Skip to content

GitLab

C
CGoGN
Commit c5be02f7 authored by Sylvain Thery's avatar Sylvain Thery
Browse files
Options

Optimize & simplify parallel foreach 

- use automatically orbit or cell & quicktraversal
- DartMarker & CellMarker now thread safe
parent 0c4ea341
Expand all Show whitespace changes
Inline Side-by-side
Showing with 792 additions and 1009 deletions
Apps/Tuto/tuto_mt.cpp
View file @ c5be02f7
...@@ -33,16 +33,14 @@ ...@@ -33,16 +33,14 @@
#include "Algo/Import/import.h" #include "Algo/Import/import.h"
#include "Algo/Geometry/boundingbox.h" #include "Algo/Geometry/boundingbox.h"
#include "Algo/Render/GL1/map_glRender.h"
#include "Utils/GLSLShader.h" #include "Utils/GLSLShader.h"
#include "Algo/Geometry/area.h" //#include "Algo/Geometry/area.h"
#include "Algo/Geometry/normal.h" #include "Algo/Geometry/normal.h"
#include "Algo/Modelisation/polyhedron.h" #include "Algo/Modelisation/polyhedron.h"
#include "Algo/Parallel/parallel_foreach.h" #include "Algo/Parallel/parallel_foreach.h"
#include "Algo/Parallel/cgogn_thread.h"
// for file input #include "Utils/chrono.h"
#include "Utils/qtInputs.h"
using namespace CGoGN ; using namespace CGoGN ;
...@@ -56,356 +54,312 @@ struct PFP: public PFP_STANDARD ...@@ -56,356 +54,312 @@ struct PFP: public PFP_STANDARD
typedef EmbeddedMap2 MAP; typedef EmbeddedMap2 MAP;
}; };
// declaration of the map
PFP::MAP myMap;
// attribute handlers PFP::MAP myMap;
VertexAttribute<PFP::VEC3> position; VertexAttribute<PFP::VEC3> position;
VertexAttribute<PFP::VEC3> position2;
VertexAttribute<PFP::VEC3> normal; VertexAttribute<PFP::VEC3> normal;
// open file
void MyQT::cb_Open()
{
// set some filters
// std::string filters("all (*.*);; trian (*.trian);; ctm (*.ctm);; off (*.off);; ply (*.ply)");
//
// std::string filename = selectFile("OpenMesh","",filters);
//
// std::vector<std::string> attrNames ;
// if(!Algo::Import::importMesh<PFP>(myMap, filename.c_str(), attrNames))
// {
// CGoGNerr << "could not import " << filename << CGoGNendl ;
// return;
// }
std::vector<std::string> attrNames ;
if(!Algo::Import::importMesh<PFP>(myMap, "/home/thery/Data/liver.trian", attrNames))
{
CGoGNerr << "could not import xxx" << CGoGNendl ;
return;
}
// recuper l'attribut pour la position des points (créé lors de l'import)
position = myMap.getAttribute<PFP::VEC3, VERTEX>(attrNames[0]) ;
if (!normal.isValid())
normal = myMap.addAttribute<PFP::VEC3, VERTEX>("normal");
Algo::Geometry::computeNormalVertices<PFP>(myMap, position, normal) ; template <typename XXX>
class ThreadNormals: public Algo::Parallel::CGoGNThread<typename XXX::MAP>
{
protected:
VertexAttribute<typename XXX::VEC3>& m_positions;
VertexAttribute<typename XXX::VEC3>& m_normals;
public:
ThreadNormals(typename XXX::MAP& map, VertexAttribute<typename XXX::VEC3>& pos, VertexAttribute<typename XXX::VEC3>& norm, unsigned int th):
Algo::Parallel::CGoGNThread<typename XXX::MAP>(map,th),
m_positions(pos),
m_normals(norm)
{}
void operator()()
{
Algo::Geometry::computeNormalVertices<XXX>(this->m_map, m_positions, m_normals, SelectorTrue(), this->tid());
}
};
// bounding box
Geom::BoundingBox<PFP::VEC3> bb = Algo::Geometry::computeBoundingBox<PFP>(myMap, position);
float lWidthObj = std::max<PFP::REAL>(std::max<PFP::REAL>(bb.size(0), bb.size(1)), bb.size(2));
Geom::Vec3f lPosObj = (bb.min() + bb.max()) / PFP::REAL(2);
// envoit info BB a l'interface
setParamObject(lWidthObj,lPosObj.data());
updateGLMatrices();
}
// new void MyQT::cb_initGL()
void MyQT::cb_New()
{ {
if (!position.isValid()) Utils::GLSLShader::setCurrentOGLVersion(2);
position = myMap.addAttribute<PFP::VEC3, VERTEX>("position");
// create a sphere // create the render
Algo::Modelisation::Polyhedron<PFP> prim(myMap, position); m_render = new Algo::Render::GL2::MapRender();
prim.cylinder_topo(16,16, true, true); // topo of sphere is a closed cylinder
prim.embedSphere(10.0f);
if (!normal.isValid()) // create VBO for position
normal = myMap.addAttribute<PFP::VEC3, VERTEX>("normal"); m_positionVBO = new Utils::VBO();
m_positionVBO->updateData(position);
Algo::Geometry::computeNormalVertices<PFP>(myMap, position, normal) ; m_normalVBO = new Utils::VBO();
// bounding box m_shader = new Utils::ShaderSimpleColor();
Geom::BoundingBox<PFP::VEC3> bb = Algo::Geometry::computeBoundingBox<PFP>(myMap, position); m_shader->setAttributePosition(m_positionVBO);
float lWidthObj = std::max<PFP::REAL>(std::max<PFP::REAL>(bb.size(0), bb.size(1)), bb.size(2)); m_shader->setColor(Geom::Vec4f(1.,1.,0.,0.));
Geom::Vec3f lPosObj = (bb.min() + bb.max()) / PFP::REAL(2);
setParamObject(lWidthObj,lPosObj.data());
updateGLMatrices(); m_lines = new Utils::ShaderVectorPerVertex();
} m_lines->setAttributePosition(m_positionVBO);
m_lines->setAttributeVector(m_normalVBO);
m_lines->setScale(2.0f);
m_lines->setColor(Geom::Vec4f(0.0f, 1.0f, 0.2f, 0.0f));
std::cout << "Je calcule les normales en meme temps que les primitives" << std::endl;
boost::thread thread1( ThreadNormals<PFP>(myMap,position,normal,1));
void MyQT::cb_initGL()
{
// Old school openGL ;)
Utils::GLSLShader::setCurrentOGLVersion(1);
glewInit(); m_render->initPrimitives<PFP>(myMap, allDarts, Algo::Render::GL2::LINES);
m_render->initPrimitives<PFP>(myMap, allDarts, Algo::Render::GL2::POINTS);
// init lighting parameters registerShader(m_shader);
float lightPosition[4]= {10.0f,10.0f,10000.0f,1.0f}; registerShader(m_lines);
float lightColor[4]= {0.9f,0.9f,0.9f,1.0f};
glLightModeli(GL_LIGHT_MODEL_TWO_SIDE, GL_TRUE); // on attend la fin du thread pour etre sur que normal est a jour
glEnable(GL_LIGHT0); thread1.join();
glLightfv(GL_LIGHT0, GL_DIFFUSE, lightColor); m_normalVBO->updateData(normal);
glLightfv(GL_LIGHT0, GL_POSITION, lightPosition);
glEnable(GL_NORMALIZE);
// glDisable(GL_CULL_FACE);
// glFrontFace(GL_CCW);
} }
void MyQT::cb_redraw() void MyQT::cb_redraw()
{ {
GLfloat diff[4]= {0.0f,1.0f,0.1f,1.0f}; m_render->draw(m_shader, Algo::Render::GL2::LINES);
GLfloat amb[4]= {0.1f,0.0f,0.1f,1.0f}; m_render->draw(m_lines, Algo::Render::GL2::POINTS);
GLfloat spec[4]= {1.0f,1.0f,1.0f,1.0f};
float shininess=125.0f;
// draw the lines
// glDisable(GL_LIGHTING);
// glColor3f(0.0f, 0.0f, 0.3f);
// glPolygonMode(GL_FRONT_AND_BACK,GL_LINE);
// glDisable(GL_LIGHTING);
//
// Algo::Render::GL1::renderTriQuadPoly<PFP>(myMap,Algo::Render::GL1::LINE, 1.0f,position, normal);
// draw the faces
glEnable(GL_POLYGON_OFFSET_FILL);
glPolygonOffset(1.0f, 1.0f);
glEnable(GL_LIGHTING);
glEnable(GL_SMOOTH);
glPolygonMode(GL_FRONT_AND_BACK,GL_FILL);
glMaterialfv(GL_FRONT, GL_DIFFUSE, diff);
glMaterialfv(GL_FRONT, GL_AMBIENT, amb);
glMaterialfv(GL_FRONT,GL_SPECULAR,spec);
glMaterialf( GL_FRONT, GL_SHININESS, shininess);
Algo::Render::GL1::renderTriQuadPoly<PFP>(myMap,Algo::Render::GL1::SMOOTH, 1.0f,position, normal);
glDisable(GL_POLYGON_OFFSET_FILL);
} }
template <typename XXX> void MyQT::cb_keyPress(int code)
class ThreadArea: public Algo::Parallel::CGoGNThread<typename XXX::MAP>
{ {
protected: Utils::Chrono ch;
const typename XXX::TVEC3& m_positions; ch.start();
float area;
public:
ThreadArea(typename XXX::MAP& map, const typename XXX::TVEC3& pos, unsigned int th) :
Algo::Parallel::CGoGNThread<typename XXX::MAP>(map,th),
m_positions(pos),
area(0.0f)
{}
void operator()() switch(code)
{ {
// 3 times just for fun !!! case 'a':
area += Algo::Geometry::totalArea<XXX>(this->m_map, m_positions, SelectorTrue(), this->m_threadId); threadSimple();
area += Algo::Geometry::totalArea<XXX>(this->m_map, m_positions, SelectorTrue(), this->m_threadId); break;
area += Algo::Geometry::totalArea<XXX>(this->m_map, m_positions, SelectorTrue(), this->m_threadId); case 'z':
threadStorage();
break;
case 'q':
threadAttrib();
break;
default:
break;
} }
std::cout << "time = "<< ch.elapsed() << std::endl;
updateGL();
}
float getTripleValue() { return area; }
};
//
// Simple attribute parallel functor and traversor
//
template <typename XXX> template <typename XXX>
class ThreadNormals: public Algo::Parallel::CGoGNThread<typename XXX::MAP> class UnshrinkFunctor : public FunctorAttribThreaded
{ {
protected: protected:
const typename XXX::TVEC3& m_positions; VertexAttribute<typename XXX::VEC3>& m_positions;
typename XXX::TVEC3& m_normals; VertexAttribute<typename XXX::VEC3>& m_positions2;
public: public:
ThreadNormals(typename XXX::MAP& map, const typename XXX::TVEC3& pos, typename XXX::TVEC3& norm, unsigned int th): UnshrinkFunctor( VertexAttribute<typename XXX::VEC3>& pos, VertexAttribute<typename XXX::VEC3>& pos2):
Algo::Parallel::CGoGNThread<typename XXX::MAP>(map,th), FunctorAttribThreaded(),m_positions(pos),m_positions2(pos2)
m_positions(pos),
m_normals(norm)
{} {}
void operator()() void parallelDo(unsigned int i, unsigned int threadID)
{ {
Algo::Geometry::computeNormalVertices<XXX>(this->m_map, m_positions, m_normals, SelectorTrue(), this->m_threadId); m_positions2[i] = 1.1f * m_positions[i];
} }
// no need to duplicate here functor can be shared (no data), call foreach with true parameter
}; };
//template<typename XXX> void MyQT::threadAttrib()
//class Thread0 {
//{ UnshrinkFunctor<PFP> funct(position,position2);
//protected: Algo::Parallel::foreach_attrib(myMap.getAttributeContainer<VERTEX>(), funct, true);
// typename XXX::MAP& m_map;
// MyGlutWin& m_mgw; myMap.swapAttributes(position,position2);
// unsigned int m_th; m_positionVBO->updateData(position);
// SelectorTrue m_selt; m_lines->setAttributePosition(m_positionVBO);
//public: updateGL();
// Thread0(typename XXX::MAP& map,MyGlutWin& mgw, unsigned int th): }
// m_map(map), m_mgw(mgw), m_th(th) {}
//
// void operator()()
// {
// CGoGNout << "Begin render init"<<CGoGNendl;
// m_mgw.useContext();
//
// // instanciation of the renderer (here using VBOs)
// m_mgw.m_render = new Algo::Render::VBO::MapRender_VBO();
//
// // update the renderer (geometry and primitives)
// m_mgw.m_render->updateData(Algo::Render::VBO::POSITIONS, position);
//
// m_mgw.m_render->initPrimitives<PFP>(m_map, m_selt, Algo::Render::VBO::TRIANGLES,m_th);
// m_mgw.m_render->initPrimitives<PFP>(m_map, m_selt, Algo::Render::VBO::LINES,m_th);
//
// m_mgw.releaseContext();
// CGoGNout<< "Render OK "<< CGoGNendl;
//
// }
//};
//
// Simple thread that traverse a map
//
template <typename XXX> template <typename XXX>
class calculFunctor1 : public Algo::Parallel::FunctorMapThreaded<typename XXX::MAP> class ShrinkFunctor : public FunctorMapThreaded<typename XXX::MAP >
{ {
protected: protected:
typename XXX::TVEC3& m_positions; VertexAttribute<typename XXX::VEC3>& m_positions;
typename XXX::TVEC3& m_normals; VertexAttribute<typename XXX::VEC3>& m_positions2;
public: public:
calculFunctor1( typename XXX::MAP& map, typename XXX::TVEC3& pos, typename XXX::TVEC3& norm, unsigned int id=0): ShrinkFunctor( typename XXX::MAP& map, VertexAttribute<typename XXX::VEC3>& pos, VertexAttribute<typename XXX::VEC3>& pos2):
Algo::Parallel::FunctorMapThreaded<typename XXX::MAP>(map,id), m_positions(pos), m_normals(norm) {} FunctorMapThreaded< typename XXX::MAP>(map),
m_positions(pos),m_positions2(pos2)
{}
bool operator()(Dart d) void parallelDo(Dart d, unsigned int threadID)
{ {
typename XXX::VEC3 n1 = Algo::Geometry::vertexNormal<XXX>(this->m_map, d, m_positions); typename XXX::VEC3 Q(0,0,0);
typename XXX::VEC3 n2 = Algo::Geometry::vertexNormal<XXX>(this->m_map, this->m_map.phi1(d), m_positions); int nb=0;
typename XXX::VEC3 n3 = Algo::Geometry::vertexNormal<XXX>(this->m_map, this->m_map.phi_1(d), m_positions); Traversor2VVaE<typename XXX::MAP> trav(this->m_map,d);
typename XXX::VEC3 n = n1+n2+n3; for (Dart e=trav.begin(); e!=trav.end(); e = trav.next())
n1 = Algo::Geometry::vertexNormal<XXX>(this->m_map, d, m_positions);
n2 = Algo::Geometry::vertexNormal<XXX>(this->m_map, this->m_map.phi1(d), m_positions);
n3 = Algo::Geometry::vertexNormal<XXX>(this->m_map, this->m_map.phi_1(d), m_positions);
n += n1+n2+n3;
n.normalize();
m_normals[d] = n;
// m_normals[d] = Algo::Geometry::vertexNormal<XXX>(this->m_map, d, m_positions);
return false;
}
Algo::Parallel::FunctorMapThreaded<typename XXX::MAP>* duplicate(unsigned int id)
{ {
calculFunctor1<XXX>* copy = new calculFunctor1<XXX>(this->m_map,m_positions,m_normals,id); Q += m_positions[e];
return reinterpret_cast<Algo::Parallel::FunctorMapThreaded<typename XXX::MAP>*>(copy); nb++;
}
m_positions2[d] = Q/nb;
} }
// no need to duplicate here functor can be shared (no data), call foreach with true parameter
}; };
void MyQT::threadSimple()
{
ShrinkFunctor<PFP> funct(myMap,position,position2);
Algo::Parallel::foreach_cell<PFP::MAP,VERTEX>(myMap, funct,true);
myMap.swapAttributes(position,position2);
m_positionVBO->updateData(position);
m_lines->setAttributePosition(m_positionVBO);
updateGL();
}
// Thread foreach with storage (computing average length of edges)
template <typename XXX> template <typename XXX>
class LengthEdgeFunctor : public Algo::Parallel::FunctorMapThreadedResult<typename XXX::MAP, std::pair<double,unsigned int> > class LengthEdgeFunctor : public FunctorMapThreaded<typename XXX::MAP >
{ {
protected: protected:
typename XXX::TVEC3& m_positions; VertexAttribute<typename XXX::VEC3>& m_positions;
double m_length; double m_length;
unsigned int m_nb; unsigned int m_nb;
public: public:
LengthEdgeFunctor( typename XXX::MAP& map, typename XXX::TVEC3& pos, unsigned int id=0): LengthEdgeFunctor( typename XXX::MAP& map, VertexAttribute<typename XXX::VEC3>& pos):
Algo::Parallel::FunctorMapThreadedResult< typename XXX::MAP, std::pair<double,unsigned int> >(map,id), FunctorMapThreaded< typename XXX::MAP>(map),
m_positions(pos), m_positions(pos), m_length(0.0), m_nb(0)
m_length(0.0), {}
m_nb(0) {}
bool operator()(Dart d) double getLength() { return m_length;}
unsigned int getNb() { return m_nb;}
void parallelDo(Dart d, unsigned int threadID)
{ {
Dart dd = this->m_map.phi2(d); Dart dd = this->m_map.phi2(d);
typename XXX::VEC3 V = m_positions[dd] - m_positions[d]; typename XXX::VEC3 V = m_positions[dd] - m_positions[d];
m_length += V.norm(); m_length += V.norm();
m_nb++; m_nb++;
return false;
} }
Algo::Parallel::FunctorMapThreaded<typename XXX::MAP>* duplicate(unsigned int id) // no need to duplicate here, we create 1 functor by thread (see bellow)
};
void MyQT::threadStorage()
{
// functor need storage so we need one per thread
std::vector<FunctorMapThreaded<PFP::MAP>*> functs;
unsigned int nbthreads = Algo::Parallel::optimalNbThreads();
for (unsigned int i=0; i<nbthreads; ++i)
{ {
LengthEdgeFunctor<XXX>* copy = new LengthEdgeFunctor<XXX>(this->m_map,m_positions,id); LengthEdgeFunctor<PFP>* lef = new LengthEdgeFunctor<PFP>(myMap,position);
return reinterpret_cast<Algo::Parallel::FunctorMapThreaded<typename XXX::MAP>*>(copy); functs.push_back(lef);
} }
std::pair<double,unsigned int> getResult() { return std::pair<double,unsigned int>(m_length,m_nb);} // LengthEdgeFunctor<PFP>* lef0 = new LengthEdgeFunctor<PFP>(myMap,position);
// LengthEdgeFunctor<PFP>* lef1 = new LengthEdgeFunctor<PFP>(myMap,position);
// LengthEdgeFunctor<PFP>* lef2 = new LengthEdgeFunctor<PFP>(myMap,position);
// LengthEdgeFunctor<PFP>* lef3 = new LengthEdgeFunctor<PFP>(myMap,position);
// functs.push_back(lef0);
// functs.push_back(lef1);
// functs.push_back(lef2);
// functs.push_back(lef3);
}; Algo::Parallel::foreach_cell<PFP::MAP,EDGE>(myMap, functs, 4);
void MyQT::menu_slot1()
{
// cree un handler pour les normales aux sommets
VertexAttribute<PFP::VEC3> normal2 = myMap.addAttribute<PFP::VEC3, VERTEX>("normal2");
// ajout de 4 threads pour les markers //compute average length from each thread result and delete functors
myMap.addThreadMarker(4); double average = 0;
unsigned int all = 0;
for (unsigned int i=0; i<nbthreads; ++i)
{
LengthEdgeFunctor<PFP>* lef = dynamic_cast<LengthEdgeFunctor<PFP>*>(functs[i]);
average += lef->getLength();
all += lef->getNb();
delete lef;
}
average /= all;
//Algorithmes en // // double average = (lef0->getLength()+lef1->getLength()+lef2->getLength()+lef3->getLength()) / (lef0->getNb()+lef1->getNb()+lef2->getNb()+lef3->getNb());
std::cout << "AVERAGE LENGTH "<< average << std::endl;
boost::thread thread1( ThreadArea<PFP>(myMap,position,1)); // delete lef0;
boost::thread thread2( ThreadNormals<PFP>(myMap,position,normal,2)); // delete lef1;
thread1.join(); // delete lef2;
thread2.join(); // delete lef3;
// parallelisation de boucle sans resultat
calculFunctor1<PFP> tf1(myMap,position,normal);
Algo::Parallel::foreach_orbit<PFP>(myMap, VERTEX, tf1,4);
CGoGNout << "ok:"<< CGoGNendl;
// parallelisation de boucle avec resultats stockes
// vector pour le resultat (ici paire double/int pour faire la moyenne des longueurs des aretes)
std::vector<std::pair<double,unsigned int> > lengthp;
LengthEdgeFunctor <PFP> tflef(myMap,position); // le foncteur
// on lance l'algo parallelise (4 threads, buffer de 16384 brins par thread)
Algo::Parallel::foreach_orbit_res< PFP,std::pair<double,unsigned int> >(myMap, EDGE, tflef, 4 , 16384,lengthp);
// on calcule la somme des resultats
std::pair<double,unsigned int> le = Algo::Parallel::sumPairResult<double,unsigned int>(lengthp);
CGoGNout << "length :" <<le.first/le.second<< CGoGNendl;
// on enleve les markers ajoutes
myMap.removeThreadMarker(4);
} }
int main(int argc, char **argv) int main(int argc, char **argv)
{ {
// interface: // interface:
QApplication app(argc, argv); QApplication app(argc, argv);
MyQT sqt; MyQT sqt;
// ajout entree dans le menu application
sqt.add_menu_entry("test threads", SLOT(menu_slot1()));
// message d'aide
sqt.setHelpMsg("Tuto:\n"
"multi-threading \n"
"utilisation GL 1.1\n"
"file sector");
sqt.show();
sqt.statusMsg("Neww to create a sphere or Load for a mesh file"); sqt.statusMsg("Neww to create a sphere or Load for a mesh file");
CGoGNStream::allToConsole(&sqt); CGoGNStream::allToConsole(&sqt);
sqt.cb_Open(); if (!position.isValid())
position = myMap.addAttribute<PFP::VEC3, VERTEX>("position");
if (!position2.isValid())
position2 = myMap.addAttribute<PFP::VEC3, VERTEX>("position2");
if (!normal.isValid())
normal = myMap.addAttribute<PFP::VEC3, VERTEX>("normal");
// create a sphere
Algo::Modelisation::Polyhedron<PFP> prim(myMap, position);
// prim.cylinder_topo(2640,2640, true, true); // test for speed timing
prim.cylinder_topo(64,64, true, true); // topo of sphere is a closed cylinder
prim.embedSphere(20.0f);
// bounding box
Geom::BoundingBox<PFP::VEC3> bb = Algo::Geometry::computeBoundingBox<PFP>(myMap, position);
float lWidthObj = std::max<PFP::REAL>(std::max<PFP::REAL>(bb.size(0), bb.size(1)), bb.size(2));
Geom::Vec3f lPosObj = (bb.min() + bb.max()) / PFP::REAL(2);
std::cout << "lPosObj=" << lPosObj << std::endl;
std::cout << "lWidthObj=" << lWidthObj << std::endl;
sqt.setParamObject(lWidthObj,lPosObj.data());