gpuCache/gpuCacheCmd.cpp

gpuCache/gpuCacheCmd.cpp
//-
//**************************************************************************/
// Copyright 2015 Autodesk, Inc. All rights reserved.
//
// Use of this software is subject to the terms of the Autodesk
// license agreement provided at the time of installation or download,
// or which otherwise accompanies this software in either electronic
// or hard copy form.
//**************************************************************************/
//+
//
// gpuCache MEL command
//
// Creates one or more cache files on disk to store attribute data for
// a span of frames.
//
#include "gpuCacheCmd.h"
#include "gpuCacheShapeNode.h"
#include "gpuCacheStrings.h"
#include "gpuCacheUtil.h"
#include "gpuCacheConfig.h"
#include "gpuCacheVBOProxy.h"
#include "gpuCacheVramQuery.h"
#include "gpuCacheMaterialBakers.h"
#include "gpuCacheSubSceneOverride.h"
#include "gpuCacheUnitBoundingBox.h"
#include "CacheWriter.h"
#include "CacheReader.h"
#include "gpuCacheGeometry.h"
#include <maya/MArgList.h>
#include <maya/MAnimControl.h>
#include <maya/MDagPathArray.h>
#include <maya/MFnDagNode.h>
#include <maya/MFnMesh.h>
#include <maya/MFnMeshData.h>
#include <maya/MFnNurbsSurface.h>
#include <maya/MFnSet.h>
#include <maya/MFnSubd.h>
#include <maya/MGlobal.h>
#include <maya/MPlugArray.h>
#include <maya/MSyntax.h>
#include <maya/MBoundingBox.h>
#include <maya/MFileObject.h>
#include <maya/MItDag.h>
#include <maya/MVector.h>
#include <maya/MFnTransform.h>
#include <maya/MFnSingleIndexedComponent.h>
#include <maya/MFnComponentListData.h>
#include <maya/MFnLambertShader.h>
#include <maya/MViewport2Renderer.h>
#include <maya/MDagPath.h>
#include <maya/MPointArray.h>
#include <maya/MUintArray.h>
#include <boost/foreach.hpp>
#include <boost/shared_ptr.hpp>
#include <boost/scoped_ptr.hpp>
#include <boost/unordered_set.hpp>
#include <cfloat>
#include <limits>
#include <list>
#include <iomanip>
#include <map>
#include <sstream>
#include <fstream>
#define MStatError(status,msg) \
if ( MS::kSuccess != (status) ) { \
MPxCommand::displayError( \
(msg) + MString(":") + (status).errorString()); \
return (status); \
}
#define MStatErrorNullObj(status,msg) \
if ( MS::kSuccess != (status) ) { \
MPxCommand::displayError( \
(msg) + MString(":") + (status).errorString()); \
return MObject::kNullObj; \
}
#define MCheckReturn(expression) \
{ \
MStatus status = (expression); \
if ( MS::kSuccess != (status) ) { \
return (status); \
} \
}
#define MUpdateProgressAndCheckInterruption(progressBar) \
{ \
(progressBar).stepProgress(); \
if ((progressBar).isCancelled()) { \
return MS::kFailure; \
} \
} \
namespace {
using namespace GPUCache;
//==============================================================================
// LOCAL FUNCTIONS
//==============================================================================
// Create a cache writer object that will write to the specified file path.
// If the directory does not exist, a new one will be created.
// If the file already exists, the existing file will be deleted. (Overwrite)
// Writing to a read-only file will return an error.
// Args:
// targetFile The target file path that the new writer will write files to.
// compressLevel Hint the compress level: -1 (Store), 0~9 (Fastest~Best).
// dataFormat Hint the file format: ogawa or hdf.
boost::shared_ptr<CacheWriter> createWriter(const MFileObject& targetFile,
const char compressLevel,
const MString& dataFormat)
{
// Get the directory of the target file.
MFileObject cacheDirectory;
cacheDirectory.setRawFullName(targetFile.resolvedPath());
// Make sure the cache folder exists.
if (!cacheDirectory.exists()) {
// Create the cache folder.
MString createFolderCmd;
createFolderCmd.format("sysFile -md \"^1s\"", EncodeString(targetFile.resolvedPath()));
MGlobal::executeCommand(createFolderCmd);
}
// Delete the existing file.
// We have already confirmed that the file is going to be overwritten.
if (MFileObject(targetFile).exists()) {
// The file already exists!
MString resolvedFullName = targetFile.resolvedFullName();
// Check if the file is writeable.
bool writeable;
{
std::ofstream ofs(resolvedFullName.asChar());
writeable = ofs.is_open();
}
// We can't overwrite a read-only file.
if (!writeable) {
MStatus stat;
MString fmt = MStringResource::getString(kCouldNotSaveFileMsg, stat);
MString msg;
msg.format(fmt, resolvedFullName);
return boost::shared_ptr<CacheWriter>();
}
// We are going to overwrite the file. Delete it!!
if (remove(resolvedFullName.asChar()) != 0) {
MStatus stat;
MString fmt = MStringResource::getString(kCouldNotSaveFileMsg, stat);
MString msg;
msg.format(fmt, resolvedFullName);
return boost::shared_ptr<CacheWriter>();
}
}
// first parameter is the file to write
// second parameter is gzip compress level, -1 or 0~9
// third parameter is data format, hdf or ogawa
boost::shared_ptr<CacheWriter> cacheWriter =
CacheWriter::create("Alembic", targetFile, compressLevel, dataFormat);
if (!cacheWriter) {
MStatus stat;
MString msg = MStringResource::getString(kCreateCacheWriterErrorMsg, stat);
return boost::shared_ptr<CacheWriter>();
}
if (!cacheWriter->valid()) {
// release the file handle
cacheWriter.reset();
MString errorMsg;
errorMsg.format("Couldn't open cache file: ^1s", targetFile.resolvedFullName());
return boost::shared_ptr<CacheWriter>();
}
return cacheWriter;
}
bool isPlugConnectedToTexture2d(const MPlug& plug)
{
MPlugArray connections;
if (plug.connectedTo(connections, true, false)) {
assert(connections.length() == 1);
//return false immediately if connections is empty, in order to fix the crash MAYA-41542
if(connections.length() == 0)
return false;
MObject srcNode = connections[0].node();
return srcNode.hasFn(MFn::kTexture2d);
}
return false;
}
MColor getTexture2dDefaultColor(const MPlug& plug)
{
MPlugArray connections;
if (plug.connectedTo(connections, true, false)) {
assert(connections.length() == 1);
//return immediately if connections is empty
if(connections.length() == 0)
return MColor(0.5, 0.5, 0.5);
MFnDependencyNode srcNode(connections[0].node());
MPlug diffusePlugR = srcNode.findPlug("defaultColorR");
MPlug diffusePlugG = srcNode.findPlug("defaultColorG");
MPlug diffusePlugB = srcNode.findPlug("defaultColorB");
assert(!diffusePlugR.isNull());
assert(!diffusePlugG.isNull());
assert(!diffusePlugB.isNull());
MStatus statusR, statusG, statusB;
float r = diffusePlugR.asFloat(MDGContext::fsNormal, &statusR);
float g = diffusePlugG.asFloat(MDGContext::fsNormal, &statusG);
float b = diffusePlugB.asFloat(MDGContext::fsNormal, &statusB);
assert(statusR == MS::kSuccess);
assert(statusG == MS::kSuccess);
assert(statusB == MS::kSuccess);
return MColor(r, g, b);
}
return MColor(0.5, 0.5, 0.5);
}
bool isPlugConnectedToTextureNode(const MPlug& plug)
{
MPlugArray connections;
if (plug.connectedTo(connections, true, false)) {
assert(connections.length() == 1);
//return false immediately if connections is empty
if(connections.length() == 0)
return false;
MObject srcNode = connections[0].node();
if (srcNode.hasFn(MFn::kTexture2d) ||
srcNode.hasFn(MFn::kTexture3d) ||
srcNode.hasFn(MFn::kTextureEnv) ||
return true;
}
}
return false;
}
MStatus getShapeDiffuseColors(const std::vector<MDagPath>& paths,
std::vector<MColor>& diffuseColors)
{
MStatus status;
diffuseColors.resize(paths.size(), Config::kDefaultGrayColor);
// Get the diffuse color for each instance
for (size_t pathIndex = 0; pathIndex < paths.size(); pathIndex++) {
MFnDagNode shape(paths[pathIndex], &status);
assert(status == MS::kSuccess);
MObject shadingGroup;
MObject shaderObj;
// Find the instObjGroups plug
MPlug instObjectGroupsParent = shape.findPlug("instObjGroups");
assert(!instObjectGroupsParent.isNull());
MPlug instObjectGroups = instObjectGroupsParent.elementByLogicalIndex(
paths[pathIndex].instanceNumber());
assert(!instObjectGroups.isNull());
// instObjGroups is connected, the whole shape is assigned a material
if (instObjectGroups.isConnected()) {
// instObjGroups[instanceNumber] -> shadingGroup
MPlugArray dstPlugs;
instObjectGroups.connectedTo(dstPlugs, false, true, &status);
if (status && dstPlugs.length() > 0) {
// Found shadingGroup assigned to the whole shape
shadingGroup = dstPlugs[0].node();
}
}
// For per-component shader assignment, we use the first shading group.
// Find the objectGroups plug
MPlug objectGroupsParent = instObjectGroups.child(0);
assert(!objectGroupsParent.isNull());
for (unsigned int parts = 0;
parts < objectGroupsParent.numElements() && shadingGroup.isNull();
parts++) {
MPlug objectGroups = objectGroupsParent[parts];
// objectGroups is connected, there is per-component material
if (objectGroups.isConnected()) {
// objectGroups[i] -> shadingGroup
MPlugArray dstPlugs;
objectGroups.connectedTo(dstPlugs, false, true, &status);
if (status && dstPlugs.length() > 0) {
// Found shadingGroup assigned to components
shadingGroup = dstPlugs[0].node();
}
}
} // for each objectGroup plug
if (!shadingGroup.isNull()) {
// Found a shading group, find its surface shader
MFnDependencyNode shadingEngine(shadingGroup, &status);
assert(status == MS::kSuccess);
// Find surfaceShader plug
MPlug surfaceShaderPlug = shadingEngine.findPlug("surfaceShader");
assert(!surfaceShaderPlug.isNull());
// outColor -> surfaceShader
if (surfaceShaderPlug.isConnected()) {
MPlugArray srcPlugs;
surfaceShaderPlug.connectedTo(srcPlugs, true, false, &status);
if (status && srcPlugs.length() > 0) {
// Found the material node
shaderObj = srcPlugs[0].node();
}
}
}
if (!shaderObj.isNull()) {
MColor diffuseColor = Config::kDefaultGrayColor;
MColor transparency = Config::kDefaultTransparency;
// Found a material node, get its color
if (shaderObj.hasFn(MFn::kLambert)) {
MFnLambertShader lambert(shaderObj, &status);
assert(status == MS::kSuccess);
MPlug colorPlug = lambert.findPlug("color");
assert(!colorPlug.isNull());
MPlug diffusePlug = lambert.findPlug("diffuse");
assert(!diffusePlug.isNull());
MPlug transparencyPlug = lambert.findPlug("transparency");
assert(!transparencyPlug.isNull());
if (isPlugConnectedToTexture2d(colorPlug)) {
diffuseColor = getTexture2dDefaultColor(colorPlug);
}
else if (!isPlugConnectedToTextureNode(colorPlug)) {
diffuseColor = lambert.color();
}
if (!isPlugConnectedToTextureNode(diffusePlug)) {
diffuseColor *= lambert.diffuseCoeff();
}
if (!isPlugConnectedToTextureNode(transparencyPlug)) {
transparency = lambert.transparency();
}
}
// Transparency RGB Luminance as alpha
diffuseColor.a = 1.0f - (transparency.r * 0.3f +
transparency.g * 0.59f + transparency.b * 0.11f);
diffuseColors[pathIndex] = diffuseColor;
}
}
return MS::kSuccess;
}
MString getSceneName()
{
"basenameEx(`file -q -sceneName`)");
if (sceneName.length() == 0) {
sceneName = MGlobal::executeCommandStringResult("untitledFileName");
}
return sceneName;
}
MString getSceneNameAsValidObjectName()
{
return MGlobal::executeCommandStringResult("formValidObjectName \"" + EncodeString(getSceneName()) + "\"");
}
size_t maxNumVerts(const ShapeData::Ptr& geom)
{
size_t maxNumVerts = 0;
BOOST_FOREACH(
const ShapeData::SampleMap::value_type& smv, geom->getSamples()
) {
maxNumVerts = std::max(maxNumVerts, smv.second->numVerts());
}
return maxNumVerts;
}
double toHumanUnits(MUint64 bytes, MString& units)
{
const MUint64 KB = 1024;
const MUint64 MB = 1024 * KB;
const MUint64 GB = 1024 * MB;
const MUint64 TB = 1024 * GB;
double value;
if (bytes >= TB) {
units = "TB";
value = double(bytes)/TB;
}
else if (bytes >= GB) {
units = "GB";
value = double(bytes)/GB;
}
else if (bytes >= MB) {
units = "MB";
value = double(bytes)/MB;
}
else if (bytes >= KB) {
units = "KB";
value = double(bytes)/KB;
}
else {
units = "bytes";
value = double(bytes);
}
return value;
}
//==============================================================================
// CLASS Baker
//==============================================================================
class Baker
{
public:
static boost::shared_ptr<Baker> create(const MObject& object,
const std::vector<MDagPath>& paths);
static bool isBakeable(const MObject& dagNode);
Baker(const MObject& object, const std::vector<MDagPath>& paths)
: fNode(object), fPaths(paths)
{}
virtual ~Baker() {}
virtual MStatus sample(const MTime& time) = 0;
virtual const SubNode::MPtr getNode(size_t instIndex) const = 0;
virtual void setWriteMaterials() {}
virtual void setUseBaseTessellation() {}
protected:
MFnDagNode fNode;
std::vector<MDagPath> fPaths;
};
//==============================================================================
// CLASS ShapeBaker
//==============================================================================
// Base class for wrappers that encapsulate the logic necessary to
// bake particular types of shapes (meshes, nurbs, subds, etc.).
class ShapeBaker : public Baker
{
public:
virtual ~ShapeBaker() {}
void enableUVs()
{
fCacheMeshSampler->enableUVs();
}
// The function is called to sample the geometry at the specified time
virtual MStatus sample(const MTime& time)
{
// Sample the shape
MCheckReturn( sampleTopologyAndAttributes() );
// Sample the diffuse color
std::vector<MColor> diffuseColors;
MCheckReturn( getShapeDiffuseColors(fPaths, diffuseColors) );
bool diffuseColorsAnimated = (fPrevDiffuseColors != diffuseColors);
// add sample to geometry
if (fCacheMeshSampler->isAnimated() || diffuseColorsAnimated) {
for (size_t i = 0; i < fGeometryInstances.size(); i++) {
fGeometryInstances[i]->addSample(
fCacheMeshSampler->getSample(
diffuseColors[i]));
}
}
fPrevDiffuseColors.swap(diffuseColors);
return MS::kSuccess;
}
// The function is called at the end of baking process to get the baked geometry
virtual const SubNode::MPtr getNode(size_t instIndex) const
{
return SubNode::create(
fNode.name(),
fGeometryInstances[instIndex]
);
}
protected:
ShapeBaker(const MObject& node, const std::vector<MDagPath>& paths)
: Baker(node, paths),
fCacheMeshSampler(
// note: the UVs can always get enabled later by calling our method enableUVs()
CacheMeshSampler::create(!Config::isIgnoringUVs())),
fGeometryInstances(paths.size())
{
// Create one geometry for each instance
for (size_t i = 0; i < paths.size(); i++) {
fGeometryInstances[i] = ShapeData::create();
}
}
virtual void setWriteMaterials()
{
// Create one geometry for each instance
for (size_t i = 0; i < fPaths.size(); i++) {
// Set material to the shape data.
MString surfaceMaterial;
InstanceMaterialLookup lookup(fPaths[i]);
if (lookup.hasWholeObjectMaterial()) {
// Whole object material assignment.
MObject material = lookup.findWholeObjectSurfaceMaterial();
if (!material.isNull()) {
MFnDependencyNode dgMaterial(material);
surfaceMaterial = dgMaterial.name();
}
}
else if (lookup.hasComponentMaterials()) {
// Per-component material assignment.
std::vector<MObject> materials;
lookup.findSurfaceMaterials(materials);
// Use the first surface material
// TODO: Support per-component material assignment.
BOOST_FOREACH (const MObject& material, materials) {
if (!material.isNull()) {
MFnDependencyNode dgMaterial(material);
surfaceMaterial = dgMaterial.name();
break;
}
}
}
if (surfaceMaterial.length() > 0) {
fGeometryInstances[i]->setMaterial(surfaceMaterial);
}
}
}
virtual void setUseBaseTessellation()
{
fCacheMeshSampler->setUseBaseTessellation();
}
virtual MStatus sampleTopologyAndAttributes() = 0;
private:
// Forbidden and not implemented.
ShapeBaker(const ShapeBaker&);
const ShapeBaker& operator=(const ShapeBaker&);
protected:
const boost::shared_ptr<CacheMeshSampler> fCacheMeshSampler;
std::vector<MColor> fPrevDiffuseColors;
std::vector<ShapeData::MPtr> fGeometryInstances;
};
//==============================================================================
// CLASS XformBaker
//==============================================================================
// Class for baking a transform MObject
class XformBaker : public Baker
{
public:
virtual ~XformBaker()
{}
XformBaker(const MObject& xformNode, const std::vector<MDagPath>& xformPaths)
: Baker(xformNode, xformPaths),
fCacheXformSamplers(CacheXformSampler::create(xformNode)),
fXformInstances(xformPaths.size())
{
for (size_t i = 0; i < fXformInstances.size(); i++) {
fXformInstances[i] = XformData::create();
}
}
virtual MStatus sample(const MTime& currentTime)
{
fCacheXformSamplers->addSample();
if (fCacheXformSamplers->isAnimated()) {
for (size_t i = 0; i < fXformInstances.size(); i++) {
fXformInstances[i]->addSample(
fCacheXformSamplers->getSample(currentTime.as(MTime::kSeconds)));
}
}
return MS::kSuccess;
}
virtual const SubNode::MPtr getNode(size_t instIndex) const
{
return SubNode::create(
fNode.name(),
fXformInstances[instIndex]
);
}
private:
boost::shared_ptr<CacheXformSampler> fCacheXformSamplers;
std::vector<XformData::MPtr> fXformInstances;
};
//==============================================================================
// CLASS MeshDataBaker
//==============================================================================
// Base class for baking a mesh MObject
class MeshDataBaker : public ShapeBaker
{
public:
virtual ~MeshDataBaker() {}
protected:
virtual MStatus sampleTopologyAndAttributes()
{
MStatus status;
MObject meshData = getMeshData(&status);
MStatError(status, "getMeshData()");
bool shapeVisibility = ShapeVisibilityChecker(fNode.object()).isVisible();
// Snapshot the topology and vertex attributes.
return fCacheMeshSampler->addSample(meshData, shapeVisibility) ?
}
virtual MObject getMeshData(MStatus* status) = 0;
MeshDataBaker(const MObject& shapeNode, const std::vector<MDagPath>& shapePaths)
: ShapeBaker(shapeNode, shapePaths)
{}
private:
// Forbidden and not implemented.
MeshDataBaker(const MeshDataBaker&);
const MeshDataBaker& operator=(const MeshDataBaker&);
};
//==============================================================================
// CLASS MeshBaker
//==============================================================================
class MeshBaker : public ShapeBaker
{
public:
MeshBaker(const MObject& meshNode, const std::vector<MDagPath>& meshPaths)
: ShapeBaker(meshNode, meshPaths), fMeshNode(meshNode)
{}
virtual ~MeshBaker() {}
protected:
virtual MStatus sampleTopologyAndAttributes()
{
return fCacheMeshSampler->addSampleFromMesh(fMeshNode) ?
}
private:
// Forbidden and not implemented.
MeshBaker(const MeshBaker&);
const MeshBaker& operator=(const MeshBaker&);
MFnMesh fMeshNode;
};
//==============================================================================
// CLASS NurbsBaker
//==============================================================================
class NurbsBaker : public MeshDataBaker
{
public:
NurbsBaker(const MObject& nurbsNode, const std::vector<MDagPath>& nurbsPaths)
: MeshDataBaker(nurbsNode, nurbsPaths)
{
// Disable Viewport 2.0 updates while baking NURBS surfaces.
}
protected:
virtual MObject getMeshData(MStatus* status)
{
MObject mesh;
MDGModifier modifier;
MFnNurbsSurface nurbsNode(fNode.object());
MObject tessellator = modifier.createNode("nurbsTessellate");
MFnDependencyNode tessellatorNode(tessellator);
modifier.connect(nurbsNode.findPlug("explicitTessellationAttributes"),
tessellatorNode.findPlug("explicitTessellationAttributes"));
modifier.connect(nurbsNode.findPlug("curvatureTolerance"),
tessellatorNode.findPlug("curvatureTolerance"));
modifier.connect(nurbsNode.findPlug("uDivisionsFactor"),
tessellatorNode.findPlug("uDivisionsFactor"));
modifier.connect(nurbsNode.findPlug("vDivisionsFactor"),
tessellatorNode.findPlug("vDivisionsFactor"));
modifier.connect(nurbsNode.findPlug("modeU"),
tessellatorNode.findPlug("uType"));
modifier.connect(nurbsNode.findPlug("modeV"),
tessellatorNode.findPlug("vType"));
modifier.connect(nurbsNode.findPlug("numberU"),
tessellatorNode.findPlug("uNumber"));
modifier.connect(nurbsNode.findPlug("numberV"),
tessellatorNode.findPlug("vNumber"));
modifier.connect(nurbsNode.findPlug("useChordHeight"),
tessellatorNode.findPlug("useChordHeight"));
modifier.connect(nurbsNode.findPlug("useChordHeightRatio"),
tessellatorNode.findPlug("useChordHeightRatio"));
modifier.connect(nurbsNode.findPlug("chordHeight"),
tessellatorNode.findPlug("chordHeight"));
modifier.connect(nurbsNode.findPlug("chordHeightRatio"),
tessellatorNode.findPlug("chordHeightRatio"));
modifier.connect(nurbsNode.findPlug("smoothEdge"),
tessellatorNode.findPlug("smoothEdge"));
modifier.connect(nurbsNode.findPlug("smoothEdgeRatio"),
tessellatorNode.findPlug("smoothEdgeRatio"));
modifier.connect(nurbsNode.findPlug("edgeSwap"),
tessellatorNode.findPlug("edgeSwap"));
modifier.connect(nurbsNode.findPlug("local"),
tessellatorNode.findPlug("inputSurface"));
// poly type - 0 means triangles
modifier.newPlugValueInt(tessellatorNode.findPlug("polygonType"),0);
// format - 2 means general fit
modifier.newPlugValueInt(tessellatorNode.findPlug("format"),2);
modifier.doIt();
tessellatorNode.findPlug("outputPolygon").getValue(mesh);
modifier.undoIt();
return mesh;
}
};
//==============================================================================
// CLASS SubdBaker
//==============================================================================
class SubdBaker : public MeshDataBaker
{
public:
SubdBaker(const MObject& subdNode, const std::vector<MDagPath>& subdPaths)
: MeshDataBaker(subdNode, subdPaths)
{}
protected:
virtual MObject getMeshData(MStatus* status)
{
MFnSubd subdNode(fNode.object());
MFnMeshData meshData;
meshData.create(status);
if (!*status) return meshData.object();
int format = -1;
int depth = -1;
int sampleCount = -1;
MPlug formatPlug = subdNode.findPlug("format");
MPlug depthPlug = subdNode.findPlug("depth");
MPlug sampleCountPlug = subdNode.findPlug("sampleCount");
formatPlug.getValue(format);
depthPlug.getValue(depth);
sampleCountPlug.getValue(sampleCount);
subdNode.tesselate(
format==0, depth, sampleCount, meshData.object(), status);
return meshData.object();
}
};
//==============================================================================
// CLASS RecursiveBaker
//==============================================================================
// This class simply extracts the hierarchy from gpuCache node.
class RecursiveBaker : public Baker
{
public:
RecursiveBaker(const MObject& shapeNode, const std::vector<MDagPath>& shapePaths)
: Baker(shapeNode, shapePaths)
{
// Find the user node
MPxNode* userNode = fNode.userNode();
assert(userNode);
ShapeNode* bakedNode = userNode ?
dynamic_cast<ShapeNode*>(userNode) : NULL;
assert(bakedNode);
// Extract the baked geometry
if (bakedNode) {
GlobalReaderCache::theCache().waitForRead(bakedNode->getCacheFileEntry().get());
fSrcTopNode = bakedNode->getCachedGeometry();
if (fSrcTopNode) {
fSampleReplicator.reset(new SampleReplicator);
fSrcTopNode->accept(*fSampleReplicator);
}
}
}
virtual ~RecursiveBaker()
{}
virtual MStatus sample(const MTime& time)
{
if (!fSrcTopNode) {
return MS::kFailure;
}
return fSampleReplicator->sample(time);
}
virtual const SubNode::MPtr getNode(size_t instIndex) const
{
// We ignore the material assigned to the gpuCache node.
if (fSrcTopNode && !fDstTopNode) {
// We replicate the hierarchy after all xform/shape data are
// filled with samples.
HierarchyReplicator hierarchyReplicator(fSampleReplicator);
fSrcTopNode->accept(hierarchyReplicator);
RecursiveBaker* nonConstThis = const_cast<RecursiveBaker*>(this);
nonConstThis->fDstTopNode = hierarchyReplicator.dstSubNode();
}
return fDstTopNode;
}
private:
// Forbidden and not implemented.
RecursiveBaker(const RecursiveBaker&);
const RecursiveBaker& operator=(const RecursiveBaker&);
class SampleReplicator : public SubNodeVisitor
{
public:
typedef boost::shared_ptr<SampleReplicator> MPtr;
SampleReplicator()
{}
virtual void visit(const XformData& srcXform,
const SubNode& srcSubNode)
{
// Create a new xform data, it will be filled later in sample()
XformData::MPtr dstXform = XformData::create();
fXforms[&srcXform] = std::make_pair(
dstXform, boost::shared_ptr<const XformSample>());
// Recursively replicate xform/shape data in the child hierarchy
BOOST_FOREACH(const SubNode::Ptr& srcChild, srcSubNode.getChildren()) {
srcChild->accept(*this);
}
}
virtual void visit(const ShapeData& srcShape,
const SubNode& srcSubNode)
{
// Create a new shape data, it will be filled later in sample()
ShapeData::MPtr dstShape = ShapeData::create();
dstShape->setMaterials(srcShape.getMaterials());
fShapes[&srcShape] = std::make_pair(
dstShape, boost::shared_ptr<const ShapeSample>());
}
MStatus sample(const MTime& time)
{
BOOST_FOREACH(XformMapping::value_type& xform, fXforms) {
// Get the already baked sample
boost::shared_ptr<const XformSample> srcXformSample =
xform.first->getSample(time);
// Only add the sample if it's different than prev sample
if (xform.second.second != srcXformSample) {
// Create a new sample with the same content but different time
boost::shared_ptr<XformSample> dstXformSample =
XformSample::create(
srcXformSample->xform(),
srcXformSample->boundingBox(),
srcXformSample->visibility());
xform.second.first->addSample(dstXformSample);
xform.second.second = srcXformSample;
}
}
BOOST_FOREACH(ShapeMapping::value_type& shape, fShapes) {
// Get the already baked sample
boost::shared_ptr<const ShapeSample> srcShapeSample =
shape.first->getSample(time);
// Only add the sample if it's different than prev sample
if (shape.second.second != srcShapeSample) {
// Create a new sample with the same content but different time
boost::shared_ptr<ShapeSample> dstShapeSample =
ShapeSample::create(
srcShapeSample->numWires(),
srcShapeSample->numVerts(),
srcShapeSample->wireVertIndices(),
srcShapeSample->triangleVertexIndexGroups(),
srcShapeSample->positions(),
srcShapeSample->boundingBox(),
srcShapeSample->diffuseColor(),
srcShapeSample->visibility());
if (srcShapeSample->normals()) {
dstShapeSample->setNormals(srcShapeSample->normals());
}
if (srcShapeSample->uvs()) {
dstShapeSample->setUVs(srcShapeSample->uvs());
}
shape.second.first->addSample(dstShapeSample);
shape.second.second = srcShapeSample;
}
}
return MS::kSuccess;
}
XformData::MPtr xform(const XformData& xform)
{
XformMapping::iterator iter = fXforms.find(&xform);
assert(iter != fXforms.end());
return (*iter).second.first;
}
ShapeData::MPtr shape(const ShapeData& shape)
{
ShapeMapping::iterator iter = fShapes.find(&shape);
assert(iter != fShapes.end());
return (*iter).second.first;
}
private:
// Forbidden and not implemented.
SampleReplicator(const SampleReplicator&);
const SampleReplicator& operator=(const SampleReplicator&);
typedef std::pair<XformData::MPtr,boost::shared_ptr<const XformSample> > XformWithPrev;
typedef std::pair<ShapeData::MPtr,boost::shared_ptr<const ShapeSample> > ShapeWithPrev;
typedef std::map<const XformData*,XformWithPrev> XformMapping;
typedef std::map<const ShapeData*,ShapeWithPrev> ShapeMapping;
XformMapping fXforms;
ShapeMapping fShapes;
};
class HierarchyReplicator : public SubNodeVisitor
{
public:
HierarchyReplicator(SampleReplicator::MPtr sampleReplicator)
: fSampleReplicator(sampleReplicator)
{}
virtual void visit(const XformData& srcXform,
const SubNode& srcSubNode)
{
// Create a new sub node for the xform
// We rename "|" to "top" as we don't want "|" to appear in hierarchy.
XformData::MPtr dstXform = fSampleReplicator->xform(srcXform);
fDstSubNode = SubNode::create(srcSubNode.getName() != "|" ?
srcSubNode.getName() : "top", dstXform);
// Recursively replicate the child hierarchy
BOOST_FOREACH(const SubNode::Ptr& srcChild, srcSubNode.getChildren())
{
HierarchyReplicator replicator(fSampleReplicator);
srcChild->accept(replicator);
SubNode::connect(fDstSubNode, replicator.dstSubNode());
}
}
virtual void visit(const ShapeData& srcShape,
const SubNode& srcSubNode)
{
// Create a new sub node for the shape
ShapeData::MPtr dstShape = fSampleReplicator->shape(srcShape);
fDstSubNode = SubNode::create(srcSubNode.getName(), dstShape);
}
SubNode::MPtr dstSubNode() const
{ return fDstSubNode; }
private:
// Forbidden and not implemented.
HierarchyReplicator(const HierarchyReplicator&);
const HierarchyReplicator& operator=(const HierarchyReplicator&);
SampleReplicator::MPtr fSampleReplicator;
SubNode::MPtr fDstSubNode;
};
SubNode::Ptr fSrcTopNode;
SubNode::MPtr fDstTopNode;
SampleReplicator::MPtr fSampleReplicator;
};
//==============================================================================
// CLASS Baker
//==============================================================================
bool Baker::isBakeable(const MObject& dagNode)
{
if (dagNode.hasFn(MFn::kTransform)
|| dagNode.hasFn(MFn::kMesh)
|| dagNode.hasFn(MFn::kSubdiv))
{
return true;
}
return false;
}
boost::shared_ptr<Baker> Baker::create(const MObject& shapeNode,
const std::vector<MDagPath>& shapePaths)
{
if (shapeNode.hasFn(MFn::kTransform)) {
return boost::make_shared<XformBaker>(shapeNode, shapePaths);
}
else if (shapeNode.hasFn(MFn::kMesh)) {
return boost::make_shared<MeshBaker>(shapeNode, shapePaths);
}
else if (shapeNode.hasFn(MFn::kNurbsSurface)) {
return boost::make_shared<NurbsBaker>(shapeNode, shapePaths);
}
else if (shapeNode.hasFn(MFn::kSubdiv)) {
return boost::make_shared<SubdBaker>(shapeNode, shapePaths);
}
MStatus status;
MFnDagNode shape(shapeNode, &status);
assert(status == MS::kSuccess);
if (shape.typeId() == ShapeNode::id) {
return boost::make_shared<RecursiveBaker>(shapeNode, shapePaths);
}
assert(false);
return boost::shared_ptr<Baker>();
}
//==============================================================================
// CLASS Writer
//==============================================================================
class Writer : boost::noncopyable
{
public:
Writer(const char compressLevel,
const MString& dataFormat,
const MTime& timePerCycle,
const MTime& startTime)
: fCompressLevel(compressLevel),
fDataFormat(dataFormat),
fTimePerCycleInSeconds(timePerCycle.as(MTime::kSeconds)),
fStartTimeInSeconds(startTime.as(MTime::kSeconds))
{}
// Write a sub-node hierarchy to the specified file.
MStatus writeNode(const SubNode::Ptr& subNode,
const MaterialGraphMap::Ptr& materials,
const MFileObject& targetFile)
{
boost::shared_ptr<CacheWriter> writer = createWriter(
targetFile,
fCompressLevel,
fDataFormat
);
if (!writer) {
return MS::kFailure;
}
writer->writeSubNodeHierarchy(
subNode,
fTimePerCycleInSeconds, fStartTimeInSeconds);
if (materials) {
writer->writeMaterials(
materials,
fTimePerCycleInSeconds, fStartTimeInSeconds);
}
return MS::kSuccess;
}
// Write a list of sub-node hierarchies to the specified file.
MStatus writeNodes(const std::vector<SubNode::Ptr>& subNodes,
const MaterialGraphMap::Ptr& materials,
const MFileObject& targetFile)
{
boost::shared_ptr<CacheWriter> writer = createWriter(
targetFile,
fCompressLevel,
fDataFormat
);
if (!writer) {
return MS::kFailure;
}
BOOST_FOREACH(const SubNode::Ptr& subNode, subNodes) {
writer->writeSubNodeHierarchy(
subNode,
fTimePerCycleInSeconds,
fStartTimeInSeconds);
}
if (materials) {
writer->writeMaterials(materials,
fTimePerCycleInSeconds, fStartTimeInSeconds);
}
return MS::kSuccess;
}
private:
const char fCompressLevel;
const MString fDataFormat;
const double fTimePerCycleInSeconds;
const double fStartTimeInSeconds;
};
//==========================================================================
// CLASS Stat
//==========================================================================
class Stat
{
public:
Stat(MUint64 bytesPerUnit)
: fMin(std::numeric_limits<MUint64>::max()),
fMax(0),
fTotal(0),
fInstancedTotal(0),
fBytesPerUnit(bytesPerUnit)
{}
void addSample(
const boost::shared_ptr<const IndexBuffer> buffer, const int indicesPerElem)
{
addSample(buffer->numIndices() / indicesPerElem, buffer.get());
};
void addSample(
const boost::shared_ptr<const VertexBuffer> buffer)
{
addSample(buffer->numVerts(), buffer.get());
};
void addSample(
const MHWRender::MIndexBuffer* buffer, size_t numIndices)
{
addSample(numIndices, (void*)buffer);
}
void addSample(
const MHWRender::MVertexBuffer* buffer, size_t numVertices)
{
addSample(numVertices, (void*)buffer);
}
void addSample(
const boost::shared_ptr<const VBOBuffer> buffer, size_t numPrimitives)
{
addSample(numPrimitives, buffer.get());
}
MUint64 getNbSamples() const { return fUniqueEntries.size(); }
MUint64 getMin() const { return fMin;}
MUint64 getMax() const { return fMin;}
MUint64 getTotal() const { return fTotal;}
MUint64 getInstancedTotal() const { return fInstancedTotal;}
double getAverage() const {
return double(getTotal())/double(getNbSamples());
}
MUint64 getSize() const {
return fTotal * fBytesPerUnit;
}
MString print(MString name) const
{
MStatus status;
MString result;
if (getNbSamples() == 0) {
MString msg;
msg.format(
MStringResource::getString(kStatsZeroBuffersMsg, status),
name);
result = msg;
}
else {
MString memUnit;
double memSize = toHumanUnits(getSize(), memUnit);
MString msg;
MString msg_buffers; msg_buffers += (double)getNbSamples();
MString msg_avrg; msg_avrg += (double)getAverage();
MString msg_min; msg_min += (double)fMin;
MString msg_max; msg_max += (double)fMax;
MString msg_total; msg_total += (double)fTotal;
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kStatsBuffersMsg, status),
name, msg_buffers, msg_avrg,
msg_min, msg_max, msg_total, msg_memSize, memUnit);
result = msg;
}
return result;
}
private:
void addSample(MUint64 value, const void* buffer)
{
if (fUniqueEntries.insert(buffer).second) {
fMin = std::min(fMin, value);
fMax = std::max(fMax, value);
fTotal += value;
}
fInstancedTotal += value;
};
boost::unordered_set<const void*> fUniqueEntries;
MUint64 fMin;
MUint64 fMax;
MUint64 fTotal;
const MUint64 fBytesPerUnit;
// Total number of instanced geometry.
MUint64 fInstancedTotal;
};
//==========================================================================
// CLASS Stats
//==========================================================================
class Stats
{
public:
Stats()
: fNbNodes(0),
fNbSubNodes(0),
fWires( 2 * sizeof(IndexBuffer::index_t)),
fTriangles( 3 * sizeof(IndexBuffer::index_t)),
fVerts( 3 * sizeof(float)),
fNormals( 3 * sizeof(float)),
fUVs( 2 * sizeof(float)),
fVP2Index( sizeof(IndexBuffer::index_t)),
fVP2Vertex( sizeof(float)),
fVBOIndex( sizeof(IndexBuffer::index_t)),
fVBOVertex( sizeof(float)),
fNbMaterialGraphs(0),
fNbMaterialNodes(0)
{}
void accumulateNode()
{
++fNbNodes;
}
void accumulateMaterialGraph(const MaterialGraph::Ptr&)
{
++fNbMaterialGraphs;
}
void accumulateMaterialNode(const MaterialNode::Ptr&)
{
++fNbMaterialNodes;
}
void accumulate(const ShapeData& shape)
{
++fNbSubNodes;
BOOST_FOREACH(const ShapeData::SampleMap::value_type& v,
shape.getSamples()) {
accumSample(v.second);
}
}
void accumulate(const ShapeData& shape,
MTime time)
{
++fNbSubNodes;
accumSample(shape.getSample(time));
}
void print(MStringArray& result, bool printInstancedInfo) const
{
MStatus status;
{
MString msg;
MString msg_nbGeom; msg_nbGeom += fNbNodes;
MString msg_nbSubNodes; msg_nbSubNodes += fNbSubNodes;
msg.format(
MStringResource::getString(kStatsNbGeomMsg, status),
msg_nbGeom, msg_nbSubNodes);
result.append(msg);
}
result.append(fWires.print( MStringResource::getString(kStatsWiresMsg, status)));
result.append(fTriangles.print(MStringResource::getString(kStatsTrianglesMsg, status)));
result.append(fVerts.print( MStringResource::getString(kStatsVerticesMsg, status)));
result.append(fNormals.print( MStringResource::getString(kStatsNormalsMsg, status)));
result.append(fUVs.print( MStringResource::getString(kStatsUVsMsg, status)));
if (printInstancedInfo) {
MString msg;
MString msgInstWires; msgInstWires += (double)fWires.getInstancedTotal();
MString msgInstTris; msgInstTris += (double)fTriangles.getInstancedTotal();
msg.format(
MStringResource::getString(kStatsTotalInstancedMsg, status),
msgInstWires, msgInstTris);
result.append(msg);
}
{
MUint64 totalMem = (fWires.getSize() +
fTriangles.getSize() +
fVerts.getSize() +
fNormals.getSize() +
fUVs.getSize());
MString memUnit;
double memSize = toHumanUnits(totalMem, memUnit);
MString msg;
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kStatsSystemTotalMsg, status),
msg_memSize, memUnit);
result.append(msg);
}
{
MUint64 totalMem = (fVBOIndex.getSize() + fVBOVertex.getSize());
result.append(fVBOIndex.print( MStringResource::getString(kStatsVBOIndexMsg, status)));
result.append(fVBOVertex.print(MStringResource::getString(kStatsVBOVertexMsg, status)));
if (Config::vp2OverrideAPI() != Config::kMPxDrawOverride) {
result.append(fVP2Index.print( MStringResource::getString(kStatsVP2IndexMsg, status)));
result.append(fVP2Vertex.print(MStringResource::getString(kStatsVP2VertexMsg, status)));
totalMem += (fVP2Index.getSize() + fVP2Vertex.getSize());
}
MString memUnit;
double memSize = toHumanUnits(totalMem, memUnit);
MString msg;
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kStatsVideoTotalMsg, status),
msg_memSize, memUnit);
result.append(msg);
}
{
MString msg_nbGraphs;
msg_nbGraphs += fNbMaterialGraphs;
MString msg_nbNodes;
msg_nbNodes += fNbMaterialNodes;
MString msg;
msg.format(
MStringResource::getString(kStatsMaterialsMsg, status),
msg_nbGraphs, msg_nbNodes);
result.append(msg);
}
}
private:
/*----- member functions -----*/
void accumSample(const boost::shared_ptr<const ShapeSample>& sample) {
accumIndexBuffer(fWires, sample->wireVertIndices(), 2);
for(size_t i=0; i<sample->numIndexGroups(); ++i) {
accumIndexBuffer(fTriangles, sample->triangleVertIndices(i), 3);
}
accumVertexBuffer(fVerts, sample->positions());
accumVertexBuffer(fNormals, sample->normals());
accumVertexBuffer(fUVs, sample->uvs());
}
void accumIndexBuffer(
Stat& stat,
const boost::shared_ptr<const IndexBuffer> indexBuffer,
const int indicesPerElem
)
{
if (indexBuffer && indexBuffer != UnitBoundingBox::indices()) {
stat.addSample(indexBuffer, indicesPerElem);
{
SubSceneOverride::lookup(indexBuffer);
if (vp2Buffer)
fVP2Index.addSample(vp2Buffer, indexBuffer->numIndices());
}
{
boost::shared_ptr<const VBOBuffer> vboBuffer =
VBOBuffer::lookup(indexBuffer);
if (vboBuffer)
fVBOIndex.addSample(vboBuffer, indexBuffer->numIndices());
}
}
}
void accumVertexBuffer(
Stat& stat,
const boost::shared_ptr<const VertexBuffer> vertexBuffer
)
{
if (vertexBuffer && vertexBuffer != UnitBoundingBox::positions()) {
stat.addSample(vertexBuffer);
{
SubSceneOverride::lookup(vertexBuffer);
if (vp2Buffer)
fVP2Vertex.addSample(
vp2Buffer, 3 * vertexBuffer->numVerts());
}
{
boost::shared_ptr<const VBOBuffer> vboBuffer =
VBOBuffer::lookup(vertexBuffer);
if (vboBuffer)
fVBOVertex.addSample(
vboBuffer, 3 * vertexBuffer->numVerts());
}
{
boost::shared_ptr<const VBOBuffer> vboBuffer =
VBOBuffer::lookupFlippedNormals(vertexBuffer);
if (vboBuffer)
fVBOVertex.addSample(
vboBuffer, 3 * vertexBuffer->numVerts());
}
}
}
/*----- data members -----*/
int fNbNodes;
int fNbSubNodes;
Stat fWires;
Stat fTriangles;
Stat fVerts;
Stat fNormals;
Stat fUVs;
Stat fVP2Index;
Stat fVP2Vertex;
Stat fVBOIndex;
Stat fVBOVertex;
int fNbMaterialGraphs;
int fNbMaterialNodes;
};
//==========================================================================
// CLASS StatsVisitor
//==========================================================================
class StatsVisitor : public SubNodeVisitor
{
public:
StatsVisitor() : fAtGivenTime(false) {}
StatsVisitor(MTime time) : fAtGivenTime(true), fTime(time) {}
void accumulateNode(const SubNode::Ptr& topNode)
{
fStats.accumulateNode();
if (topNode) {
topNode->accept(*this);
}
}
void accumulateMaterialGraph(const MaterialGraphMap::Ptr& materials)
{
if (materials) {
BOOST_FOREACH (
const MaterialGraphMap::NamedMap::value_type& val,
materials->getGraphs()) {
fStats.accumulateMaterialGraph(val.second);
accumulateMaterialNode(val.second);
}
}
}
void accumulateMaterialNode(const MaterialGraph::Ptr& material)
{
if (material) {
BOOST_FOREACH (
const MaterialGraph::NamedMap::value_type& val,
material->getNodes()) {
fStats.accumulateMaterialNode(val.second);
}
}
}
void print(MStringArray& result, bool printInstancedInfo) const
{
fStats.print(result, printInstancedInfo);
}
private:
virtual void visit(const XformData& /*xform*/,
const SubNode& subNode)
{
// Recurse into children sub nodes. Expand all instances.
BOOST_FOREACH(const SubNode::Ptr& child,
subNode.getChildren() ) {
child->accept(*this);
}
}
virtual void visit(const ShapeData& shape,
const SubNode& /*subNode*/)
{
if (fAtGivenTime) {
fStats.accumulate(shape, fTime);
}
else {
fStats.accumulate(shape);
}
}
const bool fAtGivenTime;
const MTime fTime;
Stats fStats;
};
//==========================================================================
// CLASS DumpHierarchyVisitor
//==========================================================================
class DumpHierarchyVisitor : public SubNodeVisitor
{
public:
DumpHierarchyVisitor(MStringArray& result)
: fResult(result),
fLevel(0)
{}
virtual void visit(const XformData& xform,
const SubNode& subNode)
{
using namespace std;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "xform name = " << subNode.getName().asChar()
<< ", tt = " << subNode.transparentType()
<< ", ptr = " << (void*)&subNode
<< " {" << ends;
fResult.append(MString(tmp.str().c_str()));
}
++fLevel;
{
std::string indent(kIndent*fLevel, ' ');
BOOST_FOREACH(const XformData::SampleMap::value_type& sample,
xform.getSamples()) {
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "time = " << setw(10) << sample.first
<< ", ptr = " << (void*)sample.second.get()
<< ", vis = " << sample.second->visibility()
<< ", bbox = ("
<< setw(8) << sample.second->boundingBox().min().x << ","
<< setw(8) << sample.second->boundingBox().min().y << ","
<< setw(8) << sample.second->boundingBox().min().z << ") - ("
<< setw(8) << sample.second->boundingBox().max().x << ","
<< setw(8) << sample.second->boundingBox().max().y << ","
<< setw(8) << sample.second->boundingBox().max().z << ")"
<< ends;
fResult.append(MString(tmp.str().c_str()));
}
// Recurse into children sub nodes. Expand all instances.
BOOST_FOREACH(const SubNode::Ptr& child,
subNode.getChildren() ) {
child->accept(*this);
}
}
--fLevel;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "}" << ends;
fResult.append(MString(tmp.str().c_str()));
}
}
virtual void visit(const ShapeData& shape,
const SubNode& subNode)
{
using namespace std;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "shape name = " << subNode.getName().asChar()
<< ", tt = " << subNode.transparentType()
<< ", ptr = " << (void*)&subNode
<< " {" << ends;
fResult.append(MString(tmp.str().c_str()));
}
++fLevel;
{
std::string indent(kIndent*fLevel, ' ');
BOOST_FOREACH(const ShapeData::SampleMap::value_type& sample,
shape.getSamples()) {
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "time = " << setw(10) << sample.first
<< ", ptr = " << (void*)sample.second.get()
<< ", vis = " << sample.second->visibility()
<< ", nT = " << sample.second->numTriangles()
<< ", nW = " << sample.second->numWires()
<< ", nV = " << sample.second->numVerts()
<< "," << ends;
fResult.append(MString(tmp.str().c_str()));
}
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "P = " << (void*)sample.second->positions().get()
<< ", N = " << (void*)sample.second->normals().get()
<< "," << ends;
fResult.append(MString(tmp.str().c_str()));
}
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "C = ("
<< setw(8) << sample.second->diffuseColor()[0] << ","
<< setw(8) << sample.second->diffuseColor()[1] << ","
<< setw(8) << sample.second->diffuseColor()[2] << ","
<< setw(8) << sample.second->diffuseColor()[3] << ","
<< "), bbox = ("
<< setw(8) << sample.second->boundingBox().min().x << ","
<< setw(8) << sample.second->boundingBox().min().y << ","
<< setw(8) << sample.second->boundingBox().min().z << ") - ("
<< setw(8) << sample.second->boundingBox().max().x << ","
<< setw(8) << sample.second->boundingBox().max().y << ","
<< setw(8) << sample.second->boundingBox().max().z << ")"
<< ends;
fResult.append(MString(tmp.str().c_str()));
}
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "bbox place holder = " << (sample.second->isBoundingBoxPlaceHolder() ? "yes" : "no")
<< ends;
fResult.append(MString(tmp.str().c_str()));
}
}
}
if (!shape.getMaterials().empty()) {
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "materials = ";
BOOST_FOREACH (const MString& material, shape.getMaterials()) {
tmp << material.asChar() << ' ';
}
tmp << ends;
fResult.append(MString(tmp.str().c_str()));
}
--fLevel;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "}" << ends;
fResult.append(MString(tmp.str().c_str()));
}
}
private:
static const int kIndent = 2;
MStringArray& fResult;
int fLevel;
};
//==========================================================================
// CLASS DumpMaterialVisitor
//==========================================================================
class DumpMaterialVisitor
{
public:
DumpMaterialVisitor(MStringArray& result)
: fResult(result),
fLevel(0)
{}
void dumpMaterials(const MaterialGraphMap::Ptr& materials)
{
using namespace std;
BOOST_FOREACH (const MaterialGraphMap::NamedMap::value_type& val, materials->getGraphs())
{
const MaterialGraph::Ptr& graph = val.second;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "material graph name = " << graph->name().asChar()
<< ", nNodes = " << graph->getNodes().size()
<< ", ptr = " << (void*)graph.get()
<< " {" << ends;
fResult.append(MString(tmp.str().c_str()));
}
++fLevel;
BOOST_FOREACH (const MaterialGraph::NamedMap::value_type& val, graph->getNodes())
{
dumpMaterialNode(val.second);
}
--fLevel;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "}" << ends;
fResult.append(MString(tmp.str().c_str()));
}
}
}
void dumpMaterialNode(const MaterialNode::Ptr& node)
{
using namespace std;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "material node name = " << node->name().asChar()
<< ", type = " << node->type()
<< ", ptr = " << (void*)node.get()
<< " {" << ends;
fResult.append(MString(tmp.str().c_str()));
}
++fLevel;
BOOST_FOREACH (const MaterialNode::PropertyMap::value_type& val, node->properties())
{
dumpMaterialProperty(val.second);
}
--fLevel;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "}" << ends;
fResult.append(MString(tmp.str().c_str()));
}
}
void dumpMaterialProperty(const MaterialProperty::Ptr& prop)
{
using namespace std;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "prop name = " << prop->name().asChar()
<< ", type = " << propertyTypeString(prop)
<< ", ptr = " << (void*)prop.get()
<< " {" << ends;
fResult.append(MString(tmp.str().c_str()));
}
++fLevel;
BOOST_FOREACH (const MaterialProperty::SampleMap::value_type& val, prop->getSamples())
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "time = " << setw(10) << val.first
<< ", value = " << propertyValueString(val.first, prop)
<< ", ptr = " << (void*)val.second.get()
<< ends;
fResult.append(MString(tmp.str().c_str()));
}
const MaterialNode::Ptr srcNode = prop->srcNode();
const MaterialProperty::Ptr srcProp = prop->srcProp();
if (srcNode && srcProp) {
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "src node = " << srcNode->name().asChar()
<< ", src prop = " << srcProp->name().asChar()
<< ends;
fResult.append(MString(tmp.str().c_str()));
}
--fLevel;
{
ostringstream tmp;
tmp << setw(kIndent*fLevel) << ' '
<< "}" << ends;
fResult.append(MString(tmp.str().c_str()));
}
}
std::string propertyTypeString(const MaterialProperty::Ptr& prop)
{
switch (prop->type()) {
case MaterialProperty::kBool: return "bool";
case MaterialProperty::kInt32: return "int32";
case MaterialProperty::kFloat: return "float";
case MaterialProperty::kFloat2: return "float2";
case MaterialProperty::kFloat3: return "float3";
case MaterialProperty::kRGB: return "rgb";
case MaterialProperty::kString: return "string";
default: assert(0); return "unknown";
}
}
std::string propertyValueString(double seconds, const MaterialProperty::Ptr& prop)
{
std::ostringstream tmp;
switch (prop->type()) {
case MaterialProperty::kBool:
tmp << (prop->asBool(seconds) ? "true" : "false");
break;
case MaterialProperty::kInt32:
tmp << prop->asInt32(seconds);
break;
case MaterialProperty::kFloat:
tmp << prop->asFloat(seconds);
break;
case MaterialProperty::kFloat2:
{
float x, y;
prop->asFloat2(seconds, x, y);
tmp << "(" << x << "," << y << ")";
break;
}
case MaterialProperty::kFloat3:
{
float x, y, z;
prop->asFloat3(seconds, x, y, z);
tmp << "(" << x << "," << y << "," << z << ")";
break;
}
case MaterialProperty::kRGB:
{
MColor c = prop->asColor(seconds);
tmp << "rgb(" << c.r << "," << c.g << "," << c.b << ")";
break;
}
case MaterialProperty::kString:
tmp << prop->asString(seconds).asChar();
break;
default:
assert(0);
tmp << "unknown type";
break;
}
return tmp.str();
}
private:
static const int kIndent = 2;
MStringArray& fResult;
int fLevel;
};
//==============================================================================
// CLASS ProgressBar
//==============================================================================
class ProgressBar
{
public:
ProgressBar(const MStringResourceId& msg, unsigned int max)
{
// Display a progress bar if Maya is running in UI mode
fShowProgress = (MGlobal::mayaState() == MGlobal::kInteractive);
reset(msg, max);
}
void reset(const MStringResourceId& msg, unsigned int max)
{
MStatus status;
beginProgress(MStringResource::getString(msg, status), max);
}
~ProgressBar()
{
endProgress();
}
void stepProgress() const
{
if (fShowProgress) {
MGlobal::executeCommand("progressBar -e -s 1 $gMainProgressBar");
}
}
bool isCancelled() const
{
int isCancelled = 0;
if (fShowProgress) {
MGlobal::executeCommand("progressBar -q -ic $gMainProgressBar", isCancelled);
}
if (isCancelled) {
MStatus status;
const MString interruptMsg = MStringResource::getString(kInterruptedMsg, status);
MGlobal::displayInfo(interruptMsg);
return true;
}
return false;
}
private:
// Forbidden and not implemented.
ProgressBar(const ProgressBar&);
const ProgressBar& operator=(const ProgressBar&);
ProgressBar(const MString& msg, unsigned int max)
{
beginProgress(msg, max);
}
void beginProgress(const MString& msg, unsigned int max) const
{
if (fShowProgress) {
MString maxValue, progressBarCmd;
// Progress from 0 to max
if (max <= 0) {
max = 1;
}
maxValue += max;
// Clear previous isCancelled flag
MGlobal::executeCommand("progressBar -e -bp -ii 1 $gMainProgressBar");
MGlobal::executeCommand("progressBar -e -ep $gMainProgressBar");
// Initialize the progress bar
progressBarCmd.format("progressBar -e -bp -ii 1 -st \"^1s\" -max ^2s $gMainProgressBar",
msg, maxValue);
MGlobal::executeCommand(progressBarCmd);
}
}
void endProgress() const
{
if (fShowProgress) {
MGlobal::executeCommand("progressBar -e -ep $gMainProgressBar");
}
}
bool fShowProgress; // whether to show the progress bar
};
//==============================================================================
// CLASS GroupCreator
//==============================================================================
class GroupCreator
{
public:
GroupCreator() {}
~GroupCreator() {}
void addChild(const SubNode::MPtr& childNode)
{
XformData::Ptr childXform = boost::dynamic_pointer_cast<const XformData>(
childNode->getData());
assert(childXform);
if (childXform) {
fChildNodes.push_back(childNode);
fChildXforms.push_back(childXform);
}
}
void group()
{
assert(!fGroup);
fGroup = XformData::create();
// Collect time samples
std::set<double> times;
BOOST_FOREACH(const XformData::Ptr child, fChildXforms) {
BOOST_FOREACH(const XformData::SampleMap::value_type& val, child->getSamples()) {
times.insert(val.first);
}
}
std::set<double>::const_iterator timeIt = times.begin();
std::set<double>::const_iterator timeEnd = times.end();
if (timeIt != timeEnd) {
fGroup->addSample(XformSample::create(
*timeIt,
true));
}
}
SubNode::MPtr getSubNode(const MString& name)
{
SubNode::MPtr subNode = SubNode::create(name, fGroup);
BOOST_FOREACH(const SubNode::MPtr& childNode, fChildNodes) {
SubNode::connect(subNode, childNode);
}
return subNode;
}
private:
// Prohibited and not implemented.
GroupCreator(const GroupCreator&);
const GroupCreator& operator= (const GroupCreator&);
std::vector<SubNode::MPtr> fChildNodes;
std::vector<XformData::Ptr> fChildXforms;
XformData::MPtr fGroup;
};
//==============================================================================
// CLASS XformFreezer
//==============================================================================
class XformFreezer : public SubNodeVisitor
{
public:
typedef std::vector<ShapeData::Ptr> FrozenGeometries;
typedef std::vector<std::pair<XformData::Ptr,ShapeData::Ptr> > AnimatedGeometries;
typedef std::set<double> TimeSet;
XformFreezer(const XformData::Ptr& parentXform,
FrozenGeometries& frozenGeometries,
bool dontFreezeAnimatedObjects,
AnimatedGeometries& animatedGeometries)
: fParentXform(parentXform),
fFrozenGeometries(frozenGeometries),
fDontFreezeAnimatedObjects(dontFreezeAnimatedObjects),
fAnimatedGeometries(animatedGeometries)
{}
virtual void visit(const XformData& xform,
const SubNode& subNode)
{
// Aggregate the list of sample times.
TimeSet times;
BOOST_FOREACH(const XformData::SampleMap::value_type& val,
fParentXform->getSamples() ) {
times.insert(val.first);
}
BOOST_FOREACH(const XformData::SampleMap::value_type& val,
xform.getSamples()) {
times.insert(val.first);
}
// Freeze xform sample
XformData::MPtr frozenXform = XformData::create();
BOOST_FOREACH(const double& time, times) {
// Parent xform sample
boost::shared_ptr<const XformSample> parentSample =
fParentXform->getSample(time);
// Child xform sample
boost::shared_ptr<const XformSample> sample =
xform.getSample(time);
frozenXform->addSample(XformSample::create(
time,
sample->xform() * parentSample->xform(),
MBoundingBox(), // not used
sample->visibility() && parentSample->visibility()));
}
// Recursive into children
BOOST_FOREACH(const SubNode::Ptr& child, subNode.getChildren()) {
XformFreezer xformFreezer(frozenXform, fFrozenGeometries,
fDontFreezeAnimatedObjects, fAnimatedGeometries);
child->accept(xformFreezer);
}
}
virtual void visit(const ShapeData& shape,
const SubNode& subNode)
{
// Don't freeze animated objects for motion blur.
if (fDontFreezeAnimatedObjects) {
// If the shape matches all the following conditions, we don't freeze/consolidate it.
// 1) Any of the parents (direct,indirect) is animated.
// 2) Shape is not animated.
if (fParentXform->getSamples().size() > 1 && shape.getSamples().size() <= 1) {
// Duplicate the xform data.
XformData::MPtr animatedXform = XformData::create();
BOOST_FOREACH (const XformData::SampleMap::value_type& val,
fParentXform->getSamples()) {
animatedXform->addSample(val.second);
}
// Duplicate the shape data.
ShapeData::MPtr animatedShape = ShapeData::create();
BOOST_FOREACH (const ShapeData::SampleMap::value_type& val,
shape.getSamples()) {
animatedShape->addSample(val.second);
}
animatedShape->setMaterials(shape.getMaterials());
// Give up. We don't freeze and consolidate shapes with
// animated xforms.
fAnimatedGeometries.push_back(
std::make_pair(animatedXform, animatedShape));
return;
}
}
// Aggregate the list of sample times.
TimeSet times;
BOOST_FOREACH(const XformData::SampleMap::value_type& val,
fParentXform->getSamples()) {
times.insert(val.first);
}
BOOST_FOREACH(const ShapeData::SampleMap::value_type& val,
shape.getSamples()) {
times.insert(val.first);
}
// Freeze shape sample
ShapeData::MPtr frozenShape = ShapeData::create();
TimeSet::const_iterator it = times.begin();
TimeSet::const_iterator end = times.end();
if (it != end) {
// The first xform and shape sample
boost::shared_ptr<const XformSample> xformSample =
fParentXform->getSample(*it);
boost::shared_ptr<const ShapeSample> shapeSample =
shape.getSample(*it);
// Freeze the shape sample
boost::shared_ptr<const ShapeSample> frozenSample;
if (xformSample->visibility() && shapeSample->visibility()) {
frozenSample = freezeSample(*it, xformSample, shapeSample);
}
else {
frozenSample = ShapeSample::createEmptySample(*it);
}
// Add the frozen shape sample
frozenShape->addSample(frozenSample);
++it;
for (; it != end; ++it) {
// Save the previous sample
boost::shared_ptr<const XformSample> prevXformSample = xformSample;
boost::shared_ptr<const ShapeSample> prevShapeSample = shapeSample;
// The next xform and shape sample
xformSample = fParentXform->getSample(*it);
shapeSample = shape.getSample(*it);
if (xformSample->visibility() && shapeSample->visibility()) {
if (!xformSample->xform().isEquivalent(prevXformSample->xform()) ||
xformSample->visibility() != prevXformSample->visibility() ||
shapeSample->wireVertIndices() != prevShapeSample->wireVertIndices() ||
shapeSample->triangleVertexIndexGroups() != prevShapeSample->triangleVertexIndexGroups() ||
shapeSample->positions() != prevShapeSample->positions() ||
shapeSample->normals() != prevShapeSample->normals() ||
shapeSample->diffuseColor() != prevShapeSample->diffuseColor() ||
shapeSample->visibility() != prevShapeSample->visibility())
{
// Something changed, need to re-freeze the shape sample
frozenSample = freezeSample(*it, xformSample, shapeSample);
}
else {
// Reuse the last freezeSample() result.
boost::shared_ptr<ShapeSample> newFrozenSample =
ShapeSample::create(
*it,
shapeSample->numWires(),
shapeSample->numVerts(),
shapeSample->wireVertIndices(),
shapeSample->triangleVertexIndexGroups(),
frozenSample->positions(),
frozenSample->boundingBox(),
shapeSample->diffuseColor(),
xformSample->visibility() && shapeSample->visibility());
newFrozenSample->setNormals(frozenSample->normals());
newFrozenSample->setUVs(shapeSample->uvs());
frozenSample = newFrozenSample;
}
}
else {
frozenSample = ShapeSample::createEmptySample(*it);
}
// Add the frozen shape sample
frozenShape->addSample(frozenSample);
}
}
frozenShape->setMaterials(shape.getMaterials());
fFrozenGeometries.push_back(frozenShape);
}
private:
boost::shared_ptr<const ShapeSample> freezeSample(
const double time,
const boost::shared_ptr<const XformSample>& xform,
const boost::shared_ptr<const ShapeSample>& shape)
{
const size_t numWires = shape->numWires();
const size_t numVerts = shape->numVerts();
boost::shared_ptr<IndexBuffer> wireVertIndices =
shape->wireVertIndices();
std::vector<boost::shared_ptr<IndexBuffer> > triangleVertexIndexGroups =
shape->triangleVertexIndexGroups();
boost::shared_ptr<VertexBuffer> uvs =
shape->uvs();
MColor diffuseColor = shape->diffuseColor();
bool visibility = shape->visibility() && xform->visibility();
// Check bad polys
if (numWires == 0 || numVerts == 0 || !wireVertIndices ||
triangleVertexIndexGroups.empty()) {
return ShapeSample::createEmptySample(time);
}
boost::shared_ptr<VertexBuffer> positions;
boost::shared_ptr<VertexBuffer> normals;
MBoundingBox boundingBox;
MMatrix xformMatrix = xform->xform();
if (xformMatrix.isEquivalent(MMatrix::identity)) {
// Nothing to bake for an identity transform.
positions = shape->positions();
normals = shape->normals();
boundingBox = shape->boundingBox();
}
else {
float xform[4][4];
float xformIT[4][4];
xformMatrix.get(xform);
xformMatrix.inverse().transpose().get(xformIT);
const bool isReflection = xformMatrix.det3x3() < 0.0;
if (isReflection) {
// Change the winding order of the triangles if
// the matrix contains a reflection along one the
// axis to preserve front facing.
std::vector<boost::shared_ptr<IndexBuffer> > newTriangleVertexIndexGroups;
BOOST_FOREACH(const boost::shared_ptr<IndexBuffer>& srcIdxBuf,
triangleVertexIndexGroups) {
typedef IndexBuffer::index_t index_t;
const size_t numIndices = srcIdxBuf->numIndices();
IndexBuffer::ReadInterfacePtr readable = srcIdxBuf->readableInterface();
const index_t* srcIndices = readable->get();
const boost::shared_array<IndexBuffer::index_t> dstIndices(
new index_t[numIndices]);
for (size_t i=0; i<numIndices; i+=3) {
dstIndices[i + 0] = srcIndices[i + 2];
dstIndices[i + 1] = srcIndices[i + 1];
dstIndices[i + 2] = srcIndices[i + 0];
}
boost::shared_ptr<IndexBuffer> dstIdxBuf(
IndexBuffer::create(
SharedArray<index_t>::create(dstIndices, numIndices)));
newTriangleVertexIndexGroups.push_back(dstIdxBuf);
}
triangleVertexIndexGroups.swap(newTriangleVertexIndexGroups);
}
VertexBuffer::ReadInterfacePtr srcPosRead = shape->positions()->readableInterface();
const float* srcPositions = srcPosRead->get();
VertexBuffer::ReadInterfacePtr srcNormRead = shape->normals()->readableInterface();
const float* srcNormals = srcNormRead->get();
boost::shared_array<float> dstPositions(new float[3 * numVerts]);
boost::shared_array<float> dstNormals(new float[3 * numVerts]);
float minX = +std::numeric_limits<float>::max();
float minY = +std::numeric_limits<float>::max();
float minZ = +std::numeric_limits<float>::max();
float maxX = -std::numeric_limits<float>::max();
float maxY = -std::numeric_limits<float>::max();
float maxZ = -std::numeric_limits<float>::max();
for (size_t i=0; i<numVerts; ++i) {
const float x = srcPositions[3*i + 0];
const float y = srcPositions[3*i + 1];
const float z = srcPositions[3*i + 2];
const float xp =
xform[0][0] * x + xform[1][0] * y + xform[2][0] * z + xform[3][0];
const float yp =
xform[0][1] * x + xform[1][1] * y + xform[2][1] * z + xform[3][1];
const float zp =
xform[0][2] * x + xform[1][2] * y + xform[2][2] * z + xform[3][2];
minX = std::min(xp, minX);
minY = std::min(yp, minY);
minZ = std::min(zp, minZ);
maxX = std::max(xp, maxX);
maxY = std::max(yp, maxY);
maxZ = std::max(zp, maxZ);
dstPositions[3*i + 0] = xp;
dstPositions[3*i + 1] = yp;
dstPositions[3*i + 2] = zp;
const float nx = srcNormals[3*i + 0];
const float ny = srcNormals[3*i + 1];
const float nz = srcNormals[3*i + 2];
dstNormals[3*i + 0] =
xformIT[0][0] * nx + xformIT[1][0] * ny + xformIT[2][0] * nz + xformIT[3][0];
dstNormals[3*i + 1] =
xformIT[0][1] * nx + xformIT[1][1] * ny + xformIT[2][1] * nz + xformIT[3][1];
dstNormals[3*i + 2] =
xformIT[0][2] * nx + xformIT[1][2] * ny + xformIT[2][2] * nz + xformIT[3][2];
}
positions = VertexBuffer::createPositions(
SharedArray<float>::create(dstPositions, 3 * numVerts));
normals = VertexBuffer::createNormals(
SharedArray<float>::create(dstNormals, 3 * numVerts));
boundingBox = MBoundingBox(MPoint(minX, minY, minZ),
MPoint(maxX, maxY, maxZ));
}
boost::shared_ptr<ShapeSample> frozenSample =
ShapeSample::create(
time,
numWires,
numVerts,
wireVertIndices,
triangleVertexIndexGroups,
positions,
boundingBox,
diffuseColor,
visibility);
frozenSample->setNormals(normals);
frozenSample->setUVs(uvs);
return frozenSample;
}
// Prohibited and not implemented.
XformFreezer(const XformFreezer&);
const XformFreezer& operator= (const XformFreezer&);
XformData::Ptr fParentXform;
FrozenGeometries& fFrozenGeometries;
AnimatedGeometries& fAnimatedGeometries;
bool fDontFreezeAnimatedObjects;
};
//==============================================================================
// CLASS ConsolidateBuckets
//==============================================================================
class ConsolidateBuckets
{
public:
struct BucketKey
{
typedef std::map<double,MColor> DiffuseColorMap;
typedef std::map<double,bool> VisibilityMap;
typedef std::map<double,size_t> IndexGroupMap;
typedef std::vector<MString> MaterialsAssignment;
BucketKey(const ShapeData::Ptr& shape)
{
ShapeData::SampleMap::const_iterator it = shape->getSamples().begin();
ShapeData::SampleMap::const_iterator end = shape->getSamples().end();
if (it != end) {
MColor diffuseColor = (*it).second->diffuseColor();
bool visibility = (*it).second->visibility();
size_t indexGroups = (*it).second->numIndexGroups();
fDiffuseColor[(*it).first] = diffuseColor;
fVisibility[(*it).first] = visibility;
fIndexGroup[(*it).first] = indexGroups;
++it;
for (; it != end; ++it) {
MColor prevDiffuseColor = diffuseColor;
bool prevVisibility = visibility;
size_t prevIndexGroups = indexGroups;
MColor diffuseColor = (*it).second->diffuseColor();
bool visibility = (*it).second->visibility();
size_t indexGroups = (*it).second->numIndexGroups();
if (prevDiffuseColor != diffuseColor) {
fDiffuseColor[(*it).first] = diffuseColor;
}
if (prevVisibility != visibility) {
fVisibility[(*it).first] = visibility;
}
if (prevIndexGroups != indexGroups) {
fIndexGroup[(*it).first] = indexGroups;
}
}
}
fMaterials = shape->getMaterials();
}
struct Hash : std::unary_function<BucketKey, std::size_t>
{
std::size_t operator()(const BucketKey& key) const
{
std::size_t seed = 0;
BOOST_FOREACH(const DiffuseColorMap::value_type& val, key.fDiffuseColor) {
boost::hash_combine(seed, val.first);
boost::hash_combine(seed, val.second.r);
boost::hash_combine(seed, val.second.g);
boost::hash_combine(seed, val.second.b);
boost::hash_combine(seed, val.second.a);
}
BOOST_FOREACH(const VisibilityMap::value_type& val, key.fVisibility) {
boost::hash_combine(seed, val.first);
boost::hash_combine(seed, val.second);
}
BOOST_FOREACH(const IndexGroupMap::value_type& val, key.fIndexGroup) {
boost::hash_combine(seed, val.first);
boost::hash_combine(seed, val.second);
}
BOOST_FOREACH(const MString& material, key.fMaterials) {
unsigned int length = material.length();
const char* begin = material.asChar();
size_t hash = boost::hash_range(begin, begin + length);
boost::hash_combine(seed, hash);
}
return seed;
}
};
struct EqualTo : std::binary_function<BucketKey, BucketKey, bool>
{
bool operator()(const BucketKey& x, const BucketKey& y) const
{
return x.fDiffuseColor == y.fDiffuseColor &&
x.fVisibility == y.fVisibility &&
x.fIndexGroup == y.fIndexGroup &&
x.fMaterials == y.fMaterials;
}
};
DiffuseColorMap fDiffuseColor;
VisibilityMap fVisibility;
IndexGroupMap fIndexGroup;
MaterialsAssignment fMaterials;
};
typedef std::multimap<size_t,ShapeData::Ptr> Bucket;
typedef boost::unordered_map<BucketKey,Bucket,BucketKey::Hash,BucketKey::EqualTo> BucketMap;
typedef std::list<Bucket> BucketList;
ConsolidateBuckets(const XformFreezer::FrozenGeometries& shapes)
: fShapes(shapes)
{}
void divide()
{
BOOST_FOREACH(const ShapeData::Ptr& shape , fShapes) {
BucketKey key(shape);
std::pair<BucketMap::iterator,bool> ret =
fBucketMap.insert(std::make_pair(key, Bucket()));
ret.first->second.insert(std::make_pair(maxNumVerts(shape), shape));
}
}
void getBucketList(BucketList& bucketList)
{
bucketList.clear();
BOOST_FOREACH(const BucketMap::value_type& val, fBucketMap) {
bucketList.push_back(val.second);
}
}
private:
// Prohibited and not implemented.
ConsolidateBuckets(const ConsolidateBuckets&);
const ConsolidateBuckets& operator= (const ConsolidateBuckets&);
const XformFreezer::FrozenGeometries& fShapes;
BucketMap fBucketMap;
};
//==============================================================================
// CLASS FirstSampleTime
//==============================================================================
class FirstSampleTime : public SubNodeVisitor
{
public:
FirstSampleTime()
: fTime(0)
{}
virtual void visit(const XformData& xform,
const SubNode& subNode)
{
fTime = xform.getSamples().begin()->first;
}
virtual void visit(const ShapeData& shape,
const SubNode& subNode)
{
fTime = shape.getSamples().begin()->first;
}
double get()
{ return fTime; }
private:
// Prohibited and not implemented.
FirstSampleTime(const FirstSampleTime&);
const FirstSampleTime& operator= (const FirstSampleTime&);
double fTime;
};
//==============================================================================
// CLASS Consolidator
//==============================================================================
class Consolidator
{
public:
Consolidator(SubNode::MPtr rootNode, const int threshold, const bool motionBlur)
: fRootNode(rootNode), fThreshold(threshold), fMotionBlur(motionBlur)
{}
~Consolidator()
{}
MStatus consolidate()
{
// We currently unconditionally expand all instances. This is kind
// of brute force as it assumes that the instances have a low poly
// count so that consolidating them is worthwhile and that also
// the instance count is low so that the data expansion is
// reasonable.
//
// FIXME: Obviously, a more intelligent heuristic would be needed
// at one point.
// Get the time of the first sample, useful when creating new xform
// samples.
double firstSampleTime = 0;
{
FirstSampleTime firstSampleTimeVisitor;
fRootNode->accept(firstSampleTimeVisitor);
firstSampleTime = firstSampleTimeVisitor.get();
}
// Freeze transforms.
XformFreezer::FrozenGeometries frozenGeometries;
XformFreezer::AnimatedGeometries animatedGeometries;
{
// Create an dummy identity xform data as the root of traversal
XformData::MPtr identityXformData = XformData::create();
identityXformData->addSample(XformSample::create(
firstSampleTime,
MBoundingBox(), // not used when freeze transform
true));
// Traversal the hierarchy to freeze transforms
XformFreezer xformFreezer(identityXformData, frozenGeometries,
fMotionBlur, animatedGeometries);
fRootNode->accept(xformFreezer);
}
// Divide shapes into buckets
ConsolidateBuckets::BucketList bucketList;
{
ConsolidateBuckets buckets(frozenGeometries);
buckets.divide();
buckets.getBucketList(bucketList);
}
// Set up consolidation progress bar
ProgressBar progressBar(kOptimizingMsg, (unsigned int)frozenGeometries.size());
// Consolidate each bucket
std::vector<ShapeData::Ptr> newShapes;
std::vector<ShapeData::Ptr> consolidatedShapes;
BOOST_FOREACH(ConsolidateBuckets::Bucket& bucket, bucketList) {
// Consolidate shapes until the bucket becomes empty
while (!bucket.empty()) {
const ConsolidateBuckets::Bucket::iterator largestNode = --bucket.end();
MInt64 numRemainingVerts = fThreshold - largestNode->first;
if (numRemainingVerts < 0) {
// Already too large to be consolidated.
newShapes.push_back(largestNode->second);
bucket.erase(largestNode);
MUpdateProgressAndCheckInterruption(progressBar);
}
else {
// Find nodes that could make up a consolidation group.
consolidatedShapes.push_back(largestNode->second);
bucket.erase(largestNode);
MUpdateProgressAndCheckInterruption(progressBar);
while (numRemainingVerts > 0 && !bucket.empty()) {
ConsolidateBuckets::Bucket::iterator node =
bucket.upper_bound((size_t)numRemainingVerts);
if (node == bucket.begin()) break;
--node;
numRemainingVerts -= (MInt64)node->first;
consolidatedShapes.push_back(node->second);
bucket.erase(node);
MUpdateProgressAndCheckInterruption(progressBar);
}
// Consolidate the consolidation group
consolidateGeometry(newShapes, consolidatedShapes);
}
}
}
// Attach a xform data to each new shape data
std::vector<XformData::Ptr> newXforms;
BOOST_FOREACH(const ShapeData::Ptr& newShape, newShapes) {
XformData::MPtr newXform = XformData::create();
ShapeData::SampleMap::const_iterator it = newShape->getSamples().begin();
ShapeData::SampleMap::const_iterator end = newShape->getSamples().end();
if (it != end) {
newXform->addSample(XformSample::create(
(*it).first,
true));
}
newXforms.push_back(newXform);
}
// Build a vector of all nodes (consolidated + animated nodes).
std::vector<std::pair<XformData::Ptr,ShapeData::Ptr> > finalXformsAndShapes;
for (size_t i = 0; i < newXforms.size(); i++) {
finalXformsAndShapes.push_back(std::make_pair(newXforms[i], newShapes[i]));
}
finalXformsAndShapes.insert(finalXformsAndShapes.end(),
animatedGeometries.begin(), animatedGeometries.end());
// Done
if (finalXformsAndShapes.size() == 1) {
// Only one shape, use its xform node as the consolidation root
SubNode::MPtr xformNode = SubNode::create(
fRootNode->getName(),
finalXformsAndShapes[0].first);
SubNode::MPtr shapeNode = SubNode::create(
fRootNode->getName() + "Shape",
finalXformsAndShapes[0].second);
SubNode::connect(xformNode, shapeNode);
fConsolidatedRootNode = xformNode;
}
else if (finalXformsAndShapes.size() > 1) {
// There are more than one shape
// We create one more xform node as the consolidation root
XformData::MPtr topXform = XformData::create();
std::set<double> times;
for (size_t i = 0; i < finalXformsAndShapes.size(); i++) {
BOOST_FOREACH(const XformData::SampleMap::value_type& val,
finalXformsAndShapes[i].first->getSamples()) {
times.insert(val.first);
}
BOOST_FOREACH(const ShapeData::SampleMap::value_type& val,
finalXformsAndShapes[i].second->getSamples()) {
times.insert(val.first);
}
}
std::set<double>::const_iterator timeIt = times.begin();
std::set<double>::const_iterator timeEnd = times.end();
if (timeIt != timeEnd) {
topXform->addSample(XformSample::create(
*timeIt,
true));
}
SubNode::MPtr topXformNode = SubNode::create(
fRootNode->getName(),
topXform);
// Create shapes' parent xform sub nodes.
// They are children of the consolidation root.
for (size_t i = 0; i < finalXformsAndShapes.size(); i++) {
SubNode::MPtr xformNode = SubNode::create(
fRootNode->getName() + (i + 1),
finalXformsAndShapes[i].first);
SubNode::MPtr shapeNode = SubNode::create(
fRootNode->getName() + "Shape" + (i + 1),
finalXformsAndShapes[i].second);
SubNode::connect(xformNode, shapeNode);
SubNode::connect(topXformNode, xformNode);
}
fConsolidatedRootNode = topXformNode;
}
return MS::kSuccess;
}
SubNode::MPtr consolidatedRootNode()
{ return fConsolidatedRootNode; }
private:
// Prohibited and not implemented.
Consolidator(const Consolidator&);
const Consolidator& operator= (const Consolidator&);
void consolidateGeometry(std::vector<ShapeData::Ptr>& newShapes,
std::vector<ShapeData::Ptr>& consolidatedShapes)
{
// Aggregate the list of sample times.
std::set<double> times;
BOOST_FOREACH(const ShapeData::Ptr& shape, consolidatedShapes) {
BOOST_FOREACH(const ShapeData::SampleMap::value_type& smv,
shape->getSamples()) {
times.insert(smv.first);
}
}
// Consolidated geometry.
ShapeData::MPtr newShape = ShapeData::create();
const size_t nbShapes = consolidatedShapes.size();
std::set<double>::const_iterator timeIt = times.begin();
std::set<double>::const_iterator timeEnd = times.end();
// Consolidate the first sample.
typedef IndexBuffer::index_t index_t;
boost::shared_array<index_t> wireVertIndices;
std::vector<boost::shared_array<index_t> > triangleVertIndices;
boost::shared_array<float> positions;
boost::shared_array<float> normals;
boost::shared_array<float> uvs;
MBoundingBox boundingBox;
MColor diffuseColor;
bool visibility = true;
{
size_t totalWires = 0;
size_t totalVerts = 0;
std::vector<size_t> totalTriangles;
size_t numIndexGroups = 0;
bool uvExists = false;
for (size_t i = 0; i < nbShapes; i++) {
ShapeData::Ptr& shape = consolidatedShapes[i];
const boost::shared_ptr<const ShapeSample>& sample =
shape->getSample(*timeIt);
totalWires += sample->numWires();
totalVerts += sample->numVerts();
if (numIndexGroups == 0) {
// Initialize totalTriangles, assume that
// all shapes has the same number of index groups
numIndexGroups = sample->numIndexGroups();
totalTriangles.resize(numIndexGroups, 0);
diffuseColor = sample->diffuseColor();
visibility = sample->visibility();
}
// Shapes with different number of index groups, diffuseColor and visibility
// should be divided into separate buckets.
assert(numIndexGroups == sample->numIndexGroups());
assert(fabs(diffuseColor.r - sample->diffuseColor().r) < 1e-5);
assert(fabs(diffuseColor.g - sample->diffuseColor().g) < 1e-5);
assert(fabs(diffuseColor.b - sample->diffuseColor().b) < 1e-5);
assert(fabs(diffuseColor.a - sample->diffuseColor().a) < 1e-5);
assert(visibility == sample->visibility());
for (size_t j = 0; j < totalTriangles.size(); j++) {
totalTriangles[j] += sample->numTriangles(j);
}
// Check whether UV exists
if (!uvExists && sample->uvs()) {
uvExists = true;
}
}
wireVertIndices = boost::shared_array<index_t>(
new index_t[2 * totalWires]);
triangleVertIndices.resize(totalTriangles.size());
for (size_t i = 0; i < totalTriangles.size(); i++) {
triangleVertIndices[i] = boost::shared_array<index_t>(
new index_t[3 * totalTriangles[i]]);
}
positions = boost::shared_array<float>(
new float[3 * totalVerts]);
normals = boost::shared_array<float>(
new float[3 * totalVerts]);
if (uvExists) {
uvs = boost::shared_array<float>(
new float[2 * totalVerts]);
}
{
size_t wireIdx = 0;
size_t vertIdx = 0;
std::vector<size_t> triangleIdx(numIndexGroups, 0);
for (size_t i = 0; i < nbShapes; i++) {
const boost::shared_ptr<const ShapeSample>& sample =
consolidatedShapes[i]->getSample(*timeIt);
const size_t numWires = sample->numWires();
const size_t numVerts = sample->numVerts();
// Wires
if (sample->wireVertIndices()) {
IndexBuffer::ReadInterfacePtr readable = sample->wireVertIndices()->readableInterface();
const index_t* srcWireVertIndices = readable->get();
for (size_t j = 0; j < numWires; j++) {
wireVertIndices[2*(j + wireIdx) + 0] = index_t(srcWireVertIndices[2*j + 0] + vertIdx);
wireVertIndices[2*(j + wireIdx) + 1] = index_t(srcWireVertIndices[2*j + 1] + vertIdx);
}
}
// Triangles
for (size_t group = 0; group < numIndexGroups; group++) {
const size_t numTriangles = sample->numTriangles(group);
if (sample->triangleVertIndices(group)) {
IndexBuffer::ReadInterfacePtr readable = sample->triangleVertIndices(group)->readableInterface();
const index_t* srcTriangleVertIndices = readable->get();
for (size_t j = 0; j < numTriangles; j++) {
triangleVertIndices[group][3*(j + triangleIdx[group]) + 0] = index_t(srcTriangleVertIndices[3*j + 0] + vertIdx);
triangleVertIndices[group][3*(j + triangleIdx[group]) + 1] = index_t(srcTriangleVertIndices[3*j + 1] + vertIdx);
triangleVertIndices[group][3*(j + triangleIdx[group]) + 2] = index_t(srcTriangleVertIndices[3*j + 2] + vertIdx);
}
}
}
// Positions
if (sample->positions()) {
VertexBuffer::ReadInterfacePtr readable = sample->positions()->readableInterface();
memcpy(&positions[3*vertIdx], readable->get(), 3*numVerts*sizeof(float));
}
// Normals
if (sample->normals()) {
VertexBuffer::ReadInterfacePtr readable = sample->normals()->readableInterface();
memcpy(&normals[3*vertIdx], readable->get(), 3*numVerts*sizeof(float));
}
// UVs
if (sample->uvs()) {
VertexBuffer::ReadInterfacePtr readable = sample->uvs()->readableInterface();
memcpy(&uvs[2*vertIdx], readable->get(), 2*numVerts*sizeof(float));
}
else if (uvExists) {
memset(&uvs[2*vertIdx], 0, 2*numVerts*sizeof(float));
}
wireIdx += numWires;
vertIdx += numVerts;
for (size_t i = 0; i < numIndexGroups; i++) {
triangleIdx[i] += sample->numTriangles(i);
}
boundingBox.expand(sample->boundingBox());
}
}
std::vector<boost::shared_ptr<IndexBuffer> > newTriangleVertIndices(numIndexGroups);
for (size_t i = 0; i < numIndexGroups; i++) {
newTriangleVertIndices[i] = IndexBuffer::create(
SharedArray<index_t>::create(
triangleVertIndices[i], 3 * totalTriangles[i]));
}
boost::shared_ptr<ShapeSample> newSample = ShapeSample::create(
*timeIt,
totalWires,
totalVerts,
IndexBuffer::create(
SharedArray<index_t>::create(
wireVertIndices, 2 * totalWires)),
newTriangleVertIndices,
VertexBuffer::createPositions(
SharedArray<float>::create(
positions, 3 * totalVerts)),
boundingBox,
diffuseColor,
visibility);
if (normals) {
newSample->setNormals(
VertexBuffer::createNormals(
SharedArray<float>::create(
normals, 3 * totalVerts)));
}
if (uvs) {
newSample->setUVs(
VertexBuffer::createUVs(
SharedArray<float>::create(
uvs, 2 * totalVerts)));
}
newShape->addSample(newSample);
}
// Consolidate the remaining samples.
std::set<double>::const_iterator timePrev = timeIt;
++timeIt;
for (; timeIt != timeEnd; ++timeIt) {
size_t totalWires = 0;
size_t totalVerts = 0;
std::vector<size_t> totalTriangles;
size_t numIndexGroups = 0;
bool uvExists = false;
bool wiresDirty = false;
bool trianglesDirty = false;
bool positionsDirty = false;
bool normalsDirty = false;
bool uvsDirty = false;
for (size_t i = 0; i < nbShapes; i++) {
ShapeData::Ptr& shape = consolidatedShapes[i];
const boost::shared_ptr<const ShapeSample>& sample =
shape->getSample(*timeIt);
const boost::shared_ptr<const ShapeSample>& prevSample =
shape->getSample(*timePrev);
totalWires += sample->numWires();
totalVerts += sample->numVerts();
if (numIndexGroups == 0) {
// Initialize totalTriangles, assume that
// all shapes has the same number of index groups
numIndexGroups = sample->numIndexGroups();
totalTriangles.resize(numIndexGroups, 0);
diffuseColor = sample->diffuseColor();
visibility = sample->visibility();
}
// Shapes with different number of index groups, diffuseColor and visibility
// should be divided into separate buckets.
assert(numIndexGroups == sample->numIndexGroups());
assert(fabs(diffuseColor.r - sample->diffuseColor().r) < 1e-5);
assert(fabs(diffuseColor.g - sample->diffuseColor().g) < 1e-5);
assert(fabs(diffuseColor.b - sample->diffuseColor().b) < 1e-5);
assert(fabs(diffuseColor.a - sample->diffuseColor().a) < 1e-5);
assert(visibility == sample->visibility());
for (size_t j = 0; j < totalTriangles.size(); j++) {
totalTriangles[j] += sample->numTriangles(j);
}
// Check whether UV exists
if (!uvExists && sample->uvs()) {
uvExists = true;
}
for (size_t j = 0; j < numIndexGroups; j++) {
trianglesDirty |= sample->triangleVertIndices(j) != prevSample->triangleVertIndices(j);
}
wiresDirty |= sample->wireVertIndices() != prevSample->wireVertIndices();
positionsDirty |= sample->positions() != prevSample->positions();
normalsDirty |= sample->normals() != prevSample->normals();
uvsDirty |= sample->uvs() != prevSample->uvs();
}
if (wiresDirty || trianglesDirty ||
positionsDirty || normalsDirty || uvsDirty) {
if (wiresDirty) {
wireVertIndices = boost::shared_array<index_t>(
new index_t[2 * totalWires]);
}
if (trianglesDirty) {
triangleVertIndices.resize(totalTriangles.size());
for (size_t i = 0; i < totalTriangles.size(); i++) {
triangleVertIndices[i] = boost::shared_array<index_t>(
new index_t[3 * totalTriangles[i]]);
}
}
if (positionsDirty) {
positions = boost::shared_array<float>(
new float[3 * totalVerts]);
}
if (normalsDirty) {
normals = boost::shared_array<float>(
new float[3 * totalVerts]);
}
if (uvsDirty) {
uvs.reset();
if (uvExists) {
uvs = boost::shared_array<float>(
new float[2 * totalVerts]);
}
}
boundingBox.clear();
{
size_t wireIdx = 0;
size_t vertIdx = 0;
std::vector<size_t> triangleIdx(numIndexGroups, 0);
for (size_t i = 0; i < nbShapes; i++) {
const boost::shared_ptr<const ShapeSample>& sample =
consolidatedShapes[i]->getSample(*timeIt);
const size_t numWires = sample->numWires();
const size_t numVerts = sample->numVerts();
// Wires
if (wiresDirty && sample->wireVertIndices()) {
IndexBuffer::ReadInterfacePtr readable = sample->wireVertIndices()->readableInterface();
const index_t* srcWireVertIndices = readable->get();
for (size_t j = 0; j < numWires; j++) {
wireVertIndices[2*(j + wireIdx) + 0] = index_t(srcWireVertIndices[2*j + 0] + vertIdx);
wireVertIndices[2*(j + wireIdx) + 1] = index_t(srcWireVertIndices[2*j + 1] + vertIdx);
}
}
// Triangles
if (trianglesDirty) {
for (size_t group = 0; group < numIndexGroups; group++) {
const size_t numTriangles = sample->numTriangles(group);
if (sample->triangleVertIndices(group)) {
IndexBuffer::ReadInterfacePtr readable = sample->triangleVertIndices(group)->readableInterface();
const index_t* srcTriangleVertIndices = readable->get();
for (size_t j = 0; j < numTriangles; j++) {
triangleVertIndices[group][3*(j + triangleIdx[group]) + 0] = index_t(srcTriangleVertIndices[3*j + 0] + vertIdx);
triangleVertIndices[group][3*(j + triangleIdx[group]) + 1] = index_t(srcTriangleVertIndices[3*j + 1] + vertIdx);
triangleVertIndices[group][3*(j + triangleIdx[group]) + 2] = index_t(srcTriangleVertIndices[3*j + 2] + vertIdx);
}
}
}
}
// Positions
if (positionsDirty && sample->positions()) {
VertexBuffer::ReadInterfacePtr readable = sample->positions()->readableInterface();
memcpy(&positions[3*vertIdx], readable->get(),
3*numVerts*sizeof(float));
}
// Normals
if (normalsDirty && sample->normals()) {
VertexBuffer::ReadInterfacePtr readable = sample->normals()->readableInterface();
memcpy(&normals[3*vertIdx], readable->get(),
3*numVerts*sizeof(float));
}
// UVs
if (uvsDirty) {
if (sample->uvs()) {
VertexBuffer::ReadInterfacePtr readable = sample->uvs()->readableInterface();
memcpy(&uvs[2*vertIdx], readable->get(),
2*numVerts*sizeof(float));
} else if (uvExists) {
memset(&uvs[2*vertIdx], 0, 2*numVerts*sizeof(float));
}
}
wireIdx += numWires;
vertIdx += numVerts;
for (size_t i = 0; i < numIndexGroups; i++) {
triangleIdx[i] += sample->numTriangles(i);
}
boundingBox.expand(sample->boundingBox());
} // for each nodes
}
} // if anything dirty
std::vector<boost::shared_ptr<IndexBuffer> > newTriangleVertIndices(numIndexGroups);
for (size_t i = 0; i < numIndexGroups; i++) {
newTriangleVertIndices[i] = IndexBuffer::create(
SharedArray<index_t>::create(
triangleVertIndices[i], 3 * totalTriangles[i]));
}
boost::shared_ptr<ShapeSample> newSample = ShapeSample::create(
*timeIt,
totalWires,
totalVerts,
IndexBuffer::create(
SharedArray<index_t>::create(
wireVertIndices, 2 * totalWires)),
newTriangleVertIndices,
VertexBuffer::createPositions(
SharedArray<float>::create(
positions, 3 * totalVerts)),
boundingBox,
diffuseColor,
visibility);
if (normals) {
newSample->setNormals(
VertexBuffer::createNormals(
SharedArray<float>::create(
normals, 3 * totalVerts)));
}
if (uvs) {
newSample->setUVs(
VertexBuffer::createUVs(
SharedArray<float>::create(
uvs, 2 * totalVerts)));
}
newShape->addSample(newSample);
timePrev = timeIt;
}
// All consolidated shapes should have the same materials.
newShape->setMaterials(consolidatedShapes[0]->getMaterials());
// Re-use the largest node infos.
newShapes.push_back(newShape);
consolidatedShapes.clear();
}
SubNode::MPtr fRootNode;
const int fThreshold;
const bool fMotionBlur;
SubNode::MPtr fConsolidatedRootNode;
};
//==============================================================================
// CLASS SelectionChecker
//==============================================================================
class SelectionChecker
{
public:
SelectionChecker(const MSelectionList& selection)
{
MStatus status;
MDagPath dagPath;
// A selected node should be ignored
// if its parent/grandparent is selected.
for (unsigned int i = 0; i < selection.length(); i++) {
status = selection.getDagPath(i, dagPath);
if (status == MS::kSuccess) {
std::string fullDagPath = dagPath.fullPathName().asChar();
fSelectionPaths.insert(fullDagPath);
}
}
// Check each selected DAG Path
for (unsigned int i = 0; i < selection.length(); i++) {
status = selection.getDagPath(i, dagPath);
if (status == MS::kSuccess && check(dagPath)) {
fSelection.add(dagPath);
}
}
}
const MSelectionList& selection()
{ return fSelection; }
private:
// Prohibited and not implemented.
SelectionChecker(const SelectionChecker&);
const SelectionChecker& operator= (const SelectionChecker&);
bool check(const MDagPath& dagPath)
{
// This node should not have its parent/grandparent selected
MDagPath parent = dagPath;
parent.pop();
for (; parent.length() > 0; parent.pop()) {
std::string fullDagPath = parent.fullPathName().asChar();
if (fSelectionPaths.find(fullDagPath) != fSelectionPaths.end()) {
return false;
}
}
return checkGeometry(dagPath);
}
bool checkGeometry(const MDagPath& dagPath)
{
// Check we have bakeable geometry
MFnDagNode dagNode(dagPath);
MObject object = dagPath.node();
if ((Baker::isBakeable(object) ||
dagNode.typeId() == ShapeNode::id) &&
!object.hasFn(MFn::kTransform)) {
return true;
}
// At least one descendant must be bakeable geometry
bool hasGeometry = false;
for (unsigned int i = 0; i < dagPath.childCount(); i++) {
MDagPath child = dagPath;
child.push(dagPath.child(i));
MFnDagNode childNode(child);
if (childNode.isIntermediateObject()) {
continue;
}
if (checkGeometry(child)) {
hasGeometry= true;
break;
}
}
return hasGeometry;
}
MSelectionList fSelection;
std::set<std::string> fSelectionPaths;
};
//==============================================================================
// CLASS ScopedPauseWorkerThread
//==============================================================================
class ScopedPauseWorkerThread : boost::noncopyable
{
public:
ScopedPauseWorkerThread()
{
GlobalReaderCache::theCache().pauseRead();
}
~ScopedPauseWorkerThread()
{
GlobalReaderCache::theCache().resumeRead();
}
};
//==============================================================================
// CLASS FileAndSubNodeList
//==============================================================================
// This is a list of files and corresponding hierarchy roots.
// The dummy flag means that the top-level transform is a dummy sub-node. The
// dummy sub-node should be ignored and its children should be written instead.
class FileAndSubNode
{
public:
MFileObject targetFile;
SubNode::MPtr subNode;
bool isDummy;
FileAndSubNode()
: isDummy(false)
{}
FileAndSubNode(const MFileObject& f, const SubNode::MPtr& n, bool d)
: targetFile(f), subNode(n), isDummy(d)
{}
};
typedef std::vector<FileAndSubNode> FileAndSubNodeList;
//==============================================================================
// CLASS NodePathRegistry
//==============================================================================
// This class is responsible for re-constructing the sub-node hierarchy
// according to the original dag paths.
// This class should encapsulate all the logic to determine the file paths.
//
// If we are going to write a single hierarchy,
// targetFile = [directory] / [filePrefix] [fileName] [extension]
//
// Assuming we are going to bake the following hierarchies:
// |-A
// |-B
// |-C
// |-D
// We have 4 dag paths: |A, |A|B, |C and |C|D.
// In this case, we have two hierarchy roots: |A and |C
//
// 1) Either -allDagObjects or -saveMultipleFiles false is specified.
// We are going to write two hierarchies to a single file.
// targetFile = [directory] / [filePrefix] [fileName] [extension]
// e.g. ... / filename_specified_by_fileName_arg.abc
// (containing |A, |A|B, |C and |C|D)
//
// 2) -saveMultipleFiles true is specified. (default)
// We are going to write two hierarchies to two files.
// 2.1) -clashOption numericalIncrement
// targetFile = [directory] / [filePrefix] [numericalIncrement] [extension]
// e.g. ... / scene1_0.abc (containing |A and |A|B)
// ... / scene1_1.abc (containing |C and |C|D)
//
// 2.2) -clashOption nodeName (default)
// targetFile = [directory] / [filePrefix] [nodeName] [extension]
// [nodeName] is the name of the dag node. If there is any conflicts, the
// full dag path is used as [nodeName].
// e.g. ... / scene1_A.abc (containing |A and |A|B)
// ... / scene1_C.abc (containing |C and |C|D)
//
class NodePathRegistry : boost::noncopyable
{
public:
enum Stages {
kResolveStage, // Adding MDagPath objects.
kConstructStage, // Adding sub-node objects.
kComplete // Get a list of files and hierarchy roots.
};
NodePathRegistry(const bool allDagObjects,
const bool saveMultipleFiles,
const MString& directory,
const MString& filePrefix,
const MString& fileName,
const MString& clashOption)
: fStage(kResolveStage),
fAllDagObjects(allDagObjects),
fSaveMultipleFiles(saveMultipleFiles),
fDirectory(directory),
fFilePrefix(filePrefix),
fFileName(fileName),
fClashOption(clashOption),
fExtension(".abc")
{}
~NodePathRegistry()
{}
// Add a MDagPath object to this class. When all MDagPath objects are added,
// call resolve() to move to the next stage.
void add(const MDagPath& dagPath)
{
// We can only add dag paths at resolve stage.
assert(fStage == kResolveStage);
assert(dagPath.isValid());
// The full path name of this dag path.
std::string fullPath = fullPathName(dagPath);
assert(fPathMap.find(fullPath) == fPathMap.end());
// The full path name of the parent dag path.
MDagPath parentPath = dagPath;
parentPath.pop();
std::string parentFullPath = fullPathName(parentPath);
// Insert into path map.
PathEntry pathEntry;
pathEntry.dagPath = dagPath;
pathEntry.parentPath = parentFullPath;
fPathMap[fullPath] = pathEntry;
}
// Determine the hierarchy roots and file names based on all input dag paths.
// Resolve the hierarchy root node name conflict.
void resolve()
{
// We should be at resolve stage.
assert(fStage == kResolveStage);
fStage = kConstructStage;
// Find all hierarchy root nodes.
BOOST_FOREACH (const PathMapType::value_type& v, fPathMap) {
// If the path doesn't have a parent, it's a hierarchy root.
PathMapType::const_iterator it = fPathMap.find(v.second.parentPath);
if (it == fPathMap.end()) {
assert(fRootMap.find(v.first) == fRootMap.end());
// Get the node name.
MFnDagNode dagNode(v.second.dagPath);
std::string nodeName = dagNode.name().asChar();
// Strip namespace ":" character.
std::replace(nodeName.begin(), nodeName.end(), ':', '_');
// Insert into root map.
RootEntry rootEntry;
rootEntry.nodeName = nodeName;
rootEntry.uniqueName = "Not_Specified";
rootEntry.sequence = std::numeric_limits<size_t>::max();
rootEntry.overwrite = true;
fRootMap[v.first] = rootEntry;
}
}
// Count the occurance of base names of the hierarchy roots.
std::map<std::string,size_t> nameTable; // name -> occurance
BOOST_FOREACH (const RootMapType::value_type& v, fRootMap) {
nameTable.insert(std::make_pair(v.second.nodeName, 0)).first->second++;
}
// Resolve root node name conflicts and compute sequences.
size_t counter = 0;
BOOST_FOREACH (RootMapType::value_type& v, fRootMap) {
std::string uniqueName = v.second.nodeName;
// The name conflicts with other names.
// We use full path instead of its name.
if (nameTable.find(uniqueName)->second > 1) {
uniqueName = v.first.substr(1); // remove leading |
std::replace(uniqueName.begin(), uniqueName.end(), '|', '_');
std::replace(uniqueName.begin(), uniqueName.end(), ':', '_');
}
v.second.uniqueName = uniqueName;
v.second.sequence = counter++;
}
// Determine the directory we are going to export.
const bool singleFile = fAllDagObjects || !fSaveMultipleFiles || fRootMap.size() == 1;
const MString directory = validatedDirectory();
const MString fileName = validatedFileName();
// Determine the absolute file path for hierarchy roots.
BOOST_FOREACH (RootMapType::value_type& v, fRootMap) {
MString targetFileName = fFilePrefix;
if (singleFile) {
// We are going to save to a single file.
targetFileName += fileName;
}
else {
// We are going to save each hierarchy root to a separate file.
if (fClashOption == "numericalIncrement") {
// Clash Option: Numerical Increment
targetFileName += (unsigned int)v.second.sequence;
}
else {
// Clash Option: Node Name
targetFileName += v.second.uniqueName.c_str();
}
}
MString targetFullName = directory + "/" + targetFileName + fExtension;
v.second.targetFile.setRawFullName(targetFullName);
}
}
// Pop up a dialog to prompt the user we are going to overwrite the file.
void promptOverwrite()
{
// We have resolved hierarchy roots and their target files.
assert(fStage == kConstructStage);
// Make sure we have the dialog proc.
"if (!(`exists showGpuCacheExportConfirmDialog`))\n"
"{\n"
" eval(\"source \\\"doGpuCacheExportArgList.mel\\\"\");\n"
"}\n"
);
// Prompt every file or we remember the choice.
const bool singleFile = fAllDagObjects || !fSaveMultipleFiles || fRootMap.size() == 1;
enum OverwriteChoice { kUnknownOverwrite, kAlwaysOverwrite, kNeverOverwrite };
OverwriteChoice choice = kUnknownOverwrite;
BOOST_FOREACH (RootMapType::value_type& v, fRootMap) {
// Skip non-exist files.
if (!v.second.targetFile.exists()) continue;
if (choice == kUnknownOverwrite) {
// Show the dialog.
"showGpuCacheExportConfirmDialog \"" +
EncodeString(v.second.targetFile.resolvedFullName()) +
"\""
);
if (result == "yes") {
// Overwrite this file.
v.second.overwrite = true;
// All hierarchy roots are going to be written to a single file.
// Pop up the dialog only once for all hierarchy roots.
if (singleFile) choice = kAlwaysOverwrite;
}
else if (result == "no") {
// Skip this file.
v.second.overwrite = false;
// All hierarchy roots are going to be written to a single file.
// Pop up the dialog only once for all hierarchy roots.
if (singleFile) choice = kNeverOverwrite;
}
else if (result == "yesAll") {
// Overwrite this file and all following files.
v.second.overwrite = true;
choice = kAlwaysOverwrite;
}
else if (result == "noAll" || result == "dismiss") {
// Skip this file and all following files.
v.second.overwrite = false;
choice = kNeverOverwrite;
}
else {
// Something goes wrong with the dialog proc.
// We assume overwrite.
assert(0);
}
}
else if (choice == kAlwaysOverwrite) {
v.second.overwrite = true;
}
else if (choice == kNeverOverwrite) {
v.second.overwrite = false;
}
else {
assert(0);
}
}
}
// Associate the sub-node with the MDagPath object.
void associateSubNode(const MDagPath& dagPath, const SubNode::MPtr& subNode)
{
assert(fStage == kConstructStage);
assert(dagPath.isValid());
assert(subNode);
// Set the sub node member.
std::string fullPath = fullPathName(dagPath);
PathMapType::iterator it = fPathMap.find(fullPath);
assert(it != fPathMap.end());
if (it != fPathMap.end()) {
it->second.subNode = subNode;
}
}
// Construct sub-node hierarchy according to the MDagPath hierarchy.
void constructHierarchy()
{
assert(fStage == kConstructStage);
fStage = kComplete;
// Connect child with its parent.
// Instances are already expanded.
BOOST_FOREACH (const PathMapType::value_type& v, fPathMap) {
// Find this sub node.
const SubNode::MPtr& thisSubNode = v.second.subNode;
// Find parent sub node.
PathMapType::iterator parentIt = fPathMap.find(v.second.parentPath);
if (parentIt != fPathMap.end()) {
// Find a parent, connect them.
const SubNode::MPtr& parentSubNode = parentIt->second.subNode;
SubNode::connect(parentSubNode, thisSubNode);
}
else {
// This must be a hierarchy root.
RootMapType::iterator rootIt = fRootMap.find(v.first);
assert(rootIt != fRootMap.end());
if (rootIt != fRootMap.end()) {
rootIt->second.subNode = thisSubNode;
}
}
}
}
// Get a list of file paths and corresponding hierarchy root sub-node.
void generateFileAndSubNodes(FileAndSubNodeList& list)
{
// Sub-node hierarchies are ready.
assert(fStage == kComplete);
// When -allDagObjects is set, we are going to save all dag objects in
// the scene to a single file.
// When -saveMultipleFiles is false, we are going to save all selected
// dag objects to a single file.
// In both cases, we need to create a dummy root node ("|") for root nodes.
if (fAllDagObjects || !fSaveMultipleFiles) {
if (!fRootMap.empty() && fRootMap.begin()->second.overwrite) {
// Create the "|" node
GroupCreator groupCreator;
BOOST_FOREACH (const RootMapType::value_type& v, fRootMap) {
// We are going to write root nodes to a single file.
// To prevent root node name clash, we bake the unique name to the root node.
v.second.subNode->setName(v.second.uniqueName.c_str());
// Add the root node to the dummy group.
groupCreator.addChild(v.second.subNode);
}
groupCreator.group();
// Replace all nodes with a single "|" node.
// If there is one node, we use the node name as the "|" node name.
// If there are more than one node, we use the scene name as the "|" node name.
const RootEntry& rootEntry = fRootMap.begin()->second;
const MString rootNodeName = (fRootMap.size() == 1) ?
rootEntry.subNode->getName() : getSceneNameAsValidObjectName();
// Only one file.
list.push_back(FileAndSubNode(
rootEntry.targetFile,
groupCreator.getSubNode(rootNodeName),
true /* isDummy */
));
}
}
else {
// One hierarchy root to one file.
BOOST_FOREACH (const RootMapType::value_type& v, fRootMap) {
// Skip files that we are not going to overwrite.
if (!v.second.overwrite) continue;
list.push_back(FileAndSubNode(
v.second.targetFile,
v.second.subNode,
false /* isDummy */
));
}
}
}
void dump()
{
using namespace std;
stringstream out;
out << "Current Stage: ";
switch (fStage) {
case kResolveStage: out << "ResolveStage"; break;
case kConstructStage: out << "ConstructStage"; break;
case kComplete: out << "Complete"; break;
}
out << endl;
out << "Path Map: " << endl;
BOOST_FOREACH (const PathMapType::value_type& v, fPathMap) {
out << "Path: "
<< v.first
<< ", Parent: "
<< v.second.parentPath
<< ", SubNode: "
<< (void*)v.second.subNode.get()
<< endl;
}
out << "Root Map: " << endl;
BOOST_FOREACH (const RootMapType::value_type& v, fRootMap) {
out << "Path: "
<< v.first
<< ", Node: "
<< v.second.nodeName
<< ", Unique: "
<< v.second.uniqueName
<< ", Sequence: "
<< v.second.sequence
<< ", Overwrite: "
<< v.second.overwrite
<< ", Target: "
<< v.second.targetFile.resolvedFullName().asChar()
<< ", SubNode: "
<< (void*)v.second.subNode.get()
<< endl;
}
cout << out.str() << endl;
}
private:
static std::string fullPathName(const MDagPath& dagPath)
{
MString buffer = dagPath.fullPathName();
return std::string(buffer.asChar());
}
MString validatedDirectory()
{
MFileObject directory;
if (fDirectory.length() > 0) {
// If there is a directory specified, we use that directory.
directory.setRawFullName(fDirectory);
}
else {
MString fileRule, expandName;
const MString alembicFileRule = "alembicCache";
const MString alembicFilePath = "cache/alembic";
MString queryFileRuleCmd;
queryFileRuleCmd.format("workspace -q -fre \"^1s\"", alembicFileRule);
MString queryFolderCmd;
queryFolderCmd.format("workspace -en `workspace -q -fre \"^1s\"`", alembicFileRule);
// Query the file rule for alembic cache
MGlobal::executeCommand(queryFileRuleCmd, fileRule);
if (fileRule.length() > 0) {
// We have alembic file rule, query the folder
MGlobal::executeCommand(queryFolderCmd, expandName);
}
else {
// Alembic file rule does not exist, create it
MString addFileRuleCmd;
addFileRuleCmd.format("workspace -fr \"^1s\" \"^2s\"", alembicFileRule, alembicFilePath);
MGlobal::executeCommand(addFileRuleCmd);
// Save the workspace. maya may discard file rules on exit
MGlobal::executeCommand("workspace -s");
// Query the folder
MGlobal::executeCommand(queryFolderCmd, expandName);
}
// Resolve the expanded file rule
if (expandName.length() == 0) {
expandName = alembicFilePath;
}
directory.setRawFullName(expandName);
}
return directory.resolvedFullName();
}
MString validatedFileName()
{
MString fileName;
if (fFileName.length() > 0) {
// If there is a file name specified, we use that file name.
fileName = fFileName;
}
else {
// Generate a default file name.
if (fRootMap.size() == 1) {
fileName = fRootMap.begin()->second.uniqueName.c_str();
}
else {
fileName = getSceneName();
}
}
return fileName;
}
// This map contains name/path info for all dag paths.
struct PathEntry {
MDagPath dagPath;
std::string parentPath;
SubNode::MPtr subNode;
};
typedef boost::unordered_map<std::string,PathEntry> PathMapType;
// This map contains name/path info for all hierarchy roots.
struct RootEntry {
std::string nodeName;
std::string uniqueName;
size_t sequence;
bool overwrite;
MFileObject targetFile;
SubNode::MPtr subNode;
};
typedef boost::unordered_map<std::string,RootEntry> RootMapType;
Stages fStage;
PathMapType fPathMap;
RootMapType fRootMap;
const bool fAllDagObjects;
const bool fSaveMultipleFiles;
const MString fDirectory;
const MString fFilePrefix;
const MString fFileName;
const MString fClashOption;
const MString fExtension;
};
}
namespace GPUCache {
//==============================================================================
// CLASS Command
//==============================================================================
void* Command::creator()
{
return new Command();
}
MSyntax Command::cmdSyntax()
{
MSyntax syntax;
syntax.addFlag("-dir", "-directory", MSyntax::kString );
syntax.addFlag("-f", "-fileName", MSyntax::kString );
syntax.addFlag("-smf", "-saveMultipleFiles", MSyntax::kBoolean );
syntax.addFlag("-fp", "-filePrefix", MSyntax::kString );
syntax.addFlag("-clo", "-clashOption", MSyntax::kString );
syntax.addFlag("-o", "-optimize" );
syntax.addFlag("-ot", "-optimizationThreshold", MSyntax::kUnsigned );
syntax.addFlag("-st", "-startTime", MSyntax::kTime );
syntax.addFlag("-et", "-endTime", MSyntax::kTime );
syntax.addFlag("-smr", "-simulationRate", MSyntax::kTime );
syntax.addFlag("-spm", "-sampleMultiplier", MSyntax::kLong );
syntax.addFlag("-cl", "-compressLevel", MSyntax::kLong );
syntax.addFlag("-df", "-dataFormat", MSyntax::kString );
syntax.addFlag("-sf", "-showFailed" );
syntax.addFlag("-ss", "-showStats" );
syntax.addFlag("-sgs", "-showGlobalStats" );
syntax.addFlag("-dh", "-dumpHierarchy", MSyntax::kString );
syntax.addFlag("-atr", "-animTimeRange" );
syntax.addFlag("-gma", "-gpuManufacturer" );
syntax.addFlag("-gmo", "-gpuModel" );
syntax.addFlag("-gdv", "-gpuDriverVersion" );
syntax.addFlag("-gms", "-gpuMemorySize" );
syntax.addFlag("-ado", "-allDagObjects" );
syntax.addFlag("-r", "-refresh" );
syntax.addFlag("-ra", "-refreshAll" );
syntax.addFlag("-rs", "-refreshSettings" );
syntax.addFlag("-wbr", "-waitForBackgroundReading" );
syntax.addFlag("-wm", "-writeMaterials" );
syntax.addFlag("-wuv", "-writeUVs" );
syntax.addFlag("-omb", "-optimizeAnimationsForMotionBlur" );
syntax.addFlag("-ubt", "-useBaseTessellation" );
syntax.addFlag("-p", "-prompt" );
syntax.addFlag("-lfe", "-listFileEntries" );
syntax.addFlag("-lse", "-listShapeEntries" );
syntax.makeFlagQueryWithFullArgs("-dumpHierarchy", true);
syntax.useSelectionAsDefault(true);
syntax.enableQuery(true);
syntax.enableEdit(true);
return syntax;
}
Command::Command()
: fMode(kCreate)
{}
Command::~Command()
{}
bool Command::isUndoable() const
{
return false;
}
bool Command::hasSyntax() const
{
return true;
}
void Command::AddHierarchy(
const MDagPath& dagPath,
std::map<std::string, int>* idMap,
std::vector<MObject>* sourceNodes,
std::vector<std::vector<MDagPath> >* sourcePaths,
std::vector<MObject>* gpuCacheNodes)
{
MFnDagNode dagNode(dagPath.node());
MDagPath firstDagPath;
MStatus status = dagNode.getPath(firstDagPath);
if (status != MS::kSuccess) return;
std::string firstPath(firstDagPath.partialPathName().asChar());
std::map<std::string, int>::iterator pos = idMap->find(firstPath);
if (pos != idMap->end()){
// Already traversed. Only store its DAG Path.
(*sourcePaths)[pos->second].push_back(dagPath);
}
else {
MObject object(dagNode.object());
MString msgFmt;
bool isWarning = true;
if (dagNode.typeId() == ShapeNode::id) {
if (fMode == kCreate) {
// Recursive bake a gpuCache node
(*idMap)[firstPath] = (int)sourceNodes->size();
sourceNodes->push_back(object);
sourcePaths->push_back(std::vector<MDagPath>(1, dagPath));
}
else {
// Query flag is set
gpuCacheNodes->push_back(object);
}
}
else if (Baker::isBakeable(object)) {
(*idMap)[firstPath] = (int)sourceNodes->size();
sourceNodes->push_back(object);
sourcePaths->push_back(std::vector<MDagPath>(1, dagPath));
if (fMode != kCreate && fShowFailedFlag.isSet()) {
MStatus status;
msgFmt = MStringResource::getString(kNodeWontBakeErrorMsg, status);
}
}
else if (fShowFailedFlag.isSet()) {
MStatus status;
msgFmt = MStringResource::getString(kNodeBakedFailedErrorMsg, status);
}
if (msgFmt.length() > 0) {
MString nodeName = firstDagPath.fullPathName();
MString msg;
msg.format(msgFmt, nodeName);
if (isWarning) {
}
else {
}
}
}
unsigned int numChild = dagPath.childCount();
for(unsigned int i = 0; i < numChild; ++i) {
MDagPath childPath = dagPath;
childPath.push(dagPath.child(i));
MFnDagNode childNode(childPath);
if (!childNode.isIntermediateObject())
AddHierarchy(childPath, idMap, sourceNodes, sourcePaths, gpuCacheNodes);
}
}
bool Command::AddSelected(
const MSelectionList& objects,
std::vector<MObject>* sourceNodes,
std::vector<std::vector<MDagPath> >* sourcePaths,
std::vector<MObject>* gpuCacheNodes)
{
MStatus status;
// map first DAG path to node index
std::map<std::string, int> idMap;
for (unsigned int i = 0; i<objects.length(); ++i) {
MDagPath sourceDagPath;
status = objects.getDagPath(i, sourceDagPath);
if (status == MS::kSuccess) {
AddHierarchy(sourceDagPath, &idMap, sourceNodes, sourcePaths, gpuCacheNodes);
}
}
if (fMode == kCreate) {
if (sourceNodes->empty()) {
MStatus stat;
MString msg;
if (gpuCacheNodes->empty()) {
msg = MStringResource::getString(kNoObjBakable2ErrorMsg, stat);
}
else {
msg = MStringResource::getString(kNoObjBakable1ErrorMsg, stat);
}
displayWarning(msg);
return false;
}
return true;
}
else {
if (!fRefreshSettingsFlag.isSet() && gpuCacheNodes->empty()) {
MStatus stat;
MString msg;
if (sourceNodes->empty()) {
msg = MStringResource::getString(kNoObjBaked2ErrorMsg, stat);
}
else {
msg = MStringResource::getString(kNoObjBaked1ErrorMsg, stat);
}
displayWarning(msg);
return false;
}
return true;
}
}
MStatus Command::doIt(const MArgList& args)
{
MStatus status;
MArgDatabase argsDb(syntax(), args, &status);
if (!status) return status;
unsigned int numFlags = 0;
// Save the command arguments for undo/redo purposes.
if (argsDb.isEdit()) {
if (argsDb.isQuery()) {
MStatus stat;
MString msg = MStringResource::getString(kEditQueryFlagErrorMsg, stat);
displayError(msg);
return MS::kFailure;
}
fMode = kEdit;
numFlags++;
}
else if (argsDb.isQuery()) {
fMode = kQuery;
numFlags++;
}
numFlags += fDirectoryFlag.parse(argsDb, "-directory");
if (!fDirectoryFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kDirectoryWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fFileNameFlag.parse(argsDb, "-fileName");
if (!fFileNameFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kFileNameWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fSaveMultipleFilesFlag.parse(argsDb, "-saveMultipleFiles");
if (!fSaveMultipleFilesFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kSaveMultipleFilesWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fFilePrefixFlag.parse(argsDb, "-filePrefix");
if (!fFilePrefixFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kFilePrefixWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fClashOptionFlag.parse(argsDb, "-clashOption");
if (!fClashOptionFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kClashOptionWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fOptimizeFlag.parse(argsDb, "-optimize");
if (!fOptimizeFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kOptimizeWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fOptimizationThresholdFlag.parse(argsDb, "-optimizationThreshold");
if (!fOptimizationThresholdFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kOptimizationThresholdWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fStartTimeFlag.parse(argsDb, "-startTime");
if (!fStartTimeFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kStartTimeWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fEndTimeFlag.parse(argsDb, "-endTime");
if (!fEndTimeFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kEndTimeWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fSimulationRateFlag.parse(argsDb, "-simulationRate");
if (!fSimulationRateFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kSimulationRateWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
if (fSimulationRateFlag.isSet()) {
MTime minRate(0.004, MTime::kFilm);
if (fSimulationRateFlag.arg() < minRate) {
// Simulation rate was below 1 tick, issue an appropriate error message.
MStatus stat;
MString msg, fmt = MStringResource::getString(kSimulationRateWrongValueMsg, stat);
msg.format(fmt, MString() + minRate.as(MTime::uiUnit()));
displayError(msg);
return MS::kFailure;
}
}
numFlags += fSampleMultiplierFlag.parse(argsDb, "-sampleMultiplier");
if (!fSampleMultiplierFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kSampleMultiplierWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
if (fSampleMultiplierFlag.isSet() && fSampleMultiplierFlag.arg() <= 0) {
MStatus stat;
MString msg = MStringResource::getString(kSampleMultiplierWrongValueMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fCompressLevelFlag.parse(argsDb, "-compressLevel");
if (!fCompressLevelFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kCompressLevelWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fDataFormatFlag.parse(argsDb, "-dataFormat");
if (!fDataFormatFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kDataFormatWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fShowFailedFlag.parse(argsDb, "-showFailed");
assert(fShowFailedFlag.isModeValid(fMode));
numFlags += fShowStats.parse(argsDb, "-showStats");
assert(fShowStats.isModeValid(fMode));
numFlags += fShowGlobalStats.parse(argsDb, "-showGlobalStats");
assert(fShowGlobalStats.isModeValid(fMode));
numFlags += fDumpHierarchy.parse(argsDb, "-dumpHierarchy");
assert(fDumpHierarchy.isModeValid(fMode));
numFlags += fAnimTimeRangeFlag.parse(argsDb, "-animTimeRange");
if (!fAnimTimeRangeFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kAnimTimeRangeWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fAllDagObjectsFlag.parse(argsDb, "-allDagObjects");
if (!fAllDagObjectsFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kAllDagObjectsWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fRefreshFlag.parse(argsDb, "-refresh");
if (!fRefreshFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kRefreshWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fRefreshAllFlag.parse(argsDb, "-refreshAll");
if (!fRefreshAllFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kRefreshAllWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fListFileEntriesFlag.parse(argsDb, "-listFileEntries");
if (!fListFileEntriesFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kListFileEntriesWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fListShapeEntriesFlag.parse(argsDb, "-listShapeEntries");
if (!fListShapeEntriesFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kListShapeEntriesWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fRefreshSettingsFlag.parse(argsDb, "-refreshSettings");
if (!fRefreshSettingsFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kRefreshSettingsWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fGpuManufacturerFlag.parse(argsDb, "-gpuManufacturer");
if (!fGpuManufacturerFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kGpuManufacturerWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fGpuModelFlag.parse(argsDb, "-gpuModel");
if (!fGpuModelFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kGpuModelWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fGpuDriverVersion.parse(argsDb, "-gpuDriverVersion");
if (!fGpuDriverVersion.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kGpuDriverVersionWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fGpuMemorySize.parse(argsDb, "-gpuMemorySize");
if (!fGpuMemorySize.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kGpuMemorySizeWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fWaitForBackgroundReadingFlag.parse(argsDb, "-waitForBackgroundReading");
if (!fWaitForBackgroundReadingFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kWaitForBackgroundReadingWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fWriteMaterials.parse(argsDb, "-writeMaterials");
if (!fWriteMaterials.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kWriteMaterialsWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fUVsFlag.parse(argsDb, "-writeUVs");
if( !fUVsFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kWriteUVsWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fOptimizeAnimationsForMotionBlurFlag.parse(argsDb, "-optimizeAnimationsForMotionBlur");
if (!fOptimizeAnimationsForMotionBlurFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kOptimizeAnimationsForMotionBlurWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fUseBaseTessellationFlag.parse(argsDb, "-useBaseTessellation");
if (!fUseBaseTessellationFlag.isModeValid(fMode)) {
MStatus stat;
MString msg = MStringResource::getString(kUseBaseTessellationWrongModeMsg, stat);
displayError(msg);
return MS::kFailure;
}
numFlags += fPromptFlag.parse(argsDb, "-prompt");
if (fRefreshAllFlag.isSet())
{
// Ideally, we would use MArgParser to determine number of other flags used.
// However, MArgParser returns an error when usesSelectionAsDefault() == true
// and no objects are found.
//
// Instead, we manually test for the presence of other flags.
if( numFlags > 1 )
{
MStatus stat;
MString msg = MStringResource::getString(kRefreshAllOtherFlagsMsg, stat);
displayError(msg);
return MS::kFailure;
}
refreshAll();
return MS::kSuccess;
}
if (fListFileEntriesFlag.isSet())
{
if( numFlags > 1 )
{
MStatus stat;
MString msg = MStringResource::getString(kListFileEntriesOtherFlagsMsg, stat);
displayError(msg);
return MS::kFailure;
}
listFileEntries();
return MS::kSuccess;
}
if (fListShapeEntriesFlag.isSet())
{
if( numFlags > 1 )
{
MStatus stat;
MString msg = MStringResource::getString(kListShapeEntriesOtherFlagsMsg, stat);
displayError(msg);
return MS::kFailure;
}
listShapeEntries();
return MS::kSuccess;
}
// Backup the current selection
MSelectionList selectionBackup;
MSelectionList objects;
if (fAllDagObjectsFlag.isSet()) {
// -allDagObjects flag is set, export all the top-level DAG Nodes
MStringArray result;
MGlobal::executeCommand("ls -assemblies -long", result);
for (unsigned int i = 0; i < result.length(); i++) {
objects.add(result[i]);
}
}
else {
// -allDagObjects flag is not set, export the selection or gpuCache arguments
// Duplicates are removed by merge().
MSelectionList selectedObjectArgs;
status = argsDb.getObjects(selectedObjectArgs);
MStatError(status, "argsDb.getObjects()");
if (!selectedObjectArgs.isEmpty()) {
status = objects.merge(selectedObjectArgs, MSelectionList::kMergeNormal);
MStatError(status, "objects.merge()");
}
}
if (objects.length() == 0 &&
!(fMode == kQuery && fShowGlobalStats.isSet()) &&
!(fMode == kEdit && fRefreshSettingsFlag.isSet()) &&
!(fMode == kQuery && fGpuManufacturerFlag.isSet()) &&
!(fMode == kQuery && fGpuModelFlag.isSet()) &&
!(fMode == kQuery && fGpuDriverVersion.isSet()) &&
!(fMode == kQuery && fGpuMemorySize.isSet())
) {
MString msg = MStringResource::getString(kNoObjectsMsg,status);
return MS::kFailure;
}
{
SelectionChecker selectionChecker(objects);
objects = selectionChecker.selection();
}
std::vector<MObject> sourceNodes;
std::vector<std::vector<MDagPath> > sourcePaths;
std::vector<MObject> gpuCacheNodes;
if (fMode == kCreate || fMode == kEdit || fShowStats.isSet() ||
fDumpHierarchy.isSet() || fAnimTimeRangeFlag.isSet() ||
fWaitForBackgroundReadingFlag.isSet()) {
if (!AddSelected(objects, &sourceNodes, &sourcePaths, &gpuCacheNodes))
return MS::kFailure;
}
// We flush the selection list before executing any MEL command
// through MDGModifier::commandToExecute. This saves a LOT of
// memory!!! This is due to the fact that each executed MEL
// command might take a copy of the selection list to restore it
// on undo. But, this is totally unnecessary since we invoking
// them from another MEL command that already takes care of
// restoring the selection list on undo!!!
switch (fMode) {
case kCreate: status = doCreate(sourceNodes, sourcePaths, objects); break;
case kEdit: status = doEdit(gpuCacheNodes); break;
case kQuery: status = doQuery(gpuCacheNodes); break;
}
// Restore the selection.
return status;
}
MStatus Command::doCreate(const std::vector<MObject>& sourceNodes,
const std::vector<std::vector<MDagPath> >& sourcePaths,
const MSelectionList& objects)
{
MStatus status;
// Compute the baked mesh before committing the Dag modifier so
// that the Dag modifier includes the baking.
MCheckReturn(
Command::doBaking(
sourceNodes,
sourcePaths,
fStartTimeFlag.arg(MAnimControl::animationStartTime()),
fEndTimeFlag.arg(MAnimControl::animationEndTime()),
fSimulationRateFlag.arg(MTime(1, MTime::uiUnit())),
fSampleMultiplierFlag.arg(1)));
return MS::kSuccess;
}
MStatus Command::doQuery(const std::vector<MObject>& gpuCacheNodes) const
{
// set the result of gpuCache command
if (fShowStats.isSet() ||
fShowGlobalStats.isSet() ||
fDumpHierarchy.isSet()
) {
// String array result is incompatible with double[2]
if (fAnimTimeRangeFlag.isSet()) {
MStatus stat;
MString msg = MStringResource::getString(kIncompatibleQueryMsg,stat);
return MS::kFailure;
}
MStringArray result;
if (fShowStats.isSet()) {
showStats(gpuCacheNodes, result);
}
if (fShowGlobalStats.isSet()) {
showGlobalStats(result);
}
if (fDumpHierarchy.isSet()) {
if (fDumpHierarchy.isArgValid()) {
// Dump to a text file
file.setRawFullName(fDumpHierarchy.arg());
MCheckReturn( dumpHierarchyToFile(gpuCacheNodes, file) );
result.append("Dumping hierarchy to: " + file.resolvedFullName());
}
else {
// Dump to script editor
dumpHierarchy(gpuCacheNodes, result);
}
}
{
MString output;
for (unsigned int i = 0; i < result.length(); i++) {
if (i > 0) output += "\n";
output += result[i];
}
}
}
else if (fAnimTimeRangeFlag.isSet()) {
// -animTimeRange will return double[2] in current time unit
MDoubleArray animTimeRange;
showAnimTimeRange(gpuCacheNodes, animTimeRange);
MPxCommand::setResult(animTimeRange);
}
else if (fGpuManufacturerFlag.isSet()) {
MPxCommand::setResult(VramQuery::manufacturer());
}
else if (fGpuModelFlag.isSet()) {
MPxCommand::setResult(VramQuery::model());
}
else if (fGpuDriverVersion.isSet()) {
int driverVersion[3];
VramQuery::driverVersion(driverVersion);
MString verionStr;
verionStr += driverVersion[0];
verionStr += ".";
verionStr += driverVersion[1];
verionStr += ".";
verionStr += driverVersion[2];
}
else if (fGpuMemorySize.isSet()) {
MPxCommand::setResult((int)(VramQuery::queryVram() / 1024 / 1024));
}
else if (fWaitForBackgroundReadingFlag.isSet()) {
// Wait until the background reading is finished.
BOOST_FOREACH (const MObject& node, gpuCacheNodes) {
// Request the geometry to begin reading
MFnDagNode dagNode(node);
ShapeNode* shapeNode = (ShapeNode*)dagNode.userNode();
if (shapeNode) {
shapeNode->getCachedGeometry();
}
// Wait for the reading
GlobalReaderCache::theCache().waitForRead(shapeNode->getCacheFileEntry().get());
// Pull the data
if (shapeNode) {
shapeNode->getCachedGeometry();
}
}
}
return MS::kSuccess;
}
MStatus Command::doEdit(const std::vector<MObject>& gpuCacheNodes)
{
if (fRefreshSettingsFlag.isSet()) {
Config::refresh();
}
if (fRefreshFlag.isSet()) {
refresh(gpuCacheNodes);
}
return MS::kSuccess;
}
MStatus Command::doBaking(
const std::vector<MObject>& sourceNodes,
const std::vector<std::vector<MDagPath> >& sourcePaths,
MTime startTime,
MTime endTime,
MTime simulationRate,// The time interval to do the simulation.
int samplingRate // How many time intervals to sample once.
)
{
// Check the start time and end time.
if (startTime > endTime) {
MStatus stat;
MString msg = MStringResource::getString(kStartEndTimeErrorMsg, stat);
return MS::kFailure;
}
// Find out the file names that we are going to write to.
NodePathRegistry pathRegistry(
fAllDagObjectsFlag.isSet(),
fSaveMultipleFilesFlag.arg(true),
fDirectoryFlag.arg(),
fFilePrefixFlag.arg(),
fFileNameFlag.arg(),
fClashOptionFlag.arg()
);
BOOST_FOREACH (const std::vector<MDagPath>& dagPaths, sourcePaths) {
BOOST_FOREACH (const MDagPath& path, dagPaths) {
pathRegistry.add(path);
}
}
pathRegistry.resolve();
// Prompt for overwriting files. (Default is overwrite)
if (MGlobal::mayaState() == MGlobal::kInteractive && fPromptFlag.isSet()) {
pathRegistry.promptOverwrite();
}
// Set up the progress bar for baking
ProgressBar progressBar(kExportingMsg,
(unsigned int)(sourceNodes.size() * (int)(
(endTime - startTime + simulationRate).as(MTime::kSeconds) /
simulationRate.as(MTime::kSeconds)) / samplingRate));
// First save the current time, so we can restore it later.
const MTime previousTime = MAnimControl::currentTime();
// For go to start time.
MTime currentTime = startTime;
// The DAG object bakers.
typedef std::vector< boost::shared_ptr<Baker> > Bakers;
Bakers bakers;
// The top-level baker for materials.
boost::shared_ptr<MaterialBaker> materialBaker;
if (fWriteMaterials.isSet()) {
materialBaker = boost::make_shared<MaterialBaker>();
}
for (size_t i = 0; i < sourceNodes.size(); i++) {
// Create a new DAG object baker.
const boost::shared_ptr<Baker> baker(
Baker::create(sourceNodes[i], sourcePaths[i]));
if (!baker) {
MStatus stat;
MString msg = MStringResource::getString(kCreateBakerErrorMsg, stat);
return MS::kFailure;
}
const boost::shared_ptr<ShapeBaker> shapeBaker = boost::dynamic_pointer_cast<ShapeBaker,Baker>( baker );
if( shapeBaker && fUVsFlag.isSet() )
{
shapeBaker->enableUVs();
}
if (materialBaker) {
baker->setWriteMaterials();
}
if (fUseBaseTessellationFlag.isSet()) {
baker->setUseBaseTessellation();
}
bakers.push_back(baker);
// sample all shapes at start time
MCheckReturn(baker->sample(currentTime));
// Add the connected shaders to the material baker.
if (materialBaker) {
BOOST_FOREACH (const MDagPath& path, sourcePaths[i]) {
if (path.node().hasFn(MFn::kShape)) {
MCheckReturn( materialBaker->addShapePath(path) );
}
}
}
MUpdateProgressAndCheckInterruption(progressBar);
}
// Sample all materials at start time.
if (materialBaker) {
MCheckReturn( materialBaker->sample(currentTime) );
}
// Sample the vertex attributes over time.
currentTime += simulationRate;
for (int sampleIdx = 1; currentTime<=endTime;
currentTime += simulationRate, ++sampleIdx) {
// Advance time.
if (sampleIdx % samplingRate == 0) {
BOOST_FOREACH(const boost::shared_ptr<Baker>& baker, bakers) {
MCheckReturn(baker->sample(currentTime));
MUpdateProgressAndCheckInterruption(progressBar);
} // for each baker
if (materialBaker) {
MCheckReturn( materialBaker->sample(currentTime) );
}
}
} // for each time sample
// Construct the material graphs
MaterialGraphMap::Ptr materials;
if (materialBaker) {
materialBaker->buildGraph();
materials = materialBaker->get();
}
// Construct SubNode hierarchy.
{
assert(bakers.size() == sourceNodes.size());
assert(bakers.size() == sourcePaths.size());
// Create a SubNode for each instance.
for (size_t i = 0; i < sourcePaths.size(); i++) {
for (size_t j = 0; j < sourcePaths[i].size(); j++) {
const MDagPath& path = sourcePaths[i][j];
SubNode::MPtr subNode = bakers[i]->getNode(j);
pathRegistry.associateSubNode(path, subNode);
}
}
pathRegistry.constructHierarchy();
}
// We are done with the bakers now.
bakers.clear();
materialBaker.reset();
// Restore current time.
// Preparing the root nodes and files to write.
FileAndSubNodeList fileList;
pathRegistry.generateFileAndSubNodes(fileList);
// Do consolidation
if (fOptimizeFlag.isSet()) {
const int threshold = fOptimizationThresholdFlag.arg(40000);
const bool motionBlur = fOptimizeAnimationsForMotionBlurFlag.isSet();
BOOST_FOREACH (FileAndSubNode& v, fileList) {
Consolidator consolidator(v.subNode, threshold, motionBlur);
MCheckReturn( consolidator.consolidate() );
SubNode::MPtr consolidatedRootNode = consolidator.consolidatedRootNode();
if (consolidatedRootNode) {
v.subNode = consolidatedRootNode;
v.isDummy = false;
}
}
}
// Set up progress bar for writing
//
// FIXME: The cache writer should provide more granularity for
// updating the progress bar.
progressBar.reset(kWritingMsg, (unsigned int)fileList.size());
// Write the baked geometry to the cache file.
const MTime timePerCycle = simulationRate * samplingRate;
Writer gpuCacheWriter(
(char)fCompressLevelFlag.arg(-1),
fDataFormatFlag.arg("hdf"),
timePerCycle,
startTime
);
BOOST_FOREACH (const FileAndSubNode& v, fileList) {
if (v.isDummy) {
// This is a dummy root node. We are going to write its children.
MCheckReturn(
gpuCacheWriter.writeNodes(
v.subNode->getChildren(),
materials,
v.targetFile
)
);
}
else {
// We write the node to its taget file.
MCheckReturn(
gpuCacheWriter.writeNode(
v.subNode,
materials,
v.targetFile
)
);
}
appendToResult(v.targetFile.resolvedFullName());
MUpdateProgressAndCheckInterruption(progressBar);
}
return MS::kSuccess;
}
void Command::showStats(
const std::vector<MObject>& gpuCacheNodes,
MStringArray& result
) const
{
MStatus status;
{
result.append(MStringResource::getString(kStatsAllFramesMsg, status));
StatsVisitor stats;
BOOST_FOREACH(const MObject& gpuCacheObject, gpuCacheNodes) {
MFnDagNode gpuCacheFn(gpuCacheObject);
MPxNode* node = gpuCacheFn.userNode();
assert(node);
assert(dynamic_cast<ShapeNode*>(node));
ShapeNode* gpuCacheNode =
static_cast<ShapeNode*>(node);
stats.accumulateNode(gpuCacheNode->getCachedGeometry());
stats.accumulateMaterialGraph(gpuCacheNode->getCachedMaterial());
}
stats.print(result, false);
}
{
result.append(MStringResource::getString(kStatsCurrentFrameMsg, status));
StatsVisitor stats(MAnimControl::currentTime());
BOOST_FOREACH(const MObject& gpuCacheObject, gpuCacheNodes) {
MFnDagNode gpuCacheFn(gpuCacheObject);
MPxNode* node = gpuCacheFn.userNode();
assert(node);
assert(dynamic_cast<ShapeNode*>(node));
ShapeNode* gpuCacheNode =
static_cast<ShapeNode*>(node);
stats.accumulateNode(gpuCacheNode->getCachedGeometry());
stats.accumulateMaterialGraph(gpuCacheNode->getCachedMaterial());
}
stats.print(result, true);
}
}
void Command::showGlobalStats(
MStringArray& result
) const
{
MStatus status;
// Exclude internal unit bounding box
const size_t unitBoundingBoxIndicesBytes = UnitBoundingBox::indices()->bytes();
const size_t unitBoundingBoxPositionsBytes = UnitBoundingBox::positions()->bytes();
const size_t unitBoundingBoxBytes = unitBoundingBoxIndicesBytes + unitBoundingBoxPositionsBytes;
const size_t unitBoundingBoxNbIndices = 1;
const size_t unitBoundingBoxNbVertices = 1;
const size_t unitBoundingBoxNbBuffers = unitBoundingBoxNbIndices + unitBoundingBoxNbVertices;
// System memory buffers
{
MString memUnit;
double memSize =
toHumanUnits(IndexBuffer::nbAllocatedBytes() +
VertexBuffer::nbAllocatedBytes() -
unitBoundingBoxBytes,
memUnit);
MString msg;
MString msg_buffers; msg_buffers += (double)(IndexBuffer::nbAllocated() +
VertexBuffer::nbAllocated() -
unitBoundingBoxNbBuffers);
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalSystemStatsMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
{
MString memUnit;
double memSize =
toHumanUnits(IndexBuffer::nbAllocatedBytes() - unitBoundingBoxIndicesBytes, memUnit);
MString msg;
MString msg_buffers; msg_buffers += (double)(IndexBuffer::nbAllocated() - unitBoundingBoxNbIndices);
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalSystemStatsIndexMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
{
MString memUnit;
double memSize =
toHumanUnits(VertexBuffer::nbAllocatedBytes() - unitBoundingBoxPositionsBytes, memUnit);
MString msg;
MString msg_buffers; msg_buffers +=
(double)(VertexBuffer::nbAllocated() - unitBoundingBoxNbVertices);
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalSystemStatsVertexMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
// Video memory buffers
{
MString memUnit;
double memSize = toHumanUnits(VBOBuffer::nbAllocatedBytes(),
memUnit);
MString msg;
MString msg_buffers; msg_buffers += (double)(VBOBuffer::nbAllocated());
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalVideoStatsMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
{
MString memUnit;
double memSize = toHumanUnits(VBOBuffer::nbIndexAllocatedBytes(),
memUnit);
MString msg;
MString msg_buffers; msg_buffers +=
(double)VBOBuffer::nbIndexAllocated();
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalVideoStatsIndexMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
{
MString memUnit;
double memSize = toHumanUnits(VBOBuffer::nbVertexAllocatedBytes(),
memUnit);
MString msg;
MString msg_buffers; msg_buffers +=
(double)VBOBuffer::nbVertexAllocated();
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalVideoStatsVertexMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
// Last refresh statistics
{
MString msg;
msg.format(MStringResource::getString(kGlobalRefreshStatsMsg, status));
result.append(msg);
}
{
MString memUnit;
double memSize = toHumanUnits(VBOBuffer::nbUploadedBytes(),
memUnit);
MString msg;
MString msg_buffers; msg_buffers +=
(double)VBOBuffer::nbUploaded();
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalRefreshStatsUploadMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
{
MString memUnit;
double memSize = toHumanUnits(VBOBuffer::nbEvictedBytes(),
memUnit);
MString msg;
MString msg_buffers; msg_buffers +=
(double)VBOBuffer::nbEvicted();
MString msg_memSize; msg_memSize += memSize;
msg.format(
MStringResource::getString(kGlobalRefreshStatsEvictionMsg, status),
msg_buffers, msg_memSize, memUnit);
result.append(msg);
}
}
void Command::dumpHierarchy(
const std::vector<MObject>& gpuCacheNodes,
MStringArray& result
) const
{
BOOST_FOREACH(const MObject& gpuCacheObject, gpuCacheNodes) {
MFnDagNode gpuCacheFn(gpuCacheObject);
MPxNode* node = gpuCacheFn.userNode();
assert(node);
assert(dynamic_cast<ShapeNode*>(node));
ShapeNode* gpuCacheNode =
static_cast<ShapeNode*>(node);
SubNode::Ptr rootNode = gpuCacheNode->getCachedGeometry();
if (rootNode) {
DumpHierarchyVisitor visitor(result);
rootNode->accept(visitor);
}
MaterialGraphMap::Ptr materials = gpuCacheNode->getCachedMaterial();
if (materials) {
DumpMaterialVisitor visitor(result);
visitor.dumpMaterials(materials);
}
}
}
MStatus Command::dumpHierarchyToFile(
const std::vector<MObject>& gpuCacheNodes,
const MFileObject& file
) const
{
MStringArray result;
dumpHierarchy(gpuCacheNodes, result);
std::ofstream output(file.resolvedFullName().asChar());
if (!output.is_open()) {
MStatus stat;
MString fmt = MStringResource::getString(kCouldNotSaveFileMsg, stat);
MString msg;
msg.format(fmt, file.resolvedFullName());
return MS::kFailure;
}
for (unsigned int i = 0; i < result.length(); i++) {
output << result[i].asChar() << std::endl;
}
output.close();
return MS::kSuccess;
}
void Command::showAnimTimeRange(
const std::vector<MObject>& gpuCacheNodes,
MDoubleArray& result
) const
{
TimeInterval animTimeRange(TimeInterval::kInvalid);
BOOST_FOREACH(const MObject& node, gpuCacheNodes) {
MFnDagNode dagNode(node);
if (dagNode.typeId() != ShapeNode::id) {
continue;
}
ShapeNode* userNode = dynamic_cast<ShapeNode*>(dagNode.userNode());
if (userNode == NULL) {
continue;
}
const SubNode::Ptr topNode = userNode->getCachedGeometry();
if (userNode->backgroundReadingState() != CacheFileEntry::kReadingDone) {
// Background reading in progress but we need the animation time
// range information immediately.
MString cacheFileName = MPlug(node, ShapeNode::aCacheFileName).asString();
MFileObject cacheFile;
cacheFile.setRawFullName(cacheFileName);
if (cacheFileName.length() > 0 && cacheFile.exists()) {
// Temporarily pause the worker thread and read the time range.
ScopedPauseWorkerThread pause;
GlobalReaderCache::CacheReaderProxy::Ptr proxy =
GlobalReaderCache::theCache().getCacheReaderProxy(cacheFile);
GlobalReaderCache::CacheReaderHolder holder(proxy);
boost::shared_ptr<CacheReader> reader = holder.getCacheReader();
if (reader && reader->valid()) {
TimeInterval interval(TimeInterval::kInvalid);
if (reader->readAnimTimeRange(interval)) {
animTimeRange |= interval;
}
}
}
}
else if (topNode) {
const SubNodeData::Ptr data = topNode->getData();
if (data) {
animTimeRange |= data->animTimeRange();
}
}
}
result.setLength(2);
result[0] = MTime(animTimeRange.startTime(), MTime::kSeconds).as(MTime::uiUnit());
result[1] = MTime(animTimeRange.endTime(), MTime::kSeconds).as(MTime::uiUnit());
}
void Command::refresh(const std::vector<MObject>& gpuCacheNodes)
{
BOOST_FOREACH(const MObject& node, gpuCacheNodes) {
MFnDagNode dagNode(node);
if (dagNode.typeId() != ShapeNode::id) {
continue;
}
ShapeNode* userNode = dynamic_cast<ShapeNode*>(dagNode.userNode());
if (userNode == NULL) {
continue;
}
userNode->refreshCachedGeometry( true );
}
// Schedule an idle refresh. A normal refresh will cause the Alembic file to be
// loaded immediately. We want this load operation to happen later.
}
}
void Command::refreshAll()
{
// Clear the CacheFileRegistry
CacheFileRegistry::theCache().clear();
// Force a refresh on all ShapeNodes
std::vector<MObjectHandle> shapes;
CacheShapeRegistry::theCache().getAll( shapes );
for( size_t i = 0; i < shapes.size(); i++ )
{
if( !shapes[i].isValid() )
{
continue;
}
nodeFn.setObject( shapes[i].object() );
assert( nodeFn.typeId() == ShapeNode::id );
ShapeNode* shape = (ShapeNode*) nodeFn.userNode();
assert( shape );
// File cache has already been cleared, do not request clearFileCache
shape->refreshCachedGeometry( false );
}
// Schedule an idle refresh. A normal refresh will cause the Alembic file to be
// loaded immediately. We want this load operation to happen later.
}
}
void Command::listFileEntries()
{
MStringArray output;
std::vector<CacheFileEntry::MPtr> entries;
CacheFileRegistry::theCache().getAll(entries);
size_t nEntries = entries.size();
for( size_t i = 0; i < nEntries; i++ )
{
output.append( entries[i]->fResolvedCacheFileName );
}
}
void Command::listShapeEntries()
{
MStringArray output;
std::vector<MObjectHandle> shapes;
CacheShapeRegistry::theCache().getAll(shapes);
size_t nShapes = shapes.size();
for( size_t i = 0; i < nShapes; i++ )
{
MString str;
MObject obj = shapes[i].object();
MStatus stat = nodeFn.setObject(obj);
if( stat )
{
ShapeNode* shapeNode = (ShapeNode*) nodeFn.userNode();
CacheFileEntry::MPtr entry = shapeNode->getCacheFileEntry();
str += nodeFn.name();
str += ":";
if( entry )
{
str += entry->fResolvedCacheFileName;
}
}
else
{
str += "kNullObj:";
}
output.append( str );
}
}
}