MStatus sseDeformer::compute(const MPlug& plug, MDataBlock& data)
{
MStatus status;
if (plug.attribute() != outputGeom) {
printf("Ignoring requested plug\n");
return status;
}
unsigned int index = plug.logicalIndex();
MObject thisNode = this->thisMObject();
// get input value
MPlug inPlug(thisNode,input);
inPlug.selectAncestorLogicalIndex(index,input);
MDataHandle hInput = data.inputValue(inPlug, &status);
MCheckStatus(status, "ERROR getting input mesh\n");
// get the input geometry
MDataHandle inputData = hInput.child(inputGeom);
if (inputData.type() != MFnData::kMesh) {
printf("Incorrect input geometry type\n");
return MStatus::kFailure;
}
MObject iSurf = inputData.asMesh() ;
MFnMesh inMesh;
inMesh.setObject( iSurf ) ;
MDataHandle outputData = data.outputValue(plug);
outputData.copy(inputData);
if (outputData.type() != MFnData::kMesh) {
printf("Incorrect output mesh type\n");
return MStatus::kFailure;
}
MObject oSurf = outputData.asMesh() ;
if(oSurf.isNull()) {
printf("Output surface is NULL\n");
return MStatus::kFailure;
}
MFnMesh outMesh;
outMesh.setObject( oSurf ) ;
MCheckStatus(status, "ERROR setting points\n");
// get all points at once for demo purposes. Really should get points from the current group using iterator
MFloatPointArray pts;
outMesh.getPoints(pts);
int nPoints = pts.length();
MDataHandle envData = data.inputValue(envelope, &status);
float env = envData.asFloat();
MDataHandle sseData = data.inputValue(sseEnabled, &status);
bool sseEnabled = (bool) sseData.asBool();
// NOTE: Using MTimer and possibly other classes disables
// autovectorization with Intel <=10.1 compiler on OSX and Linux!!
// Must compile this function with -fno-exceptions on OSX and
// Linux to guarantee autovectorization is done. Use -fvec_report2
// to check for vectorization status messages with Intel compiler.
MTimer timer;
timer.beginTimer();
if(sseEnabled) {
// Innter loop will autovectorize. Around 3x faster than the
// loop below it. It would be faster if first element was
// guaranteed to be aligned on 16 byte boundary.
for(int i=0; i<nPoints; i++) {
float* ptPtr = &pts[i].x;
for(int j=0; j<4; j++) {
ptPtr[j] = env * (cosf(ptPtr[j]) * sinf(ptPtr[j]) * tanf(ptPtr[j]));
}
}
} else {
// This inner loop will not autovectorize.
for(int i=0; i<nPoints; i++) {
MFloatPoint& pt = pts[i];
for(int j=0; j<3; j++) {
pt[j] = env * (cosf(pt[j]) * sinf(pt[j]) * tanf(pt[j]));
}
}
}
timer.endTimer();
if(sseEnabled) {
printf("SSE enabled, runtime %f\n", timer.elapsedTime());
} else {
printf("SSE disabled, runtime %f\n", timer.elapsedTime());
}
outMesh.setPoints(pts);
return status;
}
MStatus AbcBullet::doIt(const MArgList & args)
{
try
{
MStatus status;
MTime oldCurTime = MAnimControl::currentTime();
MArgParser argData(syntax(), args, &status);
if (argData.isFlagSet("help"))
{
MGlobal::displayInfo(util::getHelpText());
return MS::kSuccess;
}
bool verbose = argData.isFlagSet("verbose");
// If skipFrame is true, when going through the playback range of the
// scene, as much frames are skipped when possible. This could cause
// a problem for, time dependent solutions like
// particle system / hair simulation
bool skipFrame = true;
if (argData.isFlagSet("dontSkipUnwrittenFrames"))
skipFrame = false;
double startEvaluationTime = DBL_MAX;
if (argData.isFlagSet("preRollStartFrame"))
{
double startAt = 0.0;
argData.getFlagArgument("preRollStartFrame", 0, startAt);
startEvaluationTime = startAt;
}
unsigned int jobSize = argData.numberOfFlagUses("jobArg");
if (jobSize == 0)
return status;
// the frame range we will be iterating over for all jobs,
// includes frames which are not skipped and the startAt offset
std::set<double> allFrameRange;
// this will eventually hold only the animated jobs.
// its a list because we will be removing jobs from it
std::list < AbcWriteJobPtr > jobList;
for (unsigned int jobIndex = 0; jobIndex < jobSize; jobIndex++)
{
JobArgs jobArgs;
MArgList jobArgList;
argData.getFlagArgumentList("jobArg", jobIndex, jobArgList);
MString jobArgsStr = jobArgList.asString(0);
MStringArray jobArgsArray;
jobArgs.verbose = verbose;
{
// parse the job arguments
// e.g. -perFrameCallbackMel "print \"something\"" will be splitted to
// [0] -perFrameCallbackMel
// [1] print "something"
enum State {
kArgument, // parsing an argument (not quoted)
kDoubleQuotedString, // parsing a double quoted string
kSingleQuotedString, // parsing a single quoted string
};
State state = kArgument;
MString stringBuffer;
for (unsigned int charIdx = 0; charIdx < jobArgsStr.numChars();
charIdx++)
{
MString ch = jobArgsStr.substringW(charIdx, charIdx);
switch (state)
{
case kArgument:
if (ch == " ")
{
// space terminates the current argument
if (stringBuffer.length() > 0) {
jobArgsArray.append(stringBuffer);
stringBuffer.clear();
}
// goto another argument
state = kArgument;
}
else if (ch == "\"")
{
if (stringBuffer.length() > 0)
{
// double quote is part of the argument
stringBuffer += ch;
}
else
{
// goto double quoted string
state = kDoubleQuotedString;
}
}
else if (ch == "'")
{
if (stringBuffer.length() > 0)
{
// single quote is part of the argument
stringBuffer += ch;
}
else
{
// goto single quoted string
state = kSingleQuotedString;
}
}
else
{
stringBuffer += ch;
}
break;
case kDoubleQuotedString:
// double quote terminates the current string
if (ch == "\"")
{
jobArgsArray.append(stringBuffer);
stringBuffer.clear();
state = kArgument;
}
else if (ch == "\\")
{
// escaped character
MString nextCh = (++charIdx < jobArgsStr.numChars())
? jobArgsStr.substringW(charIdx, charIdx) : "\\";
if (nextCh == "n") stringBuffer += "\n";
else if (nextCh == "t") stringBuffer += "\t";
else if (nextCh == "r") stringBuffer += "\r";
else if (nextCh == "\\") stringBuffer += "\\";
else if (nextCh == "'") stringBuffer += "'";
else if (nextCh == "\"") stringBuffer += "\"";
else stringBuffer += nextCh;
}
else
{
stringBuffer += ch;
}
break;
case kSingleQuotedString:
// single quote terminates the current string
if (ch == "'")
{
jobArgsArray.append(stringBuffer);
stringBuffer.clear();
state = kArgument;
}
else if (ch == "\\")
{
// escaped character
MString nextCh = (++charIdx < jobArgsStr.numChars())
? jobArgsStr.substringW(charIdx, charIdx) : "\\";
if (nextCh == "n") stringBuffer += "\n";
else if (nextCh == "t") stringBuffer += "\t";
else if (nextCh == "r") stringBuffer += "\r";
else if (nextCh == "\\") stringBuffer += "\\";
else if (nextCh == "'") stringBuffer += "'";
else if (nextCh == "\"") stringBuffer += "\"";
else stringBuffer += nextCh;
}
else
{
stringBuffer += ch;
}
break;
}
}
// the rest of the argument
if (stringBuffer.length() > 0)
{
jobArgsArray.append(stringBuffer);
}
}
double startTime = oldCurTime.value();
double endTime = oldCurTime.value();
double strideTime = 1.0;
bool hasRange = false;
bool hasRoot = false;
std::set <double> shutterSamples;
std::string fileName;
unsigned int numJobArgs = jobArgsArray.length();
for (unsigned int i = 0; i < numJobArgs; ++i)
{
MString arg = jobArgsArray[i];
arg.toLowerCase();
if (arg == "-f" || arg == "-file")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgFile, status );
MGlobal::displayError(err);
return MS::kFailure;
}
fileName = jobArgsArray[++i].asChar();
}
else if (arg == "-fr" || arg == "-framerange")
{
if (i+2 >= numJobArgs || !jobArgsArray[i+1].isDouble() ||
!jobArgsArray[i+2].isDouble())
{
MString err = MStringResource::getString( kInvalidArgFrameRange, status );
MGlobal::displayError(err);
return MS::kFailure;
}
hasRange = true;
startTime = jobArgsArray[++i].asDouble();
endTime = jobArgsArray[++i].asDouble();
// make sure start frame is smaller or equal to endTime
if (startTime > endTime)
{
double temp = startTime;
startTime = endTime;
endTime = temp;
}
}
else if (arg == "-frs" || arg == "-framerelativesample")
{
if (i+1 >= numJobArgs || !jobArgsArray[i+1].isDouble())
{
MString err = MStringResource::getString( kInvalidArgFrameRelativeSample, status );
MGlobal::displayError(err);
return MS::kFailure;
}
shutterSamples.insert(jobArgsArray[++i].asDouble());
}
else if (arg == "-nn" || arg == "-nonormals")
{
jobArgs.noNormals = true;
}
else if (arg == "-ro" || arg == "-renderableonly")
{
jobArgs.excludeInvisible = true;
}
else if (arg == "-s" || arg == "-step")
{
if (i+1 >= numJobArgs || !jobArgsArray[i+1].isDouble())
{
MString err = MStringResource::getString( kInvalidArgStep, status );
MGlobal::displayError(err);
return MS::kFailure;
}
strideTime = jobArgsArray[++i].asDouble();
}
else if (arg == "-sl" || arg == "-selection")
{
jobArgs.useSelectionList = true;
}
else if (arg == "-sn" || arg == "-stripnamespaces")
{
if (i+1 >= numJobArgs || !jobArgsArray[i+1].isUnsigned())
{
// the strip all namespaces case
// so we pick a very LARGE number
jobArgs.stripNamespace = 0xffffffff;
}
else
{
jobArgs.stripNamespace = jobArgsArray[++i].asUnsigned();
}
}
else if (arg == "-uv" || arg == "-uvwrite")
{
jobArgs.writeUVs = true;
}
else if (arg == "-wcs" || arg == "-writecolorsets")
{
jobArgs.writeColorSets = true;
}
else if (arg == "-wfs" || arg == "-writefacesets")
{
jobArgs.writeFaceSets = true;
}
else if (arg == "-ws" || arg == "-worldspace")
{
jobArgs.worldSpace = true;
}
else if (arg == "-wv" || arg == "-writevisibility")
{
jobArgs.writeVisibility = true;
}
else if (arg == "-mfc" || arg == "-melperframecallback")
{
if (i+1 >= numJobArgs)
{
MGlobal::displayError(
"melPerFrameCallback incorrectly specified.");
return MS::kFailure;
}
jobArgs.melPerFrameCallback = jobArgsArray[++i].asChar();
}
else if (arg == "-pfc" || arg == "-pythonperframecallback")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgPythonPerframeCallback, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.pythonPerFrameCallback = jobArgsArray[++i].asChar();
}
else if (arg == "-mpc" || arg == "-melpostjobcallback")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgMelPostJobCallback, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.melPostCallback = jobArgsArray[++i].asChar();
}
else if (arg == "-ppc" || arg == "-pythonpostjobcallback")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgPythonPostJobCallback, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.pythonPostCallback = jobArgsArray[++i].asChar();
}
// geomArbParams - attribute filtering stuff
else if (arg == "-atp" || arg == "-attrprefix")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgAttrPrefix, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.prefixFilters.push_back(jobArgsArray[++i].asChar());
}
else if (arg == "-a" || arg == "-attr")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgAttr, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.attribs.insert(jobArgsArray[++i].asChar());
}
// userProperties - attribute filtering stuff
else if (arg == "-uatp" || arg == "-userattrprefix")
{
if (i+1 >= numJobArgs)
{
MString err = MStringResource::getString( kInvalidArgUserAttrPrefix, status );
MGlobal::displayError(err);
return MS::kFailure;
}
jobArgs.userPrefixFilters.push_back(jobArgsArray[++i].asChar());
}
else if (arg == "-u" || arg == "-userattr")
{
if (i+1 >= numJobArgs)
{
MGlobal::displayError("userAttr incorrectly specified.");
return MS::kFailure;
}
jobArgs.userAttribs.insert(jobArgsArray[++i].asChar());
}
else if (arg == "-rt" || arg == "-root")
{
if (i+1 >= numJobArgs)
{
MGlobal::displayError("root incorrectly specified.");
return MS::kFailure;
}
hasRoot = true;
MString root = jobArgsArray[++i];
MSelectionList sel;
if (sel.add(root) != MS::kSuccess)
{
MString warn = root;
warn += " could not be select, skipping.";
MGlobal::displayWarning(warn);
continue;
}
unsigned int numRoots = sel.length();
for (unsigned int j = 0; j < numRoots; ++j)
{
MDagPath path;
if (sel.getDagPath(j, path) != MS::kSuccess)
{
MString warn = path.fullPathName();
warn += " (part of ";
warn += root;
warn += " ) not a DAG Node, skipping.";
MGlobal::displayWarning(warn);
continue;
}
jobArgs.dagPaths.insert(path);
}
}
else if (arg == "-ef" || arg == "-eulerfilter")
{
jobArgs.filterEulerRotations = true;
}
else
{
MString warn = "Ignoring unsupported flag: ";
warn += jobArgsArray[i];
MGlobal::displayWarning(warn);
}
} // for i
if (fileName == "")
{
MString error = "-file not specified.";
MGlobal::displayError(error);
return MS::kFailure;
}
{
MString fileRule, expandName;
MString alembicFileRule = "alembicCache";
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;
}
// get the path to the alembic file rule
MFileObject directory;
directory.setRawFullName(expandName);
MString directoryName = directory.resolvedFullName();
// make sure the cache folder exists
if (!directory.exists())
{
// create the cache folder
MString createFolderCmd;
createFolderCmd.format("sysFile -md \"^1s\"", directoryName);
MGlobal::executeCommand(createFolderCmd);
}
// resolve the relative path
MFileObject absoluteFile;
absoluteFile.setRawFullName(fileName.c_str());
#if MAYA_API_VERSION < 201300
if (absoluteFile.resolvedFullName() !=
absoluteFile.expandedFullName())
{
#else
if (!MFileObject::isAbsolutePath(fileName.c_str())) {
#endif
// this is a relative path
MString absoluteFileName = directoryName + "/" +
fileName.c_str();
absoluteFile.setRawFullName(absoluteFileName);
fileName = absoluteFile.resolvedFullName().asChar();
}
else
{
fileName = absoluteFile.resolvedFullName().asChar();
}
// check the path must exist before writing
MFileObject absoluteFilePath;
absoluteFilePath.setRawFullName(absoluteFile.path());
if (!absoluteFilePath.exists()) {
MString error;
error.format("Path ^1s does not exist!", absoluteFilePath.resolvedFullName());
MGlobal::displayError(error);
return MS::kFailure;
}
// check the file is used by any AlembicNode in the scene
MItDependencyNodes dgIter(MFn::kPluginDependNode);
for (; !dgIter.isDone(); dgIter.next()) {
MFnDependencyNode alembicNode(dgIter.thisNode());
if (alembicNode.typeName() != "AlembicNode") {
continue;
}
MPlug abcFilePlug = alembicNode.findPlug("abc_File");
if (abcFilePlug.isNull()) {
continue;
}
MFileObject alembicFile;
alembicFile.setRawFullName(abcFilePlug.asString());
if (!alembicFile.exists()) {
continue;
}
if (alembicFile.resolvedFullName() == absoluteFile.resolvedFullName()) {
MString error = "Can't export to an Alembic file which is in use.";
MGlobal::displayError(error);
return MS::kFailure;
}
}
std::ofstream ofs(fileName.c_str());
if (!ofs.is_open()) {
MString error = MString("Can't write to file: ") + fileName.c_str();
MGlobal::displayError(error);
return MS::kFailure;
}
ofs.close();
}
if (shutterSamples.empty())
{
shutterSamples.insert(0.0);
}
if (jobArgs.prefixFilters.empty())
{
jobArgs.prefixFilters.push_back("ABC_");
}
// the list of frames written into the abc file
std::set<double> transSamples;
std::set <double>::const_iterator shutter;
std::set <double>::const_iterator shutterStart = shutterSamples.begin();
std::set <double>::const_iterator shutterEnd = shutterSamples.end();
for (double frame = startTime; frame <= endTime; frame += strideTime)
{
for (shutter = shutterStart; shutter != shutterEnd; ++shutter)
{
double curFrame = *shutter + frame;
transSamples.insert(curFrame);
}
}
if (transSamples.empty())
{
transSamples.insert(startTime);
}
if (jobArgs.dagPaths.size() > 1)
{
// check for validity of the DagPath relationships complexity : n^2
util::ShapeSet::const_iterator m, n;
util::ShapeSet::const_iterator end = jobArgs.dagPaths.end();
for (m = jobArgs.dagPaths.begin(); m != end; )
{
MDagPath path1 = *m;
m++;
for (n = m; n != end; n++)
{
MDagPath path2 = *n;
if (util::isAncestorDescendentRelationship(path1,path2))
{
MString errorMsg = path1.fullPathName();
errorMsg += " and ";
errorMsg += path2.fullPathName();
errorMsg += " have an ancestor relationship.";
MGlobal::displayError(errorMsg);
return MS::kFailure;
}
} // for n
} // for m
}
// no root is specified, and we aren't using a selection
// so we'll try to translate the whole Maya scene by using all
// children of the world as roots.
else if (!hasRoot && !jobArgs.useSelectionList)
{
MSelectionList sel;
#if MAYA_API_VERSION >= 201100
sel.add("|*", true);
#else
// older versions of Maya will not be able to find top level nodes
// within namespaces
sel.add("|*");
#endif
unsigned int numRoots = sel.length();
for (unsigned int i = 0; i < numRoots; ++i)
{
MDagPath path;
sel.getDagPath(i, path);
jobArgs.dagPaths.insert(path);
}
}
else if (hasRoot && jobArgs.dagPaths.empty())
{
MString errorMsg = "No valid root nodes were specified.";
MGlobal::displayError(errorMsg);
return MS::kFailure;
}
else if (jobArgs.useSelectionList)
{
MSelectionList activeList;
MGlobal::getActiveSelectionList(activeList);
if (activeList.length() == 0)
{
MString errorMsg =
"-selection specified but nothing is actively selected.";
MGlobal::displayError(errorMsg);
return MS::kFailure;
}
}
AbcA::TimeSamplingPtr transTime;
std::vector<double> samples;
for (shutter = shutterStart; shutter != shutterEnd; ++shutter)
{
samples.push_back((startTime + *shutter) * util::spf());
}
if (hasRange)
{
transTime.reset(new AbcA::TimeSampling(AbcA::TimeSamplingType(
static_cast<Alembic::Util::uint32_t>(samples.size()),
strideTime * util::spf()), samples));
}
else
{
transTime.reset(new AbcA::TimeSampling());
}
AbcWriteJobPtr job(new AbcWriteJob(fileName.c_str(),
transSamples, transTime, jobArgs));
jobList.push_front(job);
// make sure we add additional whole frames, if we arent skipping
// the inbetween ones
if (!skipFrame && !allFrameRange.empty())
{
double localMin = *(transSamples.begin());
std::set<double>::iterator last = transSamples.end();
last--;
double localMax = *last;
double globalMin = *(allFrameRange.begin());
last = allFrameRange.end();
last--;
double globalMax = *last;
// if the min of our current frame range is beyond
// what we know about, pad a few more frames
if (localMin > globalMax)
{
for (double f = globalMax; f < localMin; f++)
{
allFrameRange.insert(f);
}
}
// if the max of our current frame range is beyond
// what we know about, pad a few more frames
if (localMax < globalMin)
{
for (double f = localMax; f < globalMin; f++)
{
allFrameRange.insert(f);
}
}
}
// right now we just copy over the translation samples since
// they are guaranteed to contain all the geometry samples
allFrameRange.insert(transSamples.begin(), transSamples.end());
}
// add extra evaluation run up, if necessary
if (startEvaluationTime != DBL_MAX && !allFrameRange.empty())
{
double firstFrame = *allFrameRange.begin();
for (double f = startEvaluationTime; f < firstFrame; ++f)
{
allFrameRange.insert(f);
}
}
std::set<double>::iterator it = allFrameRange.begin();
std::set<double>::iterator itEnd = allFrameRange.end();
MComputation computation;
computation.beginComputation();
// loop through every frame in the list, if a job has that frame in it's
// list of transform or shape frames, then it will write out data and
// call the perFrameCallback, if that frame is also the last one it has
// to work on then it will also call the postCallback.
// If it doesn't have this frame, then it does nothing
MTimer timer;
for (; it != itEnd; it++)
{
timer.beginTimer();
MGlobal::viewFrame(*it);
std::list< AbcWriteJobPtr >::iterator j = jobList.begin();
std::list< AbcWriteJobPtr >::iterator jend = jobList.end();
while (j != jend)
{
if (computation.isInterruptRequested())
return MS::kFailure;
bool lastFrame = (*j)->eval(*it);
if (lastFrame)
{
j = jobList.erase(j);
}
else
j++;
}
timer.endTimer();
if (verbose)
{
double frame = *it;
MString info,arg1,arg2;
arg1.set(frame);
arg2.set(timer.elapsedTime());
info.format( "processed frame: ^1s in ^2s seconds", arg1, arg2 );
MGlobal::displayInfo(info);
}
}
computation.endComputation();
// set the time back
MGlobal::viewFrame(oldCurTime);
return MS::kSuccess;
}
catch (Alembic::Util::Exception & e)
{
MString theError("Alembic Exception encountered: ");
theError += e.what();
MGlobal::displayError(theError);
return MS::kFailure;
}
catch (std::exception & e)
{
MString theError("std::exception encountered: ");
theError += e.what();
MGlobal::displayError(theError);
return MS::kFailure;
}
}
void ComparisonStageIR::build_stage() {
assert(!is_tiled() || (is_tiled() && !track_progress()));
// timer is only allowed for serial loops (just use it to get avg iterations per second or something like that)
assert(!time_loop() || (time_loop() && !is_parallelized()));
set_stage_function(create_stage_function());
set_user_function(create_user_function());
// stuff before the loop
// build the return idx
MVar *loop_start = new MVar(MScalarType::get_long_type()); // don't make a constant b/c it should be updateable
loop_start->register_for_delete();
MStatement *set_loop_start = new MStatement(loop_start, MVar::create_constant<long>(0));
set_loop_start->register_for_delete();
MStatement *set_result = new MStatement(get_return_idx(), loop_start);
set_result->register_for_delete();
set_start_block(new MBlock("start"));
get_start_block()->register_for_delete();
get_start_block()->add_expr(set_loop_start);
get_start_block()->add_expr(set_result);
// When we don't parallelize, then make the inner loop's index outside of both the loops rather than within
// the outer loop. This is a hack for llvm because if we have an alloca call within each iteration of the outer loop,
// we will be "leaking" stack space each time that is called, so moving it outside of the loop prevents that.
// However, it makes it hard to work with when we then parallelize because the code sees that inner loop index as a
// free variable that needs to be added to the closure. This is not fun because our index is now a pointer to an index
// and then we would need to update the index by going through the pointer, etc. Basically, it would cause some hacks on the
// LLVM side (and unless this becomes something that is needed in the future, I don't want to deal with it).
// So instead, it is dealt with below. Without parallelization, the inner loop index is initialized outside of the
// nested loop, and then updated to the correct start right before the inner loop begins execution.
// When parallelization is turned on, the inner loop index is made INSIDE the outer loop. This is because the
// parallelized outer loop calls a function every iteration which is the outer loop body, and then within that the
// inner loop is created. alloca is scoped at the function level, so the inner loop index gets a single alloca
// in this function call, and then the inner loop is created.
// This may not be required of other possible back-end languages that we choose, but it will depend on their scoping rules.
//
// TL;DR LLVM has function scoping for allocainst, so if we create the inner loop index as so
// val outer_index...
// for outer_index...
// val inner_index...
// for inner_index...
// every iteration of the outer loop adds space to the stack which isn't released until the function ends. So we want
// val outer_index...
// val inner_index...
// for outer_index...
// for inner_index...
MVar *inner_start = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
MBlock *preallocation_block = create_preallocator();
get_start_block()->add_expr(preallocation_block);
MTimer *timer = nullptr;
timer = new MTimer();
timer->register_for_delete();
MFor *outer_loop_skeleton_1 = nullptr;
MFor *inner_loop_skeleton_1 = nullptr;
MFor *outer_loop_skeleton_2 = nullptr;
MFor *inner_loop_skeleton_2 = nullptr;
MBlock *inner_loop_body = nullptr;
// think of all comparisons as being in an NxM matrix where N is the left input and M is the right input.
// N is the outermost iteration
tile_size_N = MVar::create_constant<long>(2);
tile_size_M = MVar::create_constant<long>(2);
MVar *final_loop_bound;
if (!is_tiled() || !is_tileable()) { // No tiling
// To make sure that the inner loop doesn't get replace with a different bound if parallelizing, copy
// the bound to a different variable and use that
MVar *bound_copy = new MVar(MScalarType::get_long_type());
bound_copy->register_for_delete();
MStatement *set_copy = new MStatement(bound_copy, get_stage_function()->/*get_args()*/get_loaded_args()[3]);
set_copy->register_for_delete();
get_start_block()->add_expr(set_copy);
// loop components
MVar *outer_loop_start = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
outer_loop_skeleton_1 =
create_stage_for_loop(outer_loop_start, MVar::create_constant<long>(1),
get_stage_function()->/*get_args()*/get_loaded_args()[1], false, get_start_block());
if (is_parallelizable() && is_parallelized()) {
outer_loop_skeleton_1->set_exec_type(PARALLEL);
}
MVar *_inner_start = nullptr;
if ((left_input || right_input) && !_force_commutative) {
_inner_start = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
} else {
MAdd *add = new MAdd(outer_loop_skeleton_1->get_loop_index(),
MVar::create_constant<long>(1));
outer_loop_skeleton_1->get_body_block()->add_expr(add);
add->register_for_delete();
_inner_start = add->get_result();
}
if (!time_loop()) {
get_start_block()->add_expr(outer_loop_skeleton_1);
} else {
get_start_block()->add_expr(timer);
timer->get_timer_block()->add_expr(outer_loop_skeleton_1);
}
MStatement *set_inner_start = new MStatement(inner_start, _inner_start);
set_inner_start->register_for_delete();
outer_loop_skeleton_1->get_body_block()->add_expr(set_inner_start);
MBlock *temp_block = new MBlock();
temp_block->register_for_delete();
inner_loop_skeleton_1 = create_stage_for_loop(inner_start, MVar::create_constant<long>(1), bound_copy, true,
temp_block);
// TODO hack, need to add the loop index initialization before the outer loop, but we have to add the outer loop before this since
// the inner_start depends on the outer loop
get_start_block()->insert_at(temp_block, get_start_block()->get_exprs().size() - 2); // insert right before the outer loop
// stuff for calling the user function in the loop
inner_loop_body = inner_loop_skeleton_1->get_body_block();
} else if (is_tiled() && is_tileable()) { // tiling
// loop components
MDiv *_outer_1_bound =
new MDiv(get_stage_function()->/*get_args()*/get_loaded_args()[1], tile_size_N);
_outer_1_bound->register_for_delete();
MDiv *_inner_1_bound =
new MDiv(get_stage_function()->/*get_args()*/get_loaded_args()[3], tile_size_M);
_inner_1_bound->register_for_delete();
// compensate for when the number of elements isn't a multiple of the tile size
MAdd *outer_1_bound = new MAdd(_outer_1_bound->get_result(), MVar::create_constant<long>(1));
outer_1_bound->register_for_delete();
MAdd *inner_1_bound = new MAdd(_inner_1_bound->get_result(), MVar::create_constant<long>(1));
inner_1_bound->register_for_delete();
get_start_block()->add_expr(_outer_1_bound);
get_start_block()->add_expr(_inner_1_bound);
get_start_block()->add_expr(outer_1_bound);
get_start_block()->add_expr(inner_1_bound);
MVar *outer_loop_start_1 = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
outer_loop_start_1->override_name("outer_loop_start_1");
MVar *inner_loop_start_1 = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
inner_loop_start_1->override_name("inner_loop_start_1");
MVar *outer_loop_start_2 = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
outer_loop_start_2->override_name("outer_loop_start_2");
MVar *inner_loop_start_2 = initialize<long>(MScalarType::get_long_type(), 0, get_start_block());
inner_loop_start_2->override_name("inner_loop_start_2");
// n = 0 to N/tile_size_N + 1
outer_loop_skeleton_1 =
create_stage_for_loop(outer_loop_start_1, MVar::create_constant<long>(1),
outer_1_bound->get_result(), true, get_start_block());
outer_loop_skeleton_1->override_name("outer_loop_skeleton1");
//
// if (!time_loop()) {
// get_start_block()->add_expr(outer_loop_skeleton_1);
// } else {
// get_start_block()->add_expr(timer);
// timer->get_timer_block()->add_expr(outer_loop_skeleton_1);
// }
// m = 0 to M/tile_size_M + 1
inner_loop_skeleton_1 =
create_stage_for_loop(inner_loop_start_1, MVar::create_constant<long>(1),
inner_1_bound->get_result(), true, get_start_block());
inner_loop_skeleton_1->override_name("inner_loop_skeleton1");
// nn = 0 to tile_size_N
outer_loop_skeleton_2 = create_stage_for_loop(outer_loop_start_2, MVar::create_constant<long>(1),
tile_size_N, true, get_start_block());
outer_loop_skeleton_2->override_name("outer_loop_skeleton2");
// mm = 0 to tile_size_M
inner_loop_skeleton_2 = create_stage_for_loop(inner_loop_start_2, MVar::create_constant<long>(1),
tile_size_M, true, get_start_block());
inner_loop_skeleton_2->override_name("inner_loop_skeleton2");
if (!time_loop()) {
get_start_block()->add_expr(outer_loop_skeleton_1);
} else {
get_start_block()->add_expr(timer);
timer->get_timer_block()->add_expr(outer_loop_skeleton_1);
}
inner_loop_skeleton_1->get_body_block()->add_expr(outer_loop_skeleton_2);
outer_loop_skeleton_2->get_body_block()->add_expr(inner_loop_skeleton_2);
inner_loop_body = inner_loop_skeleton_2->get_body_block();
}
MBlock *user_arg_block;
std::vector<MVar *> args = create_user_function_inputs(&user_arg_block, outer_loop_skeleton_1, outer_loop_skeleton_2,
inner_loop_skeleton_1, inner_loop_skeleton_2, nullptr, false,
nullptr, nullptr, get_stage_function()->/*get_args()*/get_loaded_args()[1],
get_stage_function()->/*get_args()*/get_loaded_args()[3]);
if (!is_tiled() || !is_tileable()) {
inner_loop_body->add_expr(user_arg_block);
} // if tiled, this is already added in the create_user_function_inputs
inner_loop_body = user_arg_block;
int bucket_idx = inner_loop_body->get_exprs().size();
MFunctionCall *call = call_user_function(get_user_function(), args);
inner_loop_body->add_expr(call);
// handle the output of the user call
MBlock *processed_call = process_user_function_call(call, NULL, false);
inner_loop_body->add_expr(processed_call);
// do any other postprocessing needed in the loop before the next iteration
MBlock *extra = loop_extras();
inner_loop_body->add_expr(extra);
if (track_progress() && !is_parallelized()) {
// still return the original loop bound
MBlock *temp = new MBlock();
temp->register_for_delete();
final_loop_bound = outer_loop_skeleton_1->get_loop_bound();
outer_loop_skeleton_1->get_body_block()->add_expr(inner_loop_skeleton_1);
inner_loop_body->insert_at(apply_buckets(args[0], args[1], inner_loop_skeleton_2 ? inner_loop_skeleton_2 : inner_loop_skeleton_1), bucket_idx);
std::pair<MFor *, MFor *> splits = ProgressTracker::create_progress_tracker(outer_loop_skeleton_1,
inner_loop_skeleton_1,
get_num_tracking_splits(), temp,
true);
// find the original outer_loop_skeleton_1 in the block and remove it. Then replace with the new one in splits.first
int idx = 0;
if (!time_loop()) {
for (std::vector<MExpr *>::const_iterator iter = get_start_block()->get_exprs().cbegin();
iter != get_start_block()->get_exprs().cend(); iter++) {
if (*iter == outer_loop_skeleton_1) {
break;
}
idx++;
}
get_start_block()->remove_at(idx);
} else {
for (std::vector<MExpr *>::const_iterator iter = timer->get_timer_block()->get_exprs().cbegin();
iter != timer->get_timer_block()->get_exprs().cend(); iter++) {
if (*iter == outer_loop_skeleton_1) {
break;
}
idx++;
}
timer->get_timer_block()->remove_at(idx);
}
outer_loop_skeleton_1 = splits.first;
// do the replacement
// outer_loop_skeleton_1 added to temp block in the progress tracker function
if (!time_loop()) {
get_stage_function()->add_body_block(temp);
} else {
timer->get_timer_block()->insert_at(temp, idx);
}
} else {
outer_loop_skeleton_1->get_body_block()->add_expr(inner_loop_skeleton_1);
final_loop_bound = outer_loop_skeleton_1->get_loop_bound();
inner_loop_body->insert_at(apply_buckets(args[0], args[1], inner_loop_skeleton_2 ? inner_loop_skeleton_2 : inner_loop_skeleton_1), bucket_idx);
}
// modify this loop if it needs to be parallelized
if (is_parallelizable() && is_parallelized()) {
parallelize_main_loop(get_start_block(), outer_loop_skeleton_1, inner_loop_skeleton_1);
}
//
// if (is_tiled() && is_tileable()) {
// inner_loop_skeleton_1->get_body_block()->add_expr(outer_loop_skeleton_2);
// outer_loop_skeleton_2->get_body_block()->add_expr(inner_loop_skeleton_2);
// }
// postprocessing after the outer loop is done (no postprocessing needed after the inner loop since it just goes back to the outer loop)
MBlock *after_loop = time_loop() ? timer->get_after_timer_block() : outer_loop_skeleton_1->get_end_block();
MBlock *finished = finish_stage(nullptr, final_loop_bound);
MBlock *deletion = delete_fields();
after_loop->add_expr(deletion);
after_loop->add_expr(finished);
get_stage_function()->insert_body_block_at(get_start_block(), 1); // insert before the temp block, which would have been added if doing tracking. Insert after the stage arg loading though.
// the temp block has the loop now, so it can't come before everything else
}
MStatus splatDeformer::compute(const MPlug& plug, MDataBlock& data)
{
// do this if we are using an OpenMP implementation that is not the same as Maya's.
// Even if it is the same, it does no harm to make this call.
MThreadUtils::syncNumOpenMPThreads();
MStatus status = MStatus::kUnknownParameter;
if (plug.attribute() != outputGeom) {
return status;
}
unsigned int index = plug.logicalIndex();
MObject thisNode = this->thisMObject();
// get input value
MPlug inPlug(thisNode,input);
inPlug.selectAncestorLogicalIndex(index,input);
MDataHandle hInput = data.inputValue(inPlug, &status);
MCheckStatus(status, "ERROR getting input mesh\n");
// get the input geometry
MDataHandle inputData = hInput.child(inputGeom);
if (inputData.type() != MFnData::kMesh) {
printf("Incorrect input geometry type\n");
return MStatus::kFailure;
}
// get the input groupId - ignored for now...
MDataHandle hGroup = inputData.child(groupId);
unsigned int groupId = hGroup.asLong();
// get deforming mesh
MDataHandle deformData = data.inputValue(deformingMesh, &status);
MCheckStatus(status, "ERROR getting deforming mesh\n");
if (deformData.type() != MFnData::kMesh) {
printf("Incorrect deformer geometry type %d\n", deformData.type());
return MStatus::kFailure;
}
MObject dSurf = deformData.asMeshTransformed();
MFnMesh fnDeformingMesh;
fnDeformingMesh.setObject( dSurf ) ;
MDataHandle outputData = data.outputValue(plug);
outputData.copy(inputData);
if (outputData.type() != MFnData::kMesh) {
printf("Incorrect output mesh type\n");
return MStatus::kFailure;
}
MItGeometry iter(outputData, groupId, false);
// create fast intersector structure
MMeshIntersector intersector;
intersector.create(dSurf);
// get all points at once. Faster to query, and also better for
// threading than using iterator
MPointArray verts;
iter.allPositions(verts);
int nPoints = verts.length();
// use bool variable as lightweight object for failure check in loop below
bool failed = false;
MTimer timer; timer.beginTimer();
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(int i=0; i<nPoints; i++) {
// Cannot break out of an OpenMP loop, so if one of the
// intersections failed, skip the rest
if(failed) continue;
// mesh point object must be in loop-local scope to avoid race conditions
MPointOnMesh meshPoint;
// Do intersection. Need to use per-thread status value as
// MStatus has internal state and may trigger race conditions
// if set from multiple threads. Probably benign in this case,
// but worth being careful.
MStatus localStatus = intersector.getClosestPoint(verts[i], meshPoint);
if(localStatus != MStatus::kSuccess) {
// NOTE - we cannot break out of an OpenMP region, so set
// bad status and skip remaining iterations
failed = true;
continue;
}
// default OpenMP scheduling breaks traversal into large
// chunks, so low risk of false sharing here in array write.
verts[i] = meshPoint.getPoint();
}
timer.endTimer(); printf("Runtime for threaded loop %f\n", timer.elapsedTime());
// write values back onto output using fast set method on iterator
iter.setAllPositions(verts);
if(failed) {
printf("Closest point failed\n");
return MStatus::kFailure;
}
return status;
}
MStatus finalproject::compute(const MPlug& plug, MDataBlock& data)
{
// do this if we are using an OpenMP implementation that is not the same as Maya's.
// Even if it is the same, it does no harm to make this call.
MThreadUtils::syncNumOpenMPThreads();
MStatus status = MStatus::kUnknownParameter;
if (plug.attribute() != outputGeom) {
return status;
}
unsigned int index = plug.logicalIndex();
MObject thisNode = this->thisMObject();
// get input value
MPlug inPlug(thisNode,input);
inPlug.selectAncestorLogicalIndex(index,input);
MDataHandle hInput = data.inputValue(inPlug, &status);
MCheckStatus(status, "ERROR getting input mesh\n");
// get the input geometry
MDataHandle inputData = hInput.child(inputGeom);
if (inputData.type() != MFnData::kMesh) {
printf("Incorrect input geometry type\n");
return MStatus::kFailure;
}
// get the input groupId - ignored for now...
MDataHandle hGroup = inputData.child(groupId);
unsigned int groupId = hGroup.asLong();
// get deforming mesh
MDataHandle deformData = data.inputValue(deformingMesh, &status);
MCheckStatus(status, "ERROR getting deforming mesh\n");
if (deformData.type() != MFnData::kMesh) {
printf("Incorrect deformer geometry type %d\n", deformData.type());
return MStatus::kFailure;
}
MDataHandle offloadData = data.inputValue(offload, &status);
//gathers world space positions of the object and the magnet
MObject dSurf = deformData.asMeshTransformed();
MObject iSurf = inputData.asMeshTransformed();
MFnMesh fnDeformingMesh, fnInputMesh;
fnDeformingMesh.setObject( dSurf ) ;
fnInputMesh.setObject( iSurf ) ;
MDataHandle outputData = data.outputValue(plug);
outputData.copy(inputData);
if (outputData.type() != MFnData::kMesh) {
printf("Incorrect output mesh type\n");
return MStatus::kFailure;
}
MItGeometry iter(outputData, groupId, false);
// get all points at once. Faster to query, and also better for
// threading than using iterator
MPointArray objVerts;
iter.allPositions(objVerts);
int objNumPoints = objVerts.length();
MPointArray magVerts, tempverts;
fnDeformingMesh.getPoints(magVerts);
fnInputMesh.getPoints(tempverts);
int magNumPoints = magVerts.length();
double min = DBL_MAX, max = -DBL_MAX;
//finds min and max z-coordinate values to determine middle point (choice of z-axis was ours)
for (int i = 0; i < magNumPoints; i++) {
min = magVerts[i].z < min ? magVerts[i].z : min;
max = magVerts[i].z > max ? magVerts[i].z : max;
}
double middle = (min + max) / 2;
double polarity[magNumPoints];
//assigns polarity based on middle point of mesh
for (int i = 0; i < magNumPoints; i++) {
polarity[i] = magVerts[i].z > middle ? max / magVerts[i].z : -min / magVerts[i].z;
}
double* objdVerts = (double *)malloc(sizeof(double) * objNumPoints * 3);
double* magdVerts = (double *)malloc(sizeof(double) * magNumPoints * 3);
//creates handles to use attribute data
MDataHandle vecX = data.inputValue(transX, &status);
MDataHandle vecY = data.inputValue(transY, &status);
MDataHandle vecZ = data.inputValue(transZ, &status);
//gathers previously stored coordinates of the center of the object
double moveX = vecX.asFloat();
double moveY = vecY.asFloat();
double moveZ = vecZ.asFloat();
//translates object based on the position stored in the attribute values
for (int i=0; i<objNumPoints; i++) {
objdVerts[i * 3] = tempverts[i].x + moveX;
objdVerts[i * 3 + 1] = tempverts[i].y + moveY;
objdVerts[i * 3 + 2] = tempverts[i].z + moveZ;
}
for (int i=0; i<magNumPoints; i++) {
magdVerts[i * 3] = magVerts[i].x;
magdVerts[i * 3 + 1] = magVerts[i].y;
magdVerts[i * 3 + 2] = magVerts[i].z;
}
double teslaData = data.inputValue(tesla, &status).asDouble();
MDataHandle posiData = data.inputValue(positivelycharged, &status);
double pivot[6] = {DBL_MAX, -DBL_MAX, DBL_MAX, -DBL_MAX, DBL_MAX, -DBL_MAX};
//finds the pivot point of the object in world space prior to being affected by the magnet
for (int i = 0; i < tempverts.length(); i++) {
pivot[0] = tempverts[i].x < pivot[0] ? tempverts[i].x : pivot[0];
pivot[1] = tempverts[i].x > pivot[1] ? tempverts[i].x : pivot[1];
pivot[2] = tempverts[i].y < pivot[2] ? tempverts[i].y : pivot[2];
pivot[3] = tempverts[i].y > pivot[3] ? tempverts[i].y : pivot[3];
pivot[4] = tempverts[i].z < pivot[4] ? tempverts[i].z : pivot[4];
pivot[5] = tempverts[i].z > pivot[5] ? tempverts[i].z : pivot[5];
}
MTimer timer; timer.beginTimer();
//main function call
magnetForce(magNumPoints, objNumPoints, teslaData, magdVerts,
objdVerts, polarity, posiData.asBool(), offloadData.asBool());
timer.endTimer(); printf("Runtime for threaded loop %f\n", timer.elapsedTime());
for (int i=0; i<objNumPoints; i++) {
objVerts[i].x = objdVerts[i * 3 + 0];
objVerts[i].y = objdVerts[i * 3 + 1];
objVerts[i].z = objdVerts[i * 3 + 2];
}
//finds the pivot point of object in world space after being affected by the magnet
double objCenter[6] = {DBL_MAX, -DBL_MAX, DBL_MAX, -DBL_MAX, DBL_MAX, -DBL_MAX};
for (int i = 0; i < tempverts.length(); i++) {
objCenter[0] = objVerts[i].x < objCenter[0] ? objVerts[i].x : objCenter[0];
objCenter[1] = objVerts[i].x > objCenter[1] ? objVerts[i].x : objCenter[1];
objCenter[2] = objVerts[i].y < objCenter[2] ? objVerts[i].y : objCenter[2];
objCenter[3] = objVerts[i].y > objCenter[3] ? objVerts[i].y : objCenter[3];
objCenter[4] = objVerts[i].z < objCenter[4] ? objVerts[i].z : objCenter[4];
objCenter[5] = objVerts[i].z > objCenter[5] ? objVerts[i].z : objCenter[5];
}
//creates vector based on the two calculated pivot points
moveX = (objCenter[0] + objCenter[1]) / 2 - (pivot[0] + pivot[1]) / 2;
moveY = (objCenter[2] + objCenter[3]) / 2 - (pivot[2] + pivot[3]) / 2;
moveZ = (objCenter[4] + objCenter[5]) / 2 - (pivot[4] + pivot[5]) / 2;
//stores pivot vector for next computation
if (teslaData) {
vecX.setFloat(moveX);
vecY.setFloat(moveY);
vecZ.setFloat(moveZ);
}
// write values back onto output using fast set method on iterator
iter.setAllPositions(objVerts, MSpace::kWorld);
free(objdVerts);
free(magdVerts);
return status;
}