virtual bool fragment(Vec3f bar, TGAColor &color) {
//B3_PROFILE("fragment");
Vec4f p = m_viewportMat*(varying_tri_light_view*bar);
float depth = p[2];
p = p/p[3];
float index_x = b3Max(float(0.0), b3Min(float(m_width-1), p[0]));
float index_y = b3Max(float(0.0), b3Min(float(m_height-1), p[1]));
int idx = int(index_x) + int(index_y)*m_width; // index in the shadowbuffer array
float shadow = 0.8+0.2*(m_shadowBuffer->at(idx)<-depth+0.05); // magic coeff to avoid z-fighting
Vec3f bn = (varying_nrm*bar).normalize();
Vec2f uv = varying_uv*bar;
Vec3f reflection_direction = (bn * (bn * m_light_dir_local * 2.f) - m_light_dir_local).normalize();
float specular = pow(b3Max(reflection_direction.z, 0.f), m_model->specular(uv));
float diffuse = b3Max(0.f, bn * m_light_dir_local);
color = m_model->diffuse(uv);
color[0] *= m_colorRGBA[0];
color[1] *= m_colorRGBA[1];
color[2] *= m_colorRGBA[2];
color[3] *= m_colorRGBA[3];
for (int i = 0; i < 3; ++i)
{
color[i] = b3Min(int(m_ambient_coefficient*color[i] + shadow*(m_diffuse_coefficient*diffuse+m_specular_coefficient*specular)*color[i]*m_light_color[i]), 255);
}
return false;
}
virtual bool fragment(Vec3f bar, TGAColor &color) {
Vec3f bn = (varying_nrm*bar).normalize();
Vec2f uv = varying_uv*bar;
Vec3f reflection_direction = (bn * (bn * m_light_dir_local * 2.f) - m_light_dir_local).normalize();
float specular = pow(b3Max(reflection_direction.z, 0.f), m_model->specular(uv));
float diffuse = b3Max(0.f, bn * m_light_dir_local);
float ambient_coefficient = 0.6;
float diffuse_coefficient = 0.35;
float specular_coefficient = 0.05;
float intensity = ambient_coefficient + b3Min(diffuse * diffuse_coefficient + specular * specular_coefficient, 1.0f - ambient_coefficient);
color = m_model->diffuse(uv) * intensity;
//warning: bgra color is swapped to rgba to upload texture
color.bgra[0] *= m_colorRGBA[0];
color.bgra[1] *= m_colorRGBA[1];
color.bgra[2] *= m_colorRGBA[2];
color.bgra[3] *= m_colorRGBA[3];
color.bgra[0] *= m_light_color[0];
color.bgra[1] *= m_light_color[1];
color.bgra[2] *= m_light_color[2];
return false;
}
bool b3BroadPhase::Report(u32 proxyId)
{
if (proxyId == m_queryProxyId)
{
// The proxy can't overlap with itself.
return true;
}
// Check capacity.
if (m_pairCount == m_pairCapacity)
{
// Duplicate capacity.
m_pairCapacity *= 2;
b3Pair* oldPairs = m_pairs;
m_pairs = (b3Pair*)b3Alloc(m_pairCapacity * sizeof(b3Pair));
memcpy(m_pairs, oldPairs, m_pairCount * sizeof(b3Pair));
b3Free(oldPairs);
}
// Add overlapping pair to the pair buffer.
m_pairs[m_pairCount].proxy1 = b3Min(proxyId, m_queryProxyId);
m_pairs[m_pairCount].proxy2 = b3Max(proxyId, m_queryProxyId);
++m_pairCount;
// Keep looking for overlapping pairs.
return true;
}
bool b3BroadPhase::QueryCallback(i32 proxyId) {
if (proxyId == m_queryProxyId) {
// The proxy can't overlap with itself.
return true;
}
// Check capacity.
if (m_pairBufferCount == m_pairBufferCapacity) {
// Duplicate capacity.
m_pairBufferCapacity *= 2;
b3Pair* oldPairBuffer = m_pairBuffer;
m_pairBuffer = (b3Pair*)::b3Alloc(m_pairBufferCapacity * sizeof(b3Pair));
::memcpy(m_pairBuffer, oldPairBuffer, m_pairBufferCount * sizeof(b3Pair));
::b3Free(oldPairBuffer);
}
// Add overlapping pair to the pair buffer.
m_pairBuffer[m_pairBufferCount].proxy1 = b3Min(proxyId, m_queryProxyId);
m_pairBuffer[m_pairBufferCount].proxy2 = b3Max(proxyId, m_queryProxyId);
++m_pairBufferCount;
// Keep looking for overlapping pairs.
return true;
}
virtual bool fragment(Vec3f bar, TGAColor &color) {
Vec3f bn = (varying_nrm*bar).normalize();
Vec2f uv = varying_uv*bar;
mat<3,3,float> A;
A[0] = ndc_tri.col(1) - ndc_tri.col(0);
A[1] = ndc_tri.col(2) - ndc_tri.col(0);
A[2] = bn;
mat<3,3,float> AI = A.invert();
Vec3f i = AI * Vec3f(varying_uv[0][1] - varying_uv[0][0], varying_uv[0][2] - varying_uv[0][0], 0);
Vec3f j = AI * Vec3f(varying_uv[1][1] - varying_uv[1][0], varying_uv[1][2] - varying_uv[1][0], 0);
mat<3,3,float> B;
B.set_col(0, i.normalize());
B.set_col(1, j.normalize());
B.set_col(2, bn);
Vec3f n = (B*m_model->normal(uv)).normalize();
float diff = b3Min(b3Max(0.f, n*m_light_dir_local+0.3f),1.f);
//float diff = b3Max(0.f, n*m_light_dir_local);
color = m_model->diffuse(uv)*diff;
return false;
}
static inline b3Scalar calcArea4Points(const b3Vector3 &p0,const b3Vector3 &p1,const b3Vector3 &p2,const b3Vector3 &p3)
{
// It calculates possible 3 area constructed from random 4 points and returns the biggest one.
b3Vector3 a[3],b[3];
a[0] = p0 - p1;
a[1] = p0 - p2;
a[2] = p0 - p3;
b[0] = p2 - p3;
b[1] = p1 - p3;
b[2] = p1 - p2;
//todo: Following 3 cross production can be easily optimized by SIMD.
b3Vector3 tmp0 = a[0].cross(b[0]);
b3Vector3 tmp1 = a[1].cross(b[1]);
b3Vector3 tmp2 = a[2].cross(b[2]);
return b3Max(b3Max(tmp0.length2(),tmp1.length2()),tmp2.length2());
}
static inline void b3SolveContact(b3ContactConstraint4& cs,
const b3Vector3& posA, b3Vector3& linVelA, b3Vector3& angVelA, float invMassA, const b3Matrix3x3& invInertiaA,
const b3Vector3& posB, b3Vector3& linVelB, b3Vector3& angVelB, float invMassB, const b3Matrix3x3& invInertiaB,
float maxRambdaDt[4], float minRambdaDt[4])
{
b3Vector3 dLinVelA; dLinVelA.setZero();
b3Vector3 dAngVelA; dAngVelA.setZero();
b3Vector3 dLinVelB; dLinVelB.setZero();
b3Vector3 dAngVelB; dAngVelB.setZero();
for(int ic=0; ic<4; ic++)
{
// dont necessary because this makes change to 0
if( cs.m_jacCoeffInv[ic] == 0.f ) continue;
{
b3Vector3 angular0, angular1, linear;
b3Vector3 r0 = cs.m_worldPos[ic] - (b3Vector3&)posA;
b3Vector3 r1 = cs.m_worldPos[ic] - (b3Vector3&)posB;
b3SetLinearAndAngular( (const b3Vector3 &)-cs.m_linear, (const b3Vector3 &)r0, (const b3Vector3 &)r1, linear, angular0, angular1 );
float rambdaDt = b3CalcRelVel((const b3Vector3 &)cs.m_linear,(const b3Vector3 &) -cs.m_linear, angular0, angular1,
linVelA, angVelA, linVelB, angVelB ) + cs.m_b[ic];
rambdaDt *= cs.m_jacCoeffInv[ic];
{
float prevSum = cs.m_appliedRambdaDt[ic];
float updated = prevSum;
updated += rambdaDt;
updated = b3Max( updated, minRambdaDt[ic] );
updated = b3Min( updated, maxRambdaDt[ic] );
rambdaDt = updated - prevSum;
cs.m_appliedRambdaDt[ic] = updated;
}
b3Vector3 linImp0 = invMassA*linear*rambdaDt;
b3Vector3 linImp1 = invMassB*(-linear)*rambdaDt;
b3Vector3 angImp0 = (invInertiaA* angular0)*rambdaDt;
b3Vector3 angImp1 = (invInertiaB* angular1)*rambdaDt;
#ifdef _WIN32
b3Assert(_finite(linImp0.getX()));
b3Assert(_finite(linImp1.getX()));
#endif
{
linVelA += linImp0;
angVelA += angImp0;
linVelB += linImp1;
angVelB += angImp1;
}
}
}
}
void solveContact3(b3GpuConstraint4* cs,
b3Vector3* posAPtr, b3Vector3* linVelA, b3Vector3* angVelA, float invMassA, const b3Matrix3x3& invInertiaA,
b3Vector3* posBPtr, b3Vector3* linVelB, b3Vector3* angVelB, float invMassB, const b3Matrix3x3& invInertiaB,
b3Vector3* dLinVelA, b3Vector3* dAngVelA, b3Vector3* dLinVelB, b3Vector3* dAngVelB)
{
float minRambdaDt = 0;
float maxRambdaDt = FLT_MAX;
for(int ic=0; ic<4; ic++)
{
if( cs->m_jacCoeffInv[ic] == 0.f ) continue;
b3Vector3 angular0, angular1, linear;
b3Vector3 r0 = cs->m_worldPos[ic] - *posAPtr;
b3Vector3 r1 = cs->m_worldPos[ic] - *posBPtr;
setLinearAndAngular( cs->m_linear, r0, r1, linear, angular0, angular1 );
float rambdaDt = calcRelVel( cs->m_linear, -cs->m_linear, angular0, angular1,
*linVelA+*dLinVelA, *angVelA+*dAngVelA, *linVelB+*dLinVelB, *angVelB+*dAngVelB ) + cs->m_b[ic];
rambdaDt *= cs->m_jacCoeffInv[ic];
{
float prevSum = cs->m_appliedRambdaDt[ic];
float updated = prevSum;
updated += rambdaDt;
updated = b3Max( updated, minRambdaDt );
updated = b3Min( updated, maxRambdaDt );
rambdaDt = updated - prevSum;
cs->m_appliedRambdaDt[ic] = updated;
}
b3Vector3 linImp0 = invMassA*linear*rambdaDt;
b3Vector3 linImp1 = invMassB*(-linear)*rambdaDt;
b3Vector3 angImp0 = (invInertiaA* angular0)*rambdaDt;
b3Vector3 angImp1 = (invInertiaB* angular1)*rambdaDt;
if (invMassA)
{
*dLinVelA += linImp0;
*dAngVelA += angImp0;
}
if (invMassB)
{
*dLinVelB += linImp1;
*dAngVelB += angImp1;
}
}
}
static
__inline
void solveFriction(b3GpuConstraint4& cs,
const b3Vector3& posA, b3Vector3& linVelA, b3Vector3& angVelA, float invMassA, const b3Matrix3x3& invInertiaA,
const b3Vector3& posB, b3Vector3& linVelB, b3Vector3& angVelB, float invMassB, const b3Matrix3x3& invInertiaB,
float maxRambdaDt[4], float minRambdaDt[4])
{
if( cs.m_fJacCoeffInv[0] == 0 && cs.m_fJacCoeffInv[0] == 0 ) return;
const b3Vector3& center = (const b3Vector3&)cs.m_center;
b3Vector3 n = -(const b3Vector3&)cs.m_linear;
b3Vector3 tangent[2];
#if 1
b3PlaneSpace1 (n, tangent[0],tangent[1]);
#else
b3Vector3 r = cs.m_worldPos[0]-center;
tangent[0] = cross3( n, r );
tangent[1] = cross3( tangent[0], n );
tangent[0] = normalize3( tangent[0] );
tangent[1] = normalize3( tangent[1] );
#endif
b3Vector3 angular0, angular1, linear;
b3Vector3 r0 = center - posA;
b3Vector3 r1 = center - posB;
for(int i=0; i<2; i++)
{
setLinearAndAngular( tangent[i], r0, r1, linear, angular0, angular1 );
float rambdaDt = calcRelVel(linear, -linear, angular0, angular1,
linVelA, angVelA, linVelB, angVelB );
rambdaDt *= cs.m_fJacCoeffInv[i];
{
float prevSum = cs.m_fAppliedRambdaDt[i];
float updated = prevSum;
updated += rambdaDt;
updated = b3Max( updated, minRambdaDt[i] );
updated = b3Min( updated, maxRambdaDt[i] );
rambdaDt = updated - prevSum;
cs.m_fAppliedRambdaDt[i] = updated;
}
b3Vector3 linImp0 = invMassA*linear*rambdaDt;
b3Vector3 linImp1 = invMassB*(-linear)*rambdaDt;
b3Vector3 angImp0 = (invInertiaA* angular0)*rambdaDt;
b3Vector3 angImp1 = (invInertiaB* angular1)*rambdaDt;
#ifdef _WIN32
b3Assert(_finite(linImp0.getX()));
b3Assert(_finite(linImp1.getX()));
#endif
linVelA += linImp0;
angVelA += angImp0;
linVelB += linImp1;
angVelB += angImp1;
}
{ // angular damping for point constraint
b3Vector3 ab = ( posB - posA ).normalized();
b3Vector3 ac = ( center - posA ).normalized();
if( b3Dot( ab, ac ) > 0.95f || (invMassA == 0.f || invMassB == 0.f))
{
float angNA = b3Dot( n, angVelA );
float angNB = b3Dot( n, angVelB );
angVelA -= (angNA*0.1f)*n;
angVelB -= (angNB*0.1f)*n;
}
}
}
int main(int argc, char* argv[])
{
int sz = sizeof(b3Generic6DofConstraint);
int sz2 = sizeof(b3Point2PointConstraint);
int sz3 = sizeof(b3TypedConstraint);
int sz4 = sizeof(b3TranslationalLimitMotor);
int sz5 = sizeof(b3RotationalLimitMotor);
int sz6 = sizeof(b3Transform);
//b3OpenCLUtils::setCachePath("/Users/erwincoumans/develop/mycache");
b3SetCustomEnterProfileZoneFunc(b3ProfileManager::Start_Profile);
b3SetCustomLeaveProfileZoneFunc(b3ProfileManager::Stop_Profile);
b3SetCustomPrintfFunc(myprintf);
b3Vector3 test=b3MakeVector3(1,2,3);
test.x = 1;
test.y = 4;
printf("main start");
b3CommandLineArgs args(argc,argv);
ParticleDemo::ConstructionInfo ci;
if (args.CheckCmdLineFlag("help"))
{
Usage();
return 0;
}
selectedDemo = loadCurrentDemoEntry(sStartFileName);
args.GetCmdLineArgument("selected_demo",selectedDemo);
if (args.CheckCmdLineFlag("new_batching"))
{
useNewBatchingKernel = true;
}
bool benchmark=args.CheckCmdLineFlag("benchmark");
args.GetCmdLineArgument("max_framecount",maxFrameCount);
args.GetCmdLineArgument("shadowmap_size",shadowMapWorldSize);
args.GetCmdLineArgument("shadowmap_resolution",shadowMapWidth);
shadowMapHeight=shadowMapWidth;
if (args.CheckCmdLineFlag("disable_shadowmap"))
{
useShadowMap = false;
}
args.GetCmdLineArgument("pair_benchmark_file",gPairBenchFileName);
gDebugLauncherCL = args.CheckCmdLineFlag("debug_kernel_launch");
dump_timings=args.CheckCmdLineFlag("dump_timings");
ci.useOpenCL = !args.CheckCmdLineFlag("disable_opencl");
ci.m_useConcaveMesh = true;//args.CheckCmdLineFlag("use_concave_mesh");
if (ci.m_useConcaveMesh)
{
enableExperimentalCpuConcaveCollision = true;
}
ci.m_useInstancedCollisionShapes = !args.CheckCmdLineFlag("no_instanced_collision_shapes");
args.GetCmdLineArgument("cl_device", ci.preferredOpenCLDeviceIndex);
args.GetCmdLineArgument("cl_platform", ci.preferredOpenCLPlatformIndex);
gAllowCpuOpenCL = args.CheckCmdLineFlag("allow_opencl_cpu");
gUseLargeBatches = args.CheckCmdLineFlag("use_large_batches");
gUseJacobi = args.CheckCmdLineFlag("use_jacobi");
gUseDbvt = args.CheckCmdLineFlag("use_dbvt");
gDumpContactStats = args.CheckCmdLineFlag("dump_contact_stats");
gCalcWorldSpaceAabbOnCpu = args.CheckCmdLineFlag("calc_aabb_cpu");
gUseCalculateOverlappingPairsHost = args.CheckCmdLineFlag("calc_pairs_cpu");
gIntegrateOnCpu = args.CheckCmdLineFlag("integrate_cpu");
gConvertConstraintOnCpu = args.CheckCmdLineFlag("convert_constraints_cpu");
useUniformGrid = args.CheckCmdLineFlag("use_uniform_grid");
args.GetCmdLineArgument("x_dim", ci.arraySizeX);
args.GetCmdLineArgument("y_dim", ci.arraySizeY);
args.GetCmdLineArgument("z_dim", ci.arraySizeZ);
args.GetCmdLineArgument("x_gap", ci.gapX);
args.GetCmdLineArgument("y_gap", ci.gapY);
args.GetCmdLineArgument("z_gap", ci.gapZ);
gPause = args.CheckCmdLineFlag("paused");
gDebugForceLoadingFromSource = args.CheckCmdLineFlag("load_cl_kernels_from_disk");
gDebugSkipLoadingBinary = args.CheckCmdLineFlag("disable_cached_cl_kernels");
#ifndef B3_NO_PROFILE
b3ProfileManager::Reset();
#endif //B3_NO_PROFILE
window = new b3gDefaultOpenGLWindow();
b3gWindowConstructionInfo wci(g_OpenGLWidth,g_OpenGLHeight);
window->createWindow(wci);
window->setResizeCallback(MyResizeCallback);
window->setMouseMoveCallback(MyMouseMoveCallback);
window->setMouseButtonCallback(MyMouseButtonCallback);
window->setKeyboardCallback(MyKeyboardCallback);
window->setWindowTitle("Bullet 3.x GPU Rigid Body http://bulletphysics.org");
printf("-----------------------------------------------------\n");
#ifndef __APPLE__
glewInit();
#endif
gui = new GwenUserInterface();
printf("started GwenUserInterface");
GLPrimitiveRenderer prim(g_OpenGLWidth,g_OpenGLHeight);
stash = initFont(&prim);
if (gui)
{
gui->init(g_OpenGLWidth,g_OpenGLHeight,stash,window->getRetinaScale());
printf("init fonts");
gui->setToggleButtonCallback(MyButtonCallback);
gui->registerToggleButton(MYPAUSE,"Pause");
gui->registerToggleButton(MYPROFILE,"Profile");
gui->registerToggleButton(MYRESET,"Reset");
int numItems = sizeof(allDemos)/sizeof(ParticleDemo::CreateFunc*);
demoNames.clear();
for (int i=0;i<numItems;i++)
{
GpuDemo* demo = allDemos[i]();
demoNames.push_back(demo->getName());
delete demo;
}
gui->registerComboBox(MYCOMBOBOX1,numItems,&demoNames[0],selectedDemo);
gui->setComboBoxCallback(MyComboBoxCallback);
}
do
{
bool syncOnly = false;
gReset = false;
{
GLint err;
glEnable(GL_BLEND);
err = glGetError();
b3Assert(err==GL_NO_ERROR);
glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
glDisable(GL_DEPTH_TEST);
err = glGetError();
b3Assert(err==GL_NO_ERROR);
window->startRendering();
glClearColor(1,1,1,1);
glClear(GL_COLOR_BUFFER_BIT| GL_DEPTH_BUFFER_BIT);//|GL_STENCIL_BUFFER_BIT);
glEnable(GL_DEPTH_TEST);
sth_begin_draw(stash);
//sth_draw_text(stash, droidRegular,12.f, dx, dy-50, "How does this OpenGL True Type font look? ", &dx,width,height);
int spacing = 0;//g_OpenGLHeight;
float sx,sy,dx,dy,lh;
sx = 0;
sy = g_OpenGLHeight;
dx = sx; dy = sy;
//if (1)
const char* msg[] = {"Please wait, initializing the OpenCL demo",
"Please make sure to run the demo on a high-end discrete GPU with OpenCL support",
"The first time it can take a bit longer to compile the OpenCL kernels.",
"Check the console if it takes longer than 1 minute or if a demos has issues.",
"Please share the full commandline output when reporting issues:",
"App_Bullet3_OpenCL_Demos_* >> error.log",
"",
"",
#ifdef _DEBUG
"Some of the demos load a large .obj file,",
"please use an optimized build of this app for faster parsing",
"",
"",
#endif
"You can press F1 to create a single screenshot,",
"or press F2 toggle screenshot (useful to create movies)",
"",
"",
"There are various command-line options such as --benchmark",
"See http://github.com/erwincoumans/bullet3 for more information"
};
int fontSize = 68;
int nummsg = sizeof(msg)/sizeof(const char*);
for (int i=0;i<nummsg;i++)
{
char txt[512];
sprintf(txt,"%s",msg[i]);
//sth_draw_text(stash, droidRegular,i, 10, dy-spacing, txt, &dx,g_OpenGLWidth,g_OpenGLHeight);
sth_draw_text(stash, droidRegular,fontSize, 10, spacing, txt, &dx,g_OpenGLWidth,g_OpenGLHeight);
spacing+=fontSize;
fontSize = 32;
}
sth_end_draw(stash);
sth_flush_draw(stash);
window->endRendering();
}
static bool once=true;
//glClearColor(0.3f, 0.3f, 0.3f, 1.0f);
glClearColor(1,1,1,1);
glClear(GL_COLOR_BUFFER_BIT|GL_DEPTH_BUFFER_BIT);
window->setWheelCallback(b3DefaultWheelCallback);
{
GpuDemo* demo = allDemos[selectedDemo]();
sDemo = demo;
// demo->myinit();
bool useGpu = false;
//int maxObjectCapacity=128*1024;
int maxObjectCapacity=1024*1024;
maxObjectCapacity = b3Max(maxObjectCapacity,ci.arraySizeX*ci.arraySizeX*ci.arraySizeX+10);
{
ci.m_instancingRenderer = new GLInstancingRenderer(maxObjectCapacity);//render.getInstancingRenderer();
ci.m_window = window;
ci.m_gui = gui;
ci.m_instancingRenderer->init();
ci.m_instancingRenderer->resize(g_OpenGLWidth,g_OpenGLHeight);
ci.m_instancingRenderer->InitShaders();
ci.m_primRenderer = &prim;
// render.init();
}
{
demo->initPhysics(ci);
}
printf("-----------------------------------------------------\n");
FILE* csvFile = 0;
FILE* detailsFile = 0;
if (benchmark)
{
char prefixFileName[1024];
char csvFileName[1024];
char detailsFileName[1024];
b3OpenCLDeviceInfo info;
b3OpenCLUtils::getDeviceInfo(demo->getInternalData()->m_clDevice,&info);
//todo: move this time stuff into the Platform/Window class
#ifdef _WIN32
SYSTEMTIME time;
GetLocalTime(&time);
char buf[1024];
DWORD dwCompNameLen = 1024;
if (0 != GetComputerName(buf, &dwCompNameLen))
{
printf("%s", buf);
} else
{
printf("unknown", buf);
}
sprintf(prefixFileName,"%s_%s_%s_%d_%d_%d_date_%d-%d-%d_time_%d-%d-%d",info.m_deviceName,buf,demoNames[selectedDemo],ci.arraySizeX,ci.arraySizeY,ci.arraySizeZ,time.wDay,time.wMonth,time.wYear,time.wHour,time.wMinute,time.wSecond);
#else
timeval now;
gettimeofday(&now,0);
struct tm* ptm;
ptm = localtime (&now.tv_sec);
char buf[1024];
#ifdef __APPLE__
sprintf(buf,"MacOSX");
#else
sprintf(buf,"Unix");
#endif
sprintf(prefixFileName,"%s_%s_%s_%d_%d_%d_date_%d-%d-%d_time_%d-%d-%d",info.m_deviceName,buf,demoNames[selectedDemo],ci.arraySizeX,ci.arraySizeY,ci.arraySizeZ,
ptm->tm_mday,
ptm->tm_mon+1,
ptm->tm_year+1900,
ptm->tm_hour,
ptm->tm_min,
ptm->tm_sec);
#endif
sprintf(csvFileName,"%s.csv",prefixFileName);
sprintf(detailsFileName,"%s.txt",prefixFileName);
printf("Open csv file %s and details file %s\n", csvFileName,detailsFileName);
//GetSystemTime(&time2);
csvFile=fopen(csvFileName,"w");
detailsFile = fopen(detailsFileName,"w");
if (detailsFile)
defaultOutput = detailsFile;
//if (f)
// fprintf(f,"%s (%dx%dx%d=%d),\n", g_deviceName,ci.arraySizeX,ci.arraySizeY,ci.arraySizeZ,ci.arraySizeX*ci.arraySizeY*ci.arraySizeZ);
}
fprintf(defaultOutput,"Demo settings:\n");
fprintf(defaultOutput," SelectedDemo=%d, demoname = %s\n", selectedDemo, demo->getName());
fprintf(defaultOutput," x_dim=%d, y_dim=%d, z_dim=%d\n",ci.arraySizeX,ci.arraySizeY,ci.arraySizeZ);
fprintf(defaultOutput," x_gap=%f, y_gap=%f, z_gap=%f\n",ci.gapX,ci.gapY,ci.gapZ);
fprintf(defaultOutput,"\nOpenCL settings:\n");
fprintf(defaultOutput," Preferred cl_device index %d\n", ci.preferredOpenCLDeviceIndex);
fprintf(defaultOutput," Preferred cl_platform index%d\n", ci.preferredOpenCLPlatformIndex);
fprintf(defaultOutput,"\n");
if (demo->getInternalData()->m_platformId)
{
b3OpenCLUtils::printPlatformInfo( demo->getInternalData()->m_platformId);
fprintf(defaultOutput,"\n");
b3OpenCLUtils::printDeviceInfo( demo->getInternalData()->m_clDevice);
fprintf(defaultOutput,"\n");
}
do
{
GLint err = glGetError();
assert(err==GL_NO_ERROR);
if (exportFrame || exportMovie)
{
if (!renderTexture)
{
renderTexture = new GLRenderToTexture();
GLuint renderTextureId;
glGenTextures(1, &renderTextureId);
// "Bind" the newly created texture : all future texture functions will modify this texture
glBindTexture(GL_TEXTURE_2D, renderTextureId);
// Give an empty image to OpenGL ( the last "0" )
//glTexImage2D(GL_TEXTURE_2D, 0,GL_RGB, g_OpenGLWidth,g_OpenGLHeight, 0,GL_RGBA, GL_UNSIGNED_BYTE, 0);
//glTexImage2D(GL_TEXTURE_2D, 0,GL_RGBA32F, g_OpenGLWidth,g_OpenGLHeight, 0,GL_RGBA, GL_FLOAT, 0);
glTexImage2D(GL_TEXTURE_2D, 0,GL_RGBA32F, g_OpenGLWidth,g_OpenGLHeight, 0,GL_RGBA, GL_FLOAT, 0);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
//glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
//glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
renderTexture->init(g_OpenGLWidth,g_OpenGLHeight,renderTextureId, RENDERTEXTURE_COLOR);
}
bool result = renderTexture->enable();
}
err = glGetError();
assert(err==GL_NO_ERROR);
b3ProfileManager::Reset();
b3ProfileManager::Increment_Frame_Counter();
// render.reshape(g_OpenGLWidth,g_OpenGLHeight);
ci.m_instancingRenderer->resize(g_OpenGLWidth,g_OpenGLHeight);
prim.setScreenSize(g_OpenGLWidth,g_OpenGLHeight);
err = glGetError();
assert(err==GL_NO_ERROR);
window->startRendering();
err = glGetError();
assert(err==GL_NO_ERROR);
glClear(GL_COLOR_BUFFER_BIT| GL_DEPTH_BUFFER_BIT);//|GL_STENCIL_BUFFER_BIT);
glEnable(GL_DEPTH_TEST);
err = glGetError();
assert(err==GL_NO_ERROR);
if (!gPause)
{
B3_PROFILE("clientMoveAndDisplay");
demo->clientMoveAndDisplay();
}
else
{
}
{
B3_PROFILE("renderScene");
demo->renderScene();
}
err = glGetError();
assert(err==GL_NO_ERROR);
/*if (demo->getDynamicsWorld() && demo->getDynamicsWorld()->getNumCollisionObjects())
{
B3_PROFILE("renderPhysicsWorld");
b3AlignedObjectArray<b3CollisionObject*> arr = demo->getDynamicsWorld()->getCollisionObjectArray();
b3CollisionObject** colObjArray = &arr[0];
render.renderPhysicsWorld(demo->getDynamicsWorld()->getNumCollisionObjects(),colObjArray, syncOnly);
syncOnly = true;
}
*/
if (exportFrame || exportMovie)
{
char fileName[1024];
sprintf(fileName,"screenShot%d.png",frameIndex++);
writeTextureToPng(g_OpenGLWidth,g_OpenGLHeight,fileName);
exportFrame = false;
renderTexture->disable();
}
{
B3_PROFILE("gui->draw");
if (gui && gDrawGui)
gui->draw(g_OpenGLWidth,g_OpenGLHeight);
}
err = glGetError();
assert(err==GL_NO_ERROR);
{
B3_PROFILE("window->endRendering");
window->endRendering();
}
err = glGetError();
assert(err==GL_NO_ERROR);
{
B3_PROFILE("glFinish");
}
if (dump_timings)
{
b3ProfileManager::dumpAll(stdout);
}
if (csvFile)
{
static int frameCount=0;
if (frameCount>0)
{
DumpSimulationTime(csvFile);
if (detailsFile)
{
fprintf(detailsFile,"\n==================================\nFrame %d:\n", frameCount);
b3ProfileManager::dumpAll(detailsFile);
}
}
if (frameCount>=maxFrameCount)
window->setRequestExit();
frameCount++;
}
if (gStep)
gPause=true;
} while (!window->requestedExit() && !gReset);
demo->exitPhysics();
b3ProfileManager::CleanupMemory();
delete ci.m_instancingRenderer;
delete demo;
sDemo = 0;
if (detailsFile)
{
fclose(detailsFile);
detailsFile=0;
}
if (csvFile)
{
fclose(csvFile);
csvFile=0;
}
}
} while (gReset);
if (gui)
gui->setComboBoxCallback(0);
{
delete gui;
gui=0;
exitFont();
window->closeWindow();
delete window;
window = 0;
}
return 0;
}
void b3GpuPgsContactSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem inertiaBuf, int numContacts, cl_mem contactBuf, const b3Config& config, int static0Index)
{
B3_PROFILE("solveContacts");
m_data->m_bodyBufferGPU->setFromOpenCLBuffer(bodyBuf,numBodies);
m_data->m_inertiaBufferGPU->setFromOpenCLBuffer(inertiaBuf,numBodies);
m_data->m_pBufContactOutGPU->setFromOpenCLBuffer(contactBuf,numContacts);
if (optionalSortContactsDeterminism)
{
if (!gCpuSortContactsDeterminism)
{
B3_PROFILE("GPU Sort contact constraints (determinism)");
m_data->m_pBufContactOutGPUCopy->resize(numContacts);
m_data->m_contactKeyValues->resize(numContacts);
m_data->m_pBufContactOutGPU->copyToCL(m_data->m_pBufContactOutGPUCopy->getBufferCL(),numContacts,0,0);
{
b3LauncherCL launcher(m_data->m_queue, m_data->m_setDeterminismSortDataChildShapeBKernel,"m_setDeterminismSortDataChildShapeBKernel");
launcher.setBuffer(m_data->m_pBufContactOutGPUCopy->getBufferCL());
launcher.setBuffer(m_data->m_contactKeyValues->getBufferCL());
launcher.setConst(numContacts);
launcher.launch1D( numContacts, 64 );
}
m_data->m_solverGPU->m_sort32->execute(*m_data->m_contactKeyValues);
{
b3LauncherCL launcher(m_data->m_queue, m_data->m_setDeterminismSortDataChildShapeAKernel,"m_setDeterminismSortDataChildShapeAKernel");
launcher.setBuffer(m_data->m_pBufContactOutGPUCopy->getBufferCL());
launcher.setBuffer(m_data->m_contactKeyValues->getBufferCL());
launcher.setConst(numContacts);
launcher.launch1D( numContacts, 64 );
}
m_data->m_solverGPU->m_sort32->execute(*m_data->m_contactKeyValues);
{
b3LauncherCL launcher(m_data->m_queue, m_data->m_setDeterminismSortDataBodyBKernel,"m_setDeterminismSortDataBodyBKernel");
launcher.setBuffer(m_data->m_pBufContactOutGPUCopy->getBufferCL());
launcher.setBuffer(m_data->m_contactKeyValues->getBufferCL());
launcher.setConst(numContacts);
launcher.launch1D( numContacts, 64 );
}
m_data->m_solverGPU->m_sort32->execute(*m_data->m_contactKeyValues);
{
b3LauncherCL launcher(m_data->m_queue, m_data->m_setDeterminismSortDataBodyAKernel,"m_setDeterminismSortDataBodyAKernel");
launcher.setBuffer(m_data->m_pBufContactOutGPUCopy->getBufferCL());
launcher.setBuffer(m_data->m_contactKeyValues->getBufferCL());
launcher.setConst(numContacts);
launcher.launch1D( numContacts, 64 );
}
m_data->m_solverGPU->m_sort32->execute(*m_data->m_contactKeyValues);
{
B3_PROFILE("gpu reorderContactKernel (determinism)");
b3Int4 cdata;
cdata.x = numContacts;
//b3BufferInfoCL bInfo[] = { b3BufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ), b3BufferInfoCL( m_data->m_solverGPU->m_contactBuffer2->getBufferCL())
// , b3BufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
b3LauncherCL launcher(m_data->m_queue,m_data->m_solverGPU->m_reorderContactKernel,"m_reorderContactKernel");
launcher.setBuffer(m_data->m_pBufContactOutGPUCopy->getBufferCL());
launcher.setBuffer(m_data->m_pBufContactOutGPU->getBufferCL());
launcher.setBuffer(m_data->m_contactKeyValues->getBufferCL());
launcher.setConst( cdata );
launcher.launch1D( numContacts, 64 );
}
} else
{
B3_PROFILE("CPU Sort contact constraints (determinism)");
b3AlignedObjectArray<b3Contact4> cpuConstraints;
m_data->m_pBufContactOutGPU->copyToHost(cpuConstraints);
bool sort = true;
if (sort)
{
cpuConstraints.quickSort(b3ContactCmp);
for (int i=0;i<cpuConstraints.size();i++)
{
cpuConstraints[i].m_batchIdx = i;
}
}
m_data->m_pBufContactOutGPU->copyFromHost(cpuConstraints);
if (m_debugOutput==100)
{
for (int i=0;i<cpuConstraints.size();i++)
{
printf("c[%d].m_bodyA = %d, m_bodyB = %d, batchId = %d\n",i,cpuConstraints[i].m_bodyAPtrAndSignBit,cpuConstraints[i].m_bodyBPtrAndSignBit, cpuConstraints[i].m_batchIdx);
}
}
m_debugOutput++;
}
}
int nContactOut = m_data->m_pBufContactOutGPU->size();
bool useSolver = true;
if (useSolver)
{
float dt=1./60.;
b3ConstraintCfg csCfg( dt );
csCfg.m_enableParallelSolve = true;
csCfg.m_batchCellSize = 6;
csCfg.m_staticIdx = static0Index;
b3OpenCLArray<b3RigidBodyData>* bodyBuf = m_data->m_bodyBufferGPU;
void* additionalData = 0;//m_data->m_frictionCGPU;
const b3OpenCLArray<b3InertiaData>* shapeBuf = m_data->m_inertiaBufferGPU;
b3OpenCLArray<b3GpuConstraint4>* contactConstraintOut = m_data->m_contactCGPU;
int nContacts = nContactOut;
int maxNumBatches = 0;
if (!gUseLargeBatches)
{
if( m_data->m_solverGPU->m_contactBuffer2)
{
m_data->m_solverGPU->m_contactBuffer2->resize(nContacts);
}
if( m_data->m_solverGPU->m_contactBuffer2 == 0 )
{
m_data->m_solverGPU->m_contactBuffer2 = new b3OpenCLArray<b3Contact4>(m_data->m_context,m_data->m_queue, nContacts );
m_data->m_solverGPU->m_contactBuffer2->resize(nContacts);
}
//clFinish(m_data->m_queue);
{
B3_PROFILE("batching");
//@todo: just reserve it, without copy of original contact (unless we use warmstarting)
const b3OpenCLArray<b3RigidBodyData>* bodyNative = bodyBuf;
{
//b3OpenCLArray<b3RigidBodyData>* bodyNative = b3OpenCLArrayUtils::map<adl::TYPE_CL, true>( data->m_device, bodyBuf );
//b3OpenCLArray<b3Contact4>* contactNative = b3OpenCLArrayUtils::map<adl::TYPE_CL, true>( data->m_device, contactsIn );
const int sortAlignment = 512; // todo. get this out of sort
if( csCfg.m_enableParallelSolve )
{
int sortSize = B3NEXTMULTIPLEOF( nContacts, sortAlignment );
b3OpenCLArray<unsigned int>* countsNative = m_data->m_solverGPU->m_numConstraints;
b3OpenCLArray<unsigned int>* offsetsNative = m_data->m_solverGPU->m_offsets;
if (!gCpuSetSortData)
{ // 2. set cell idx
B3_PROFILE("GPU set cell idx");
struct CB
{
int m_nContacts;
int m_staticIdx;
float m_scale;
b3Int4 m_nSplit;
};
b3Assert( sortSize%64 == 0 );
CB cdata;
cdata.m_nContacts = nContacts;
cdata.m_staticIdx = csCfg.m_staticIdx;
cdata.m_scale = 1.f/csCfg.m_batchCellSize;
cdata.m_nSplit.x = B3_SOLVER_N_SPLIT_X;
cdata.m_nSplit.y = B3_SOLVER_N_SPLIT_Y;
cdata.m_nSplit.z = B3_SOLVER_N_SPLIT_Z;
m_data->m_solverGPU->m_sortDataBuffer->resize(nContacts);
b3BufferInfoCL bInfo[] = { b3BufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ), b3BufferInfoCL( bodyBuf->getBufferCL()), b3BufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
b3LauncherCL launcher(m_data->m_queue, m_data->m_solverGPU->m_setSortDataKernel,"m_setSortDataKernel" );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(b3BufferInfoCL) );
launcher.setConst( cdata.m_nContacts );
launcher.setConst( cdata.m_scale );
launcher.setConst(cdata.m_nSplit);
launcher.setConst(cdata.m_staticIdx);
launcher.launch1D( sortSize, 64 );
} else
{
m_data->m_solverGPU->m_sortDataBuffer->resize(nContacts);
b3AlignedObjectArray<b3SortData> sortDataCPU;
m_data->m_solverGPU->m_sortDataBuffer->copyToHost(sortDataCPU);
b3AlignedObjectArray<b3Contact4> contactCPU;
m_data->m_pBufContactOutGPU->copyToHost(contactCPU);
b3AlignedObjectArray<b3RigidBodyData> bodiesCPU;
bodyBuf->copyToHost(bodiesCPU);
float scale = 1.f/csCfg.m_batchCellSize;
b3Int4 nSplit;
nSplit.x = B3_SOLVER_N_SPLIT_X;
nSplit.y = B3_SOLVER_N_SPLIT_Y;
nSplit.z = B3_SOLVER_N_SPLIT_Z;
SetSortDataCPU(&contactCPU[0], &bodiesCPU[0], &sortDataCPU[0], nContacts,scale,nSplit,csCfg.m_staticIdx);
m_data->m_solverGPU->m_sortDataBuffer->copyFromHost(sortDataCPU);
}
if (!gCpuRadixSort)
{ // 3. sort by cell idx
B3_PROFILE("gpuRadixSort");
//int n = B3_SOLVER_N_SPLIT*B3_SOLVER_N_SPLIT;
//int sortBit = 32;
//if( n <= 0xffff ) sortBit = 16;
//if( n <= 0xff ) sortBit = 8;
//adl::RadixSort<adl::TYPE_CL>::execute( data->m_sort, *data->m_sortDataBuffer, sortSize );
//adl::RadixSort32<adl::TYPE_CL>::execute( data->m_sort32, *data->m_sortDataBuffer, sortSize );
b3OpenCLArray<b3SortData>& keyValuesInOut = *(m_data->m_solverGPU->m_sortDataBuffer);
this->m_data->m_solverGPU->m_sort32->execute(keyValuesInOut);
} else
{
b3OpenCLArray<b3SortData>& keyValuesInOut = *(m_data->m_solverGPU->m_sortDataBuffer);
b3AlignedObjectArray<b3SortData> hostValues;
keyValuesInOut.copyToHost(hostValues);
hostValues.quickSort(sortfnc);
keyValuesInOut.copyFromHost(hostValues);
}
if (gUseScanHost)
{
// 4. find entries
B3_PROFILE("cpuBoundSearch");
b3AlignedObjectArray<unsigned int> countsHost;
countsNative->copyToHost(countsHost);
b3AlignedObjectArray<b3SortData> sortDataHost;
m_data->m_solverGPU->m_sortDataBuffer->copyToHost(sortDataHost);
//m_data->m_solverGPU->m_search->executeHost(*m_data->m_solverGPU->m_sortDataBuffer,nContacts,*countsNative,B3_SOLVER_N_CELLS,b3BoundSearchCL::COUNT);
m_data->m_solverGPU->m_search->executeHost(sortDataHost,nContacts,countsHost,B3_SOLVER_N_CELLS,b3BoundSearchCL::COUNT);
countsNative->copyFromHost(countsHost);
//adl::BoundSearch<adl::TYPE_CL>::execute( data->m_search, *data->m_sortDataBuffer, nContacts, *countsNative,
// B3_SOLVER_N_SPLIT*B3_SOLVER_N_SPLIT, adl::BoundSearchBase::COUNT );
//unsigned int sum;
//m_data->m_solverGPU->m_scan->execute(*countsNative,*offsetsNative, B3_SOLVER_N_CELLS);//,&sum );
b3AlignedObjectArray<unsigned int> offsetsHost;
offsetsHost.resize(offsetsNative->size());
m_data->m_solverGPU->m_scan->executeHost(countsHost,offsetsHost, B3_SOLVER_N_CELLS);//,&sum );
offsetsNative->copyFromHost(offsetsHost);
//printf("sum = %d\n",sum);
} else
{
// 4. find entries
B3_PROFILE("gpuBoundSearch");
m_data->m_solverGPU->m_search->execute(*m_data->m_solverGPU->m_sortDataBuffer,nContacts,*countsNative,B3_SOLVER_N_CELLS,b3BoundSearchCL::COUNT);
m_data->m_solverGPU->m_scan->execute(*countsNative,*offsetsNative, B3_SOLVER_N_CELLS);//,&sum );
}
if (nContacts)
{ // 5. sort constraints by cellIdx
if (gReorderContactsOnCpu)
{
B3_PROFILE("cpu m_reorderContactKernel");
b3AlignedObjectArray<b3SortData> sortDataHost;
m_data->m_solverGPU->m_sortDataBuffer->copyToHost(sortDataHost);
b3AlignedObjectArray<b3Contact4> inContacts;
b3AlignedObjectArray<b3Contact4> outContacts;
m_data->m_pBufContactOutGPU->copyToHost(inContacts);
outContacts.resize(inContacts.size());
for (int i=0;i<nContacts;i++)
{
int srcIdx = sortDataHost[i].y;
outContacts[i] = inContacts[srcIdx];
}
m_data->m_solverGPU->m_contactBuffer2->copyFromHost(outContacts);
/* "void ReorderContactKernel(__global struct b3Contact4Data* in, __global struct b3Contact4Data* out, __global int2* sortData, int4 cb )\n"
"{\n"
" int nContacts = cb.x;\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" if( gIdx < nContacts )\n"
" {\n"
" int srcIdx = sortData[gIdx].y;\n"
" out[gIdx] = in[srcIdx];\n"
" }\n"
"}\n"
*/
} else
{
B3_PROFILE("gpu m_reorderContactKernel");
b3Int4 cdata;
cdata.x = nContacts;
b3BufferInfoCL bInfo[] = {
b3BufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ),
b3BufferInfoCL( m_data->m_solverGPU->m_contactBuffer2->getBufferCL())
, b3BufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
b3LauncherCL launcher(m_data->m_queue,m_data->m_solverGPU->m_reorderContactKernel,"m_reorderContactKernel");
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(b3BufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( nContacts, 64 );
}
}
}
}
//clFinish(m_data->m_queue);
// {
// b3AlignedObjectArray<unsigned int> histogram;
// m_data->m_solverGPU->m_numConstraints->copyToHost(histogram);
// printf(",,,\n");
// }
if (nContacts)
{
if (gUseCpuCopyConstraints)
{
for (int i=0;i<nContacts;i++)
{
m_data->m_pBufContactOutGPU->copyFromOpenCLArray(*m_data->m_solverGPU->m_contactBuffer2);
// m_data->m_solverGPU->m_contactBuffer2->getBufferCL();
// m_data->m_pBufContactOutGPU->getBufferCL()
}
} else
{
B3_PROFILE("gpu m_copyConstraintKernel");
b3Int4 cdata; cdata.x = nContacts;
b3BufferInfoCL bInfo[] = {
b3BufferInfoCL( m_data->m_solverGPU->m_contactBuffer2->getBufferCL() ),
b3BufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() )
};
b3LauncherCL launcher(m_data->m_queue, m_data->m_solverGPU->m_copyConstraintKernel,"m_copyConstraintKernel" );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(b3BufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( nContacts, 64 );
//we use the clFinish for proper benchmark/profile
clFinish(m_data->m_queue);
}
}
bool compareGPU = false;
if (nContacts)
{
if (!gCpuBatchContacts)
{
B3_PROFILE("gpu batchContacts");
maxNumBatches = 150;//250;
m_data->m_solverGPU->batchContacts( m_data->m_pBufContactOutGPU, nContacts, m_data->m_solverGPU->m_numConstraints, m_data->m_solverGPU->m_offsets, csCfg.m_staticIdx );
clFinish(m_data->m_queue);
} else
{
B3_PROFILE("cpu batchContacts");
static b3AlignedObjectArray<b3Contact4> cpuContacts;
b3OpenCLArray<b3Contact4>* contactsIn = m_data->m_solverGPU->m_contactBuffer2;
{
B3_PROFILE("copyToHost");
contactsIn->copyToHost(cpuContacts);
}
b3OpenCLArray<unsigned int>* countsNative = m_data->m_solverGPU->m_numConstraints;
b3OpenCLArray<unsigned int>* offsetsNative = m_data->m_solverGPU->m_offsets;
b3AlignedObjectArray<unsigned int> nNativeHost;
b3AlignedObjectArray<unsigned int> offsetsNativeHost;
{
B3_PROFILE("countsNative/offsetsNative copyToHost");
countsNative->copyToHost(nNativeHost);
offsetsNative->copyToHost(offsetsNativeHost);
}
int numNonzeroGrid=0;
if (gUseLargeBatches)
{
m_data->m_batchSizes.resize(B3_MAX_NUM_BATCHES);
int totalNumConstraints = cpuContacts.size();
int simdWidth =numBodies+1;//-1;//64;//-1;//32;
int numBatches = sortConstraintByBatch3( &cpuContacts[0], totalNumConstraints, totalNumConstraints+1,csCfg.m_staticIdx ,numBodies,&m_data->m_batchSizes[0]); // on GPU
maxNumBatches = b3Max(numBatches,maxNumBatches);
static int globalMaxBatch = 0;
if (maxNumBatches>globalMaxBatch )
{
globalMaxBatch = maxNumBatches;
b3Printf("maxNumBatches = %d\n",maxNumBatches);
}
} else
{
m_data->m_batchSizes.resize(B3_SOLVER_N_CELLS*B3_MAX_NUM_BATCHES);
B3_PROFILE("cpu batch grid");
for(int i=0; i<B3_SOLVER_N_CELLS; i++)
{
int n = (nNativeHost)[i];
int offset = (offsetsNativeHost)[i];
if( n )
{
numNonzeroGrid++;
int simdWidth =numBodies+1;//-1;//64;//-1;//32;
int numBatches = sortConstraintByBatch3( &cpuContacts[0]+offset, n, simdWidth,csCfg.m_staticIdx ,numBodies,&m_data->m_batchSizes[i*B3_MAX_NUM_BATCHES]); // on GPU
maxNumBatches = b3Max(numBatches,maxNumBatches);
static int globalMaxBatch = 0;
if (maxNumBatches>globalMaxBatch )
{
globalMaxBatch = maxNumBatches;
b3Printf("maxNumBatches = %d\n",maxNumBatches);
}
//we use the clFinish for proper benchmark/profile
}
}
//clFinish(m_data->m_queue);
}
{
B3_PROFILE("m_contactBuffer->copyFromHost");
m_data->m_solverGPU->m_contactBuffer2->copyFromHost((b3AlignedObjectArray<b3Contact4>&)cpuContacts);
}
}
}
}
}
//printf("maxNumBatches = %d\n", maxNumBatches);
if (gUseLargeBatches)
{
if (nContacts)
{
B3_PROFILE("cpu batchContacts");
static b3AlignedObjectArray<b3Contact4> cpuContacts;
// b3OpenCLArray<b3Contact4>* contactsIn = m_data->m_solverGPU->m_contactBuffer2;
{
B3_PROFILE("copyToHost");
m_data->m_pBufContactOutGPU->copyToHost(cpuContacts);
}
b3OpenCLArray<unsigned int>* countsNative = m_data->m_solverGPU->m_numConstraints;
b3OpenCLArray<unsigned int>* offsetsNative = m_data->m_solverGPU->m_offsets;
int numNonzeroGrid=0;
{
m_data->m_batchSizes.resize(B3_MAX_NUM_BATCHES);
int totalNumConstraints = cpuContacts.size();
int simdWidth =numBodies+1;//-1;//64;//-1;//32;
int numBatches = sortConstraintByBatch3( &cpuContacts[0], totalNumConstraints, totalNumConstraints+1,csCfg.m_staticIdx ,numBodies,&m_data->m_batchSizes[0]); // on GPU
maxNumBatches = b3Max(numBatches,maxNumBatches);
static int globalMaxBatch = 0;
if (maxNumBatches>globalMaxBatch )
{
globalMaxBatch = maxNumBatches;
b3Printf("maxNumBatches = %d\n",maxNumBatches);
}
}
{
B3_PROFILE("m_contactBuffer->copyFromHost");
m_data->m_solverGPU->m_contactBuffer2->copyFromHost((b3AlignedObjectArray<b3Contact4>&)cpuContacts);
}
}
}
if (nContacts)
{
B3_PROFILE("gpu convertToConstraints");
m_data->m_solverGPU->convertToConstraints( bodyBuf,
shapeBuf, m_data->m_solverGPU->m_contactBuffer2,
contactConstraintOut,
additionalData, nContacts,
(b3SolverBase::ConstraintCfg&) csCfg );
clFinish(m_data->m_queue);
}
if (1)
{
int numIter = 4;
m_data->m_solverGPU->m_nIterations = numIter;//10
if (!gCpuSolveConstraint)
{
B3_PROFILE("GPU solveContactConstraint");
/*m_data->m_solverGPU->solveContactConstraint(
m_data->m_bodyBufferGPU,
m_data->m_inertiaBufferGPU,
m_data->m_contactCGPU,0,
nContactOut ,
maxNumBatches);
*/
//m_data->m_batchSizesGpu->copyFromHost(m_data->m_batchSizes);
if (gUseLargeBatches)
{
solveContactConstraintBatchSizes(m_data->m_bodyBufferGPU,
m_data->m_inertiaBufferGPU,
m_data->m_contactCGPU,0,
nContactOut ,
maxNumBatches,numIter,&m_data->m_batchSizes);
} else
{
solveContactConstraint(
m_data->m_bodyBufferGPU,
m_data->m_inertiaBufferGPU,
m_data->m_contactCGPU,0,
nContactOut ,
maxNumBatches,numIter,&m_data->m_batchSizes);//m_data->m_batchSizesGpu);
}
}
else
{
B3_PROFILE("Host solveContactConstraint");
m_data->m_solverGPU->solveContactConstraintHost(m_data->m_bodyBufferGPU, m_data->m_inertiaBufferGPU, m_data->m_contactCGPU,0, nContactOut ,maxNumBatches,&m_data->m_batchSizes);
}
}
#if 0
if (0)
{
B3_PROFILE("read body velocities back to CPU");
//read body updated linear/angular velocities back to CPU
m_data->m_bodyBufferGPU->read(
m_data->m_bodyBufferCPU->m_ptr,numOfConvexRBodies);
adl::DeviceUtils::waitForCompletion( m_data->m_deviceCL );
}
#endif
}
}
GLInstanceGraphicsShape* btgCreateGraphicsShapeFromWavefrontObj(std::vector<tinyobj::shape_t>& shapes, bool flatShading)
{
b3AlignedObjectArray<GLInstanceVertex>* vertices = new b3AlignedObjectArray<GLInstanceVertex>;
{
// int numVertices = obj->vertexCount;
// int numIndices = 0;
b3AlignedObjectArray<int>* indicesPtr = new b3AlignedObjectArray<int>;
for (int s=0;s<(int)shapes.size();s++)
{
tinyobj::shape_t& shape = shapes[s];
int faceCount = shape.mesh.indices.size();
for (int f=0;f<faceCount;f+=3)
{
//btVector3 normal(face.m_plane[0],face.m_plane[1],face.m_plane[2]);
if (1)
{
btVector3 normal(0,1,0);
int vtxBaseIndex = vertices->size();
GLInstanceVertex vtx0;
vtx0.xyzw[0] = shape.mesh.positions[shape.mesh.indices[f]*3+0];
vtx0.xyzw[1] = shape.mesh.positions[shape.mesh.indices[f]*3+1];
vtx0.xyzw[2] = shape.mesh.positions[shape.mesh.indices[f]*3+2];
vtx0.xyzw[3] = 0.f;
if (shape.mesh.texcoords.size())
{
vtx0.uv[0] = shape.mesh.texcoords[shape.mesh.indices[f]*2+0];
vtx0.uv[1] = shape.mesh.texcoords[shape.mesh.indices[f]*2+1];
} else
{
vtx0.uv[0] = 0.5;
vtx0.uv[1] = 0.5;
}
GLInstanceVertex vtx1;
vtx1.xyzw[0] = shape.mesh.positions[shape.mesh.indices[f+1]*3+0];
vtx1.xyzw[1] = shape.mesh.positions[shape.mesh.indices[f+1]*3+1];
vtx1.xyzw[2] = shape.mesh.positions[shape.mesh.indices[f+1]*3+2];
vtx1.xyzw[3]= 0.f;
if (shape.mesh.texcoords.size())
{
vtx1.uv[0] = shape.mesh.texcoords[shape.mesh.indices[f+1]*2+0];
vtx1.uv[1] = shape.mesh.texcoords[shape.mesh.indices[f+1]*2+1];
} else
{
vtx1.uv[0] = 0.5f;
vtx1.uv[1] = 0.5f;
}
GLInstanceVertex vtx2;
vtx2.xyzw[0] = shape.mesh.positions[shape.mesh.indices[f+2]*3+0];
vtx2.xyzw[1] = shape.mesh.positions[shape.mesh.indices[f+2]*3+1];
vtx2.xyzw[2] = shape.mesh.positions[shape.mesh.indices[f+2]*3+2];
vtx2.xyzw[3] = 0.f;
if (shape.mesh.texcoords.size())
{
vtx2.uv[0] = shape.mesh.texcoords[shape.mesh.indices[f+2]*2+0];
vtx2.uv[1] = shape.mesh.texcoords[shape.mesh.indices[f+2]*2+1];
} else
{
vtx2.uv[0] = 0.5;
vtx2.uv[1] = 0.5;
}
btVector3 v0(vtx0.xyzw[0],vtx0.xyzw[1],vtx0.xyzw[2]);
btVector3 v1(vtx1.xyzw[0],vtx1.xyzw[1],vtx1.xyzw[2]);
btVector3 v2(vtx2.xyzw[0],vtx2.xyzw[1],vtx2.xyzw[2]);
unsigned int maxIndex = 0;
maxIndex = b3Max(maxIndex,shape.mesh.indices[f]*3+0);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f]*3+1);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f]*3+2);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+1]*3+0);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+1]*3+1);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+1]*3+2);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+2]*3+0);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+2]*3+1);
maxIndex = b3Max(maxIndex,shape.mesh.indices[f+2]*3+2);
bool hasNormals = (shape.mesh.normals.size() && maxIndex<shape.mesh.normals.size() );
if (flatShading || !hasNormals)
{
normal = (v1-v0).cross(v2-v0);
btScalar len2 = normal.length2();
//skip degenerate triangles
if (len2 > SIMD_EPSILON)
{
normal.normalize();
} else
{
normal.setValue(0,0,0);
}
vtx0.normal[0] = normal[0];
vtx0.normal[1] = normal[1];
vtx0.normal[2] = normal[2];
vtx1.normal[0] = normal[0];
vtx1.normal[1] = normal[1];
vtx1.normal[2] = normal[2];
vtx2.normal[0] = normal[0];
vtx2.normal[1] = normal[1];
vtx2.normal[2] = normal[2];
} else
{
vtx0.normal[0] = shape.mesh.normals[shape.mesh.indices[f]*3+0];
vtx0.normal[1] = shape.mesh.normals[shape.mesh.indices[f]*3+1];
vtx0.normal[2] = shape.mesh.normals[shape.mesh.indices[f]*3+2]; //shape.mesh.indices[f+1]*3+0
vtx1.normal[0] = shape.mesh.normals[shape.mesh.indices[f+1]*3+0];
vtx1.normal[1] = shape.mesh.normals[shape.mesh.indices[f+1]*3+1];
vtx1.normal[2] = shape.mesh.normals[shape.mesh.indices[f+1]*3+2];
vtx2.normal[0] = shape.mesh.normals[shape.mesh.indices[f+2]*3+0];
vtx2.normal[1] = shape.mesh.normals[shape.mesh.indices[f+2]*3+1];
vtx2.normal[2] = shape.mesh.normals[shape.mesh.indices[f+2]*3+2];
}
vertices->push_back(vtx0);
vertices->push_back(vtx1);
vertices->push_back(vtx2);
indicesPtr->push_back(vtxBaseIndex);
indicesPtr->push_back(vtxBaseIndex+1);
indicesPtr->push_back(vtxBaseIndex+2);
}
}
}
GLInstanceGraphicsShape* gfxShape = new GLInstanceGraphicsShape;
gfxShape->m_vertices = vertices;
gfxShape->m_numvertices = vertices->size();
gfxShape->m_indices = indicesPtr;
gfxShape->m_numIndices = indicesPtr->size();
for (int i=0;i<4;i++)
gfxShape->m_scaling[i] = 1;//bake the scaling into the vertices
return gfxShape;
}
}
void b3CollideHulls(b3Manifold& manifold,
const b3Transform& xf1, const b3HullShape* s1,
const b3Transform& xf2, const b3HullShape* s2)
{
B3_ASSERT(manifold.pointCount == 0);
const b3Hull* hull1 = s1->m_hull;
float32 r1 = s1->m_radius;
const b3Hull* hull2 = s2->m_hull;
float32 r2 = s2->m_radius;
float32 totalRadius = r1 + r2;
b3FaceQuery faceQuery1 = b3QueryFaceSeparation(xf1, hull1, xf2, hull2);
if (faceQuery1.separation > totalRadius)
{
return;
}
b3FaceQuery faceQuery2 = b3QueryFaceSeparation(xf2, hull2, xf1, hull1);
if (faceQuery2.separation > totalRadius)
{
return;
}
b3EdgeQuery edgeQuery = b3QueryEdgeSeparation(xf1, hull1, xf2, hull2);
if (edgeQuery.separation > totalRadius)
{
return;
}
const float32 kTol = 0.1f * B3_LINEAR_SLOP;
if (edgeQuery.separation > b3Max(faceQuery1.separation, faceQuery2.separation) + kTol)
{
b3BuildEdgeContact(manifold, xf1, edgeQuery.index1, s1, xf2, edgeQuery.index2, s2);
}
else
{
if (faceQuery1.separation + kTol > faceQuery2.separation)
{
b3BuildFaceContact(manifold, xf1, faceQuery1.index, s1, xf2, s2, false);
}
else
{
b3BuildFaceContact(manifold, xf2, faceQuery2.index, s2, xf1, s1, true);
}
}
// Heuristic succeded.
if (manifold.pointCount > 0)
{
return;
}
// Heuristic failed. Fallback.
if (edgeQuery.separation > b3Max(faceQuery1.separation, faceQuery2.separation))
{
b3BuildEdgeContact(manifold, xf1, edgeQuery.index1, s1, xf2, edgeQuery.index2, s2);
}
else
{
if (faceQuery1.separation > faceQuery2.separation)
{
b3BuildFaceContact(manifold, xf1, faceQuery1.index, s1, xf2, s2, false);
}
else
{
b3BuildFaceContact(manifold, xf2, faceQuery2.index, s2, xf1, s1, true);
}
}
// When both convex hulls are not simplified clipping might fail and create no contact points.
// For example, when a hull contains tiny faces, coplanar faces, and/or non-sharped edges.
// So we simply create a contact point between the segments.
// The hulls might overlap, but is better than solving no contact points.
if (manifold.pointCount == 0)
{
b3BuildEdgeContact(manifold, xf1, edgeQuery.index1, s1, xf2, edgeQuery.index2, s2);
}
// If the shapes are overlapping then at least on point must be created.
B3_ASSERT(manifold.pointCount > 0);
}
