexBuffers[i]) {
+                _blendedVertexBuffers[i] = std::make_shared<gpu::Buffer>();
+            }
             _modelMeshRenderItems << std::make_shared<ModelMeshPartPayload>(shared_from_this(), i, partIndex, shapeID, transform, offset);
             auto material = getGeometry()->getShapeMaterial(shapeID);
             _modelMeshMaterialNames.push_back(material ? material->getName() : "");
@@ -1725,11 +1708,10 @@ void ModelBlender::noteRequiresBlend(ModelPointer model) {
     }
 }
 
-void ModelBlender::setBlendedVertices(ModelPointer model, int blendNumber, const Geometry::WeakPointer& geometry, 
+void ModelBlender::setBlendedVertices(ModelPointer model, int blendNumber, const Geometry::WeakPointer& geometry,
-                                      const QVector<glm::vec3>& vertices, const QVector<glm::vec3>& normals, 
+                                      const QVector<glm::vec3>& vertices, const QVector<NormalType>& normalsAndTangents) {
-                                      const QVector<glm::vec3>& tangents) {
     if (model) {
-        model->setBlendedVertices(blendNumber, geometry, vertices, normals, tangents);
+        model->setBlendedVertices(blendNumber, geometry, vertices, normalsAndTangents);
     }
     _pendingBlenders--;
     {
@@ -1745,4 +1727,3 @@ void ModelBlender::setBlendedVertices(ModelPointer model, int blendNumber, const
         }
     }
 }
-

libraries/render-utils/src/Model.h

@@ -135,7 +135,7 @@ public:
 
     /// Sets blended vertices computed in a separate thread.
     void setBlendedVertices(int blendNumber, const Geometry::WeakPointer& geometry,
-        const QVector<glm::vec3>& vertices, const QVector<glm::vec3>& normals, const QVector<glm::vec3>& tangents);
+        const QVector<glm::vec3>& vertices, const QVector<NormalType>& normalsAndTangents);
 
     bool isLoaded() const { return (bool)_renderGeometry && _renderGeometry->isGeometryLoaded(); }
     bool isAddedToScene() const { return _addedToScene; }
@@ -413,7 +413,7 @@ protected:
 
     QUrl _url;
 
-    gpu::Buffers _blendedVertexBuffers;
+    std::unordered_map<int, gpu::BufferPointer> _blendedVertexBuffers;
 
     QVector<QVector<QSharedPointer<Texture> > > _dilatedTextures;
 
@@ -503,9 +503,12 @@ public:
     /// Adds the specified model to the list requiring vertex blends.
     void noteRequiresBlend(ModelPointer model);
 
+    bool shouldComputeBlendshapes() { return _computeBlendshapes; }
+
 public slots:
     void setBlendedVertices(ModelPointer model, int blendNumber, const Geometry::WeakPointer& geometry,
-        const QVector<glm::vec3>& vertices, const QVector<glm::vec3>& normals, const QVector<glm::vec3>& tangents);
+        const QVector<glm::vec3>& vertices, const QVector<NormalType>& normalsAndTangents);
+    void setComputeBlendshapes(bool computeBlendshapes) { _computeBlendshapes = computeBlendshapes; }
 
 private:
     using Mutex = std::mutex;
@@ -517,6 +520,8 @@ private:
     std::set<ModelWeakPointer, std::owner_less<ModelWeakPointer>> _modelsRequiringBlends;
     int _pendingBlenders;
     Mutex _mutex;
+
+    bool _computeBlendshapes { true };
 };
 
 

libraries/render-utils/src/SoftAttachmentModel.cpp

@@ -77,8 +77,9 @@ void SoftAttachmentModel::updateClusterMatrices() {
     }
 
     // post the blender if we're not currently waiting for one to finish
-    if (geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
+    auto modelBlender = DependencyManager::get<ModelBlender>();
+    if (modelBlender->shouldComputeBlendshapes() && geometry.hasBlendedMeshes() && _blendshapeCoefficients != _blendedBlendshapeCoefficients) {
         _blendedBlendshapeCoefficients = _blendshapeCoefficients;
-        DependencyManager::get<ModelBlender>()->noteRequiresBlend(getThisPointer());
+        modelBlender->noteRequiresBlend(getThisPointer());
     }
 }