From Time Moore, "This submission implements 3 optimizations for meshes. INDEX_MESH turns DrawArrays style geometry into DrawElements, uniquifying the vertices in the process. This is useful for certain loaders, like ac3d, which just spit out DrawArrays. VERTEX_POSTTRANSFORM and VERTEX_PRETRANSFORM optimize mesh triangle and vertex order for the caches on a modern GPU, using Tom Forsyth's algorithm. I describe this and the big difference it makes (38% improvement on a very large mesh) in my blog,

http://shiny-dynamics.blogspot.com/2010/03/vertex-cache-optimization-for-osg.html."

include/osgUtil/MeshOptimizers

@@ -0,0 +1,101 @@
+/* -*-c++-*- OpenSceneGraph - Copyright (C) 1998-2006 Robert Osfield 
+ *
+ * This library is open source and may be redistributed and/or modified under  
+ * the terms of the OpenSceneGraph Public License (OSGPL) version 0.0 or 
+ * (at your option) any later version.  The full license is in LICENSE file
+ * included with this distribution, and on the openscenegraph.org website.
+ * 
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the 
+ * OpenSceneGraph Public License for more details.
+*/
+
+#ifndef OSGUTIL_MESHOPTIMIZERS
+#define OSGUTIL_MESHOPTIMIZERS 1
+
+#include <set>
+#include <vector>
+
+#include <osg/Geode>
+#include <osg/Geometry>
+#include <osg/NodeVisitor>
+
+#include <osgUtil/Optimizer>
+
+namespace osgUtil
+{
+// Helper that collects all the unique Geometry objects in a subgraph.
+class OSGUTIL_EXPORT GeometryCollector : public BaseOptimizerVisitor
+{
+public:
+    GeometryCollector(Optimizer* optimizer,
+                      Optimizer::OptimizationOptions options)
+        : BaseOptimizerVisitor(optimizer, options) {}
+    void reset();
+    void apply(osg::Geode& geode);
+    typedef std::set<osg::Geometry*> GeometryList;
+    GeometryList& getGeometryList() { return _geometryList; };
+protected:
+    GeometryList _geometryList;
+};
+
+// Convert geometry that uses DrawArrays to DrawElements i.e.,
+// construct a real mesh. This removes duplicate vertices.
+class OSGUTIL_EXPORT IndexMeshVisitor : public GeometryCollector
+{
+public:
+    IndexMeshVisitor(Optimizer* optimizer = 0)
+        : GeometryCollector(optimizer, Optimizer::INDEX_MESH)
+    {
+    }
+    void makeMesh(osg::Geometry& geom);
+    void makeMesh();
+};
+
+// Optimize the triangle order in a mesh for best use of the GPU's
+// post-transform cache. This uses Tom Forsyth's algorithm described
+// at http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html
+class OSGUTIL_EXPORT VertexCacheVisitor : public GeometryCollector
+{
+public:
+    VertexCacheVisitor(Optimizer* optimizer = 0)
+        : GeometryCollector(optimizer, Optimizer::VERTEX_POSTTRANSFORM)
+    {
+    }
+
+    void optimizeVertices(osg::Geometry& geom);
+    void optimizeVertices();
+private:
+    void doVertexOptimization(osg::Geometry& geom,
+                              std::vector<unsigned>& vertDrawList);
+};
+
+// Gather statistics on post-transform cache misses for geometry
+class OSGUTIL_EXPORT VertexCacheMissVisitor : public osg::NodeVisitor
+{
+public:
+    VertexCacheMissVisitor(unsigned cacheSize = 16);
+    void reset();
+    virtual void apply(osg::Geode& geode);
+    void doGeometry(osg::Geometry& geom);
+    unsigned misses;
+    unsigned triangles;
+protected:
+    const unsigned _cacheSize;
+};
+
+// Optimize the use of the GPU pre-transform cache by arranging vertex
+// attributes in the order they are used.
+class OSGUTIL_EXPORT VertexAccessOrderVisitor : public GeometryCollector
+{
+public:
+    VertexAccessOrderVisitor(Optimizer* optimizer = 0)
+        : GeometryCollector(optimizer, Optimizer::VERTEX_PRETRANSFORM)
+    {
+    }
+    void optimizeOrder();
+    void optimizeOrder(osg::Geometry& geom);
+};
+}
+#endif

include/osgUtil/Optimizer

@@ -85,6 +85,9 @@ class OSGUTIL_EXPORT Optimizer
             TEXTURE_ATLAS_BUILDER =     (1 << 15),
             STATIC_OBJECT_DETECTION =   (1 << 16),
             FLATTEN_STATIC_TRANSFORMS_DUPLICATING_SHARED_SUBGRAPHS = (1 << 17),
+            INDEX_MESH =                (1 << 18),
+            VERTEX_POSTTRANSFORM =      (1 << 19),
+            VERTEX_PRETRANSFORM =       (1 << 20),
             DEFAULT_OPTIMIZATIONS = FLATTEN_STATIC_TRANSFORMS |
                                 REMOVE_REDUNDANT_NODES |
                                 REMOVE_LOADED_PROXY_NODES |

src/osgUtil/CMakeLists.txt

@@ -24,6 +24,7 @@ SET(LIB_PUBLIC_HEADERS
     ${HEADER_PATH}/IntersectVisitor
     ${HEADER_PATH}/IncrementalCompileOperation
     ${HEADER_PATH}/LineSegmentIntersector
+    ${HEADER_PATH}/MeshOptimizers
     ${HEADER_PATH}/OperationArrayFunctor
     ${HEADER_PATH}/Optimizer
     ${HEADER_PATH}/PlaneIntersector
@@ -67,6 +68,7 @@ ADD_LIBRARY(${LIB_NAME}
     IntersectVisitor.cpp
     IncrementalCompileOperation.cpp
     LineSegmentIntersector.cpp
+    MeshOptimizers.cpp
     Optimizer.cpp
     PlaneIntersector.cpp
     PolytopeIntersector.cpp

src/osgUtil/Optimizer.cpp

@@ -39,6 +39,7 @@
 #include <osgUtil/TriStripVisitor>
 #include <osgUtil/Tessellator>
 #include <osgUtil/Statistics>
+#include <osgUtil/MeshOptimizers>
 
 #include <typeinfo>
 #include <algorithm>
@@ -54,7 +55,7 @@ void Optimizer::reset()
 {
 }
 
-static osg::ApplicationUsageProxy Optimizer_e0(osg::ApplicationUsage::ENVIRONMENTAL_VARIABLE,"OSG_OPTIMIZER \"<type> [<type>]\"","OFF | DEFAULT | FLATTEN_STATIC_TRANSFORMS | FLATTEN_STATIC_TRANSFORMS_DUPLICATING_SHARED_SUBGRAPHS | REMOVE_REDUNDANT_NODES | COMBINE_ADJACENT_LODS | SHARE_DUPLICATE_STATE | MERGE_GEOMETRY | MERGE_GEODES | SPATIALIZE_GROUPS  | COPY_SHARED_NODES  | TRISTRIP_GEOMETRY | OPTIMIZE_TEXTURE_SETTINGS | REMOVE_LOADED_PROXY_NODES | TESSELLATE_GEOMETRY | CHECK_GEOMETRY |  FLATTEN_BILLBOARDS | TEXTURE_ATLAS_BUILDER | STATIC_OBJECT_DETECTION");
+static osg::ApplicationUsageProxy Optimizer_e0(osg::ApplicationUsage::ENVIRONMENTAL_VARIABLE,"OSG_OPTIMIZER \"<type> [<type>]\"","OFF | DEFAULT | FLATTEN_STATIC_TRANSFORMS | FLATTEN_STATIC_TRANSFORMS_DUPLICATING_SHARED_SUBGRAPHS | REMOVE_REDUNDANT_NODES | COMBINE_ADJACENT_LODS | SHARE_DUPLICATE_STATE | MERGE_GEOMETRY | MERGE_GEODES | SPATIALIZE_GROUPS  | COPY_SHARED_NODES  | TRISTRIP_GEOMETRY | OPTIMIZE_TEXTURE_SETTINGS | REMOVE_LOADED_PROXY_NODES | TESSELLATE_GEOMETRY | CHECK_GEOMETRY |  FLATTEN_BILLBOARDS | TEXTURE_ATLAS_BUILDER | STATIC_OBJECT_DETECTION | INDEX_MESH | VERTEX_POSTTRANSFORM | VERTEX_PRETRANSFORM");
 
 void Optimizer::optimize(osg::Node* node)
 {
@@ -125,6 +126,15 @@ void Optimizer::optimize(osg::Node* node)
         if(str.find("~STATIC_OBJECT_DETECTION")!=std::string::npos) options ^= STATIC_OBJECT_DETECTION;
         else if(str.find("STATIC_OBJECT_DETECTION")!=std::string::npos) options |= STATIC_OBJECT_DETECTION;
 
+        if(str.find("~INDEX_MESH")!=std::string::npos) options ^= INDEX_MESH;
+        else if(str.find("INDEX_MESH")!=std::string::npos) options |= INDEX_MESH;
+
+        if(str.find("~VERTEX_POSTTRANSFORM")!=std::string::npos) options ^= VERTEX_POSTTRANSFORM;
+        else if(str.find("VERTEX_POSTTRANSFORM")!=std::string::npos) options |= VERTEX_POSTTRANSFORM;
+
+        if(str.find("~VERTEX_PRETRANSFORM")!=std::string::npos) options ^= VERTEX_PRETRANSFORM;
+        else if(str.find("VERTEX_PRETRANSFORM")!=std::string::npos) options |= VERTEX_PRETRANSFORM;
+
     }
     else
     {
@@ -347,6 +357,30 @@ void Optimizer::optimize(osg::Node* node, unsigned int options)
         sv.divide();
     }
 
+    if (options & INDEX_MESH)
+    {
+        OSG_NOTIFY(osg::INFO)<<"Optimizer::optimize() doing INDEX_MESH"<<std::endl;
+        IndexMeshVisitor imv(this);
+        node->accept(imv);
+        imv.makeMesh();
+    }
+
+    if (options & VERTEX_POSTTRANSFORM)
+    {
+        OSG_NOTIFY(osg::INFO)<<"Optimizer::optimize() doing VERTEX_POSTTRANSFORM"<<std::endl;
+        VertexCacheVisitor vcv;
+        node->accept(vcv);
+        vcv.optimizeVertices();
+    }
+
+    if (options & VERTEX_PRETRANSFORM)
+    {
+        OSG_NOTIFY(osg::INFO)<<"Optimizer::optimize() doing VERTEX_PRETRANSFORM"<<std::endl;
+        VertexAccessOrderVisitor vaov;
+        node->accept(vaov);
+        vaov.optimizeOrder();
+    }
+
     if (osg::getNotifyLevel()>=osg::INFO)
     {
         stats.reset();