void initGlWindow(int argc, char ** argv){
#ifdef USE_OPENGL
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
if( !glfwInit() )
{
cerr << "Failed to initalize GLFW" << endl;
exit( 1 );
}
if ( !glfwOpenWindow( width, height,
8, 8, 8, 8, 8, 8,
GLFW_WINDOW ) )
{
cerr << "Failed to open window" << endl;
exit( 1 );
}
glewInit();
if (! glewIsSupported("GL_VERSION_2_0 "))
{
cerr << "ERROR: Support for necessary OpenGL extensions missing." << endl;
exit( 1 );
}
//glEnable(GL_POINT_SMOOTH);
glViewport(0, 0, height, width);
glLoadIdentity();
glOrtho(-1.0, 1.0, -1.0, 1.0, 0.0, 1.0);
#endif
};
void GLManager::initGlWindow(){
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
if( !glfwInit() )
{
std::cerr << "Failed to initalize GLFW" << std::endl;
exit( 1 );
}
glfwOpenWindowHint( GLFW_WINDOW_NO_RESIZE, GL_TRUE );
if ( !glfwOpenWindow( width, height,
8, 8, 8, 8, 8, 8,
GLFW_WINDOW ) )
{
std::cerr << "Failed to open window" << std::endl;
exit( 1 );
}
glewInit();
if (! glewIsSupported("GL_VERSION_2_0 "))
{
std::cerr << "ERROR: Support for necessary OpenGL extensions missing." << std::endl;
exit( 1 );
}
glViewport(0, 0, height, width);
glLoadIdentity();
glOrtho(0.0, 1.0, 0.0, 1.0, 0.0, 1.0);
};
////////////////////////////////////////////////////////////////////////////////
//! Run a simple test for CUDA
////////////////////////////////////////////////////////////////////////////////
CUTBoolean runTest(int argc, char** argv)
{
if (!cutCheckCmdLineFlag(argc, (const char **)argv, "noqatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
g_bQAReadback = true;
fpsLimit = frameCheckNumber;
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
if (CUTFalse == initGL(argc, argv)) {
return CUTFalse;
}
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") )
cutilGLDeviceInit(argc, argv);
else {
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
}
// Create the CUTIL timer
cutilCheckError( cutCreateTimer( &timer));
// register callbacks
glutDisplayFunc(display);
glutKeyboardFunc(keyboard);
glutMouseFunc(mouse);
glutMotionFunc(motion);
if (g_bQAReadback) {
g_CheckRender = new CheckBackBuffer(window_width, window_height, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
// create VBO
createVBO(&vbo);
// run the cuda part
runCuda(vbo);
// check result of Cuda step
checkResultCuda(argc, argv, vbo);
atexit(cleanup);
// start rendering mainloop
glutMainLoop();
cudaThreadExit();
return CUTTrue;
}
void runAutoTest(int argc, char **argv)
{
printf("[%s] (automated testing w/ readback)\n", sSDKsample);
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilDeviceInit(argc, argv);
} else {
cudaSetDevice( cutGetMaxGflopsDeviceId() );
}
loadDefaultImage( argv[0] );
if (argc > 1) {
char *filename;
if (cutGetCmdLineArgumentstr(argc, (const char **)argv, "file", &filename)) {
initializeData(filename);
}
} else {
loadDefaultImage( argv[0]);
}
g_CheckRender = new CheckBackBuffer(imWidth, imHeight, sizeof(Pixel), false);
g_CheckRender->setExecPath(argv[0]);
Pixel *d_result;
cutilSafeCall( cudaMalloc( (void **)&d_result, imWidth*imHeight*sizeof(Pixel)) );
while (g_SobelDisplayMode <= 2)
{
printf("AutoTest: %s <%s>\n", sSDKsample, filterMode[g_SobelDisplayMode]);
sobelFilter(d_result, imWidth, imHeight, g_SobelDisplayMode, imageScale );
cutilSafeCall( cudaThreadSynchronize() );
cudaMemcpy(g_CheckRender->imageData(), d_result, imWidth*imHeight*sizeof(Pixel), cudaMemcpyDeviceToHost);
g_CheckRender->savePGM(sOriginal[g_Index], false, NULL);
if (!g_CheckRender->PGMvsPGM(sOriginal[g_Index], sReference[g_Index], MAX_EPSILON_ERROR, 0.15f)) {
g_TotalErrors++;
}
g_Index++;
g_SobelDisplayMode = (SobelDisplayMode)g_Index;
}
cutilSafeCall( cudaFree( d_result ) );
delete g_CheckRender;
if (!g_TotalErrors)
printf("TEST PASSED!\n");
else
printf("TEST FAILED!\n");
}
/**
* Selects a GPU for the CUDA execution.
* @param id id of the selected GPU.
*/
void selectGPU(int id) {
int deviceId;
if (id == -1) {
deviceId = cutGetMaxGflopsDeviceId();
} else {
deviceId = id;
}
cutilSafeCall(cudaSetDevice( deviceId ));
cutilCheckMsg("cudaSetDevice failed");
}
void Application::_init()
{
// Pick the best CUDA device
const int deviceIdx = cutGetMaxGflopsDeviceId();
CudaSafeCall( cudaSetDevice( deviceIdx ) );
// CUDA configuration
CudaSafeCall( cudaDeviceSetCacheConfig( cudaFuncCachePreferShared ) );
return;
}
void initCuda(){
// Use device with highest Gflops/s
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
initPBO(&pbo);
// Clean up on program exit
atexit(cleanupCuda);
runCuda();
}
bool initCUDA( int argc, char **argv)
{
return true;
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device"))
{
cutilGLDeviceInit(argc, argv);
}
else
{
cudaGLSetGLDevice (cutGetMaxGflopsDeviceId());
}
return true;
}
void
runGraphicsTest(int argc, char** argv)
{
printf("MarchingCubes ");
if (g_bFBODisplay) printf("[w/ FBO] ");
if (g_bOpenGLQA) printf("[Readback Comparisons] ");
printf("\n");
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
printf("[%s]\n", argv[0]);
printf(" Does not explicitly support -device=n in OpenGL mode\n");
printf(" To use -device=n, the sample must be running w/o OpenGL\n\n");
printf(" > %s -device=n -qatest\n", argv[0]);
printf("exiting...\n");
exit(0);
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
if(CUTFalse == initGL(&argc, argv)) {
return;
}
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
// register callbacks
glutDisplayFunc(display);
glutKeyboardFunc(keyboard);
glutMouseFunc(mouse);
glutMotionFunc(motion);
glutIdleFunc(idle);
glutReshapeFunc(reshape);
initMenus();
// Initialize CUDA buffers for Marching Cubes
initMC(argc, argv);
cutilCheckError( cutCreateTimer( &timer));
if (g_bOpenGLQA) {
g_CheckRender = new CheckBackBuffer(window_width, window_height, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
// start rendering mainloop
glutMainLoop();
cudaThreadExit();
}
bool initCUDA(void)
{
#if __DEVICE_EMULATION__
return true;
#else
int count = 0;
int i = 0;
cudaGetDeviceCount(&count);
if(count == 0) {
fprintf(stderr, "Nu exista nici un device.\n");
return false;
}
printf("Exista %d device-uri.\n",count);
for(i = 0; i < count; i++) {
cudaDeviceProp prop;
if(cudaGetDeviceProperties(&prop, i) == cudaSuccess) {
if(prop.major >= 1) {
break;
}
}
if(!prop.canMapHostMemory) {
fprintf(stderr, "Device %d cannot map host memory!\n",0);
exit(EXIT_FAILURE);
}
}
if(i == count) {
fprintf(stderr, "Nu exista nici un device care suporta CUDA.\n");
return false;
}
cudaSetDevice(cutGetMaxGflopsDeviceId());
cudaSetDeviceFlags(cudaDeviceMapHost);
checkCUDAError("cudaSetDeviceFlags");
printf("CUDA initializat cu succes\n");
// Create the CUTIL timer
cutilCheckError( cutCreateTimer( &timer));
return true;
#endif
}
void initCuda(){
// Use device with highest Gflops/s
#if CUDA_VERSION >= 5000
cudaGLSetGLDevice( gpuGetMaxGflopsDeviceId() );
#else
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
#endif
initPBO(&pbo);
// Clean up on program exit
atexit(cleanupCuda);
runCuda();
}
void initCuda(){
// Use device with highest Gflops/s
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
initPBO(&pbo);
dptr=NULL;
cudaGLMapBufferObject((void**)&dptr, pbo);
clearPBOpos(dptr,width,height);
cudaGLUnmapBufferObject(pbo);
// Clean up on program exit
atexit(cleanupCuda);
SetScissorWindow(glm::vec4(300,300,500,500));
texture.mapptr = stbi_load("cow.jpeg",&texture.width, &texture.height,&texture.depth,0);
runCuda();
}
void initCuda(){
// Use device with highest Gflops/s
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
initPBO(&pbo);
paraMap = new float[(int)renderCam->resolution.x *(int)renderCam->resolution.y];
effectiveRayMap =new int[(int)renderCam->resolution.x *(int)renderCam->resolution.y];
initialRayMap = new ray[(int)renderCam->resolution.x * (int)renderCam->resolution.y];
generateRayMap(renderCam, targetFrame);
// Clean up on program exit
atexit(cleanupCuda);
runCuda();
}
RemoteCUDARunner::RemoteCUDARunner():GPURunner<unsigned long,int>(TYPE_CUDA),m_metahashsize(0)
{
m_in=0;
m_devin=0;
m_out=0;
m_devout=0;
m_metahash=0;
m_devmetahash=0;
cutilSafeCall(cudaGetDeviceCount(&m_devicecount));
if(m_devicecount>0)
{
if(m_deviceindex<0 || m_deviceindex>=m_devicecount)
{
m_deviceindex=cutGetMaxGflopsDeviceId();
std::cout << "Setting CUDA device to Max GFlops device at index " << m_deviceindex << std::endl;
}
else
{
std::cout << "Setting CUDA device to device at index " << m_deviceindex << std::endl;
}
cudaDeviceProp props;
cudaGetDeviceProperties(&props,m_deviceindex);
std::cout << "Device info for " << props.name << " :" << std::endl;
std::cout << "Compute Capability : " << props.major << "." << props.minor << std::endl;
std::cout << "Clock Rate (hz) : " << props.clockRate << std::endl;
if(props.major>999)
{
std::cout << "CUDA seems to be running in CPU emulation mode" << std::endl;
}
cutilSafeCall(cudaSetDevice(m_deviceindex));
}
else
{
m_deviceindex=-1;
std::cout << "No CUDA capable device detected" << std::endl;
}
}
CUDARunner::CUDARunner():GPURunner<unsigned long,int>(TYPE_CUDA)
{
m_in=0;
m_devin=0;
m_out=0;
m_devout=0;
cutilSafeCall(cudaGetDeviceCount(&m_devicecount));
if(m_devicecount>0)
{
if(m_deviceindex<0 || m_deviceindex>=m_devicecount)
{
m_deviceindex=cutGetMaxGflopsDeviceId();
printf("Setting CUDA device to Max GFlops device at index %u\n",m_deviceindex);
}
else
{
printf("Setting CUDA device to device at index %u\n",m_deviceindex);
}
cudaDeviceProp props;
cudaGetDeviceProperties(&props,m_deviceindex);
printf("Device info for %s :\nCompute Capability : %d.%d\nClock Rate (hz) : %d\n",props.name,props.major,props.minor,props.clockRate);
if(props.major>999)
{
printf("CUDA seems to be running in CPU emulation mode\n");
}
cutilSafeCall(cudaSetDevice(m_deviceindex));
}
else
{
m_deviceindex=-1;
printf("No CUDA capable device detected\n");
}
}
int DrawScene::InitGL( )
{
GLenum err = glewInit();
if (GLEW_OK != err)
{
/* Problem: glewInit failed, something is seriously wrong. */
::MessageBox(NULL,"glewInit failed, something is seriously wrong.",
"glew error occured.",MB_OK );
return false;
}
#if CUDA_ENABLE
// cuda初始化
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
#endif
for (int i=0;i<COUNT_MODEL;i++)
{
m_model[i].loadModelData(FILENAME_MS3D);
m_model[i].reloadTextures(); // Loads Model Textures
}
glEnable(GL_TEXTURE_2D); // Enable Texture Mapping ( NEW )
glShadeModel(GL_SMOOTH); // Enable Smooth Shading
glClearColor(0.0f, 0.0f, 0.0f, 0.5f); // Black Background
glClearDepth(1.0f); // Depth Buffer Setup
glEnable(GL_DEPTH_TEST); // Enables Depth Testing
glDepthFunc(GL_LEQUAL); // The Type Of Depth Testing To Do
glHint(GL_PERSPECTIVE_CORRECTION_HINT, GL_NICEST); // Really Nice Perspective Calculations
#if ENABLE_CONSOLE_WINDOW
AllocConsole();
freopen( "CONOUT$","w",stdout);
#endif
m_particleMngr.addParticleNode("", PARTICLE_TYPE_FOUNTAIN );
return TRUE;
}
//------------------------------------------------------------------------------
static void
#if GLFW_VERSION_MAJOR>=3
keyboard(GLFWwindow *, int key, int scancode, int event, int mods) {
#else
#define GLFW_KEY_ESCAPE GLFW_KEY_ESC
keyboard(int key, int event) {
#endif
if (event == GLFW_RELEASE) return;
if (g_hud.KeyDown(tolower(key))) return;
switch (key) {
case 'Q': g_running = 0; break;
case 'F': fitFrame(); break;
case '+':
case '=': g_tessLevel++; break;
case '-': g_tessLevel = std::max(g_tessLevelMin, g_tessLevel-1); break;
case '.': g_moveModels = std::max(g_moveModels*2, 1); break;
case ',': g_moveModels = std::max(g_moveModels/2, 0); break;
case 'I': g_modelCount = std::max(g_modelCount/2, 1); rebuild(); break;
case 'O': g_modelCount = std::min(g_modelCount*2, MAX_MODELS); rebuild(); break;
case GLFW_KEY_ESCAPE: g_hud.SetVisible(!g_hud.IsVisible()); break;
}
}
//------------------------------------------------------------------------------
static void
callbackDisplayStyle(int b)
{
if (g_displayStyle == kVaryingColor or b == kVaryingColor or
g_displayStyle == kFaceVaryingColor or b == kFaceVaryingColor) {
// need to rebuild for varying reconstruct
g_displayStyle = b;
rebuild();
return;
}
g_displayStyle = b;
}
static void
callbackKernel(int k)
{
g_kernel = k;
#ifdef OPENSUBDIV_HAS_OPENCL
if (g_kernel == kCL and g_clContext == NULL) {
if (initCL(&g_clContext, &g_clQueue) == false) {
printf("Error in initializing OpenCL\n");
exit(1);
}
}
#endif
#ifdef OPENSUBDIV_HAS_CUDA
if (g_kernel == kCUDA and g_cudaInitialized == false) {
g_cudaInitialized = true;
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
}
#endif
rebuild();
}
void WaterPlaneCUDA::configure(Vector upperLeft, Vector lowerRight, float dampFactor, float resolution)
{
cudaSetDevice(cutGetMaxGflopsDeviceId());
cudaGLSetGLDevice(cutGetMaxGflopsDeviceId());
timeSinceLast = timePassed = 0;
unsigned int free, total;
int gpuCount, i;
CUresult res;
CUdevice dev;
CUcontext ctx;
cuInit(0);
cuDeviceGetCount(&gpuCount);
printf("Detected %d GPU\n",gpuCount);
for (i=0; i<gpuCount; i++)
{
cuDeviceGet(&dev,i);
cuCtxCreate(&ctx, 0, dev);
res = cuMemGetInfo(&free, &total);
if(res != CUDA_SUCCESS)
printf("!!!! cuMemGetInfo failed! (status = %x)", res);
printf("^^^^ Device: %d\n",i);
printf("^^^^ Free : %lu bytes (%lu KB) (%lu MB)\n", free, inKB(free), inMB(free));
printf("^^^^ Total: %lu bytes (%lu KB) (%lu MB)\n", total, inKB(total), inMB(total));
printf("^^^^ %f%% free, %f%% used\n", 100.0*free/(double)total, 100.0*(total - free)/(double)total);
cuCtxDetach(ctx);
}
this->stepSize = 1.0f/resolution;
this->resolutionFactor = resolution;
//reale Z - Achse ist x - Achse der WaterPlaneCUDA
this->sizeX = (unsigned int) abs(upperLeft.z - lowerRight.z);
//reale X -Achse ist y- Achse der WaterPlaneCUDA
this->sizeY = (unsigned int) abs(upperLeft.x - lowerRight.x);
//Anzahl der Netzpunkte in X -Richtung
this->pointsX = (unsigned int)(sizeX * resolution);
//Anzahl der Netzpunkte in Y -Richtung
pointsY = (unsigned int)(sizeY * resolution);
uLeft = upperLeft;
lRight = lowerRight;
//Der "Meeresspiegel"
baseHeight = lRight.y;
//Das Höhenfeld der WaterPlaneCUDA
waveMap = NULL;
initBuffer();
gpu_newVertices = new float3[pointsX*pointsY];
gpu_oldVertices = new float3[pointsX*pointsY];
gpu_normals = new float3[pointsX*pointsY];
for (int i=0;i<pointsX*pointsY;i++) {
gpu_newVertices[i]=make_float3(0,0,0);
gpu_oldVertices[i]=make_float3(0,0,0);
gpu_normals[i]=make_float3(0,1.0,0);
}
cutilSafeCall(cudaMalloc((void**)&gpu_newVertices,pointsX*pointsY*sizeof(float3)));
cutilSafeCall(cudaMalloc((void**)&gpu_oldVertices,pointsX*pointsY*sizeof(float3)));
cutilSafeCall(cudaMalloc((void**)&gpu_normals,pointsX*pointsY*sizeof(float3)));
drawMesh();
}
////////////////////////////////////////////////////////////////////////////////
//! Run test
////////////////////////////////////////////////////////////////////////////////
void runGraphicsTest(int argc, char** argv)
{
printf("[%s] ", sSDKsample);
if (g_bOpenGLQA) printf("[OpenGL Readback Comparisons] ");
printf("\n");
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device") ) {
printf("[%s]\n", argv[0]);
printf(" Does not explicitly support -device=n in OpenGL mode\n");
printf(" To use -device=n, the sample must be running w/o OpenGL\n\n");
printf(" > %s -device=n -qatest\n", argv[0]);
printf("exiting...\n");
exit(0);
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
if(CUTFalse == initGL( &argc, argv )) {
cudaThreadExit();
return;
}
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
// create FFT plan
CUFFT_SAFE_CALL(cufftPlan2d(&fftPlan, meshW, meshH, CUFFT_C2R) );
// allocate memory
fftInputW = (meshW / 2)+1;
fftInputH = meshH;
fftInputSize = (fftInputW*fftInputH)*sizeof(float2);
cutilSafeCall(cudaMalloc((void **)&d_h0, fftInputSize) );
cutilSafeCall(cudaMalloc((void **)&d_ht, fftInputSize) );
h_h0 = (float2 *) malloc(fftInputSize);
generate_h0();
cutilSafeCall(cudaMemcpy(d_h0, h_h0, fftInputSize, cudaMemcpyHostToDevice) );
cutilSafeCall(cudaMalloc((void **)&d_slope, meshW*meshH*sizeof(float2)) );
cutCreateTimer(&timer);
cutStartTimer(timer);
prevTime = cutGetTimerValue(timer);
// create vertex buffers and register with CUDA
createVBO(&heightVertexBuffer, meshW*meshH*sizeof(float));
// DEPRECATED: cutilSafeCall(cudaGLRegisterBufferObject(heightVertexBuffer));
cutilSafeCall(cudaGraphicsGLRegisterBuffer(&cuda_heightVB_resource, heightVertexBuffer, cudaGraphicsMapFlagsWriteDiscard));
createVBO(&slopeVertexBuffer, meshW*meshH*sizeof(float2));
// DEPRECATED: cutilSafeCall(cudaGLRegisterBufferObject(slopeVertexBuffer));
cutilSafeCall(cudaGraphicsGLRegisterBuffer(&cuda_slopeVB_resource, slopeVertexBuffer, cudaGraphicsMapFlagsWriteDiscard));
// create vertex and index buffer for mesh
createMeshPositionVBO(&posVertexBuffer, meshW, meshH);
createMeshIndexBuffer(&indexBuffer, meshW, meshH);
// Creating the Auto-Validation Code
if (g_bOpenGLQA) {
g_CheckRender = new CheckBackBuffer(windowH, windowH, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
runCuda();
// register callbacks
glutDisplayFunc(display);
glutKeyboardFunc(keyboard);
glutMouseFunc(mouse);
glutMotionFunc(motion);
glutReshapeFunc(reshape);
glutIdleFunc(idle);
// start rendering mainloop
glutMainLoop();
cudaThreadExit();
}
//////////////////////////////////////////////////////////////////////////////
// Program main
//////////////////////////////////////////////////////////////////////////////
int
main( int argc, char** argv)
{
printf("Run \"nbody -benchmark [-n=<numBodies>]\" to measure perfomance.\n\n");
bool benchmark =
(cutCheckCmdLineFlag(argc, (const char**) argv, "benchmark") != 0);
bool compareToCPU =
((cutCheckCmdLineFlag(argc, (const char**) argv, "compare") != 0) ||
!(cutCheckCmdLineFlag(argc, (const char**) argv, "noqatest") != 0));
bool regression =
(cutCheckCmdLineFlag(argc, (const char**) argv, "regression") != 0);
int devID;
cudaDeviceProp props;
// nBody has a mode that allows it to be run without using GL interop
if (benchmark || compareToCPU || regression) {
/*
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilDeviceInit(argc, argv);
} else {
devID = cutGetMaxGflopsDeviceId();
cudaSetDevice( devID );
} */
}
else
{
// This mode shows the OpenGL results rendered
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
glutInit(&argc, argv);
glutInitDisplayMode(GLUT_RGB | GLUT_DEPTH | GLUT_DOUBLE);
glutInitWindowSize(720, 480);
glutCreateWindow("CUDA n-body system");
GLenum err = glewInit();
if (GLEW_OK != err)
{
printf("GLEW Error: %s\n", glewGetErrorString(err));
}
else
{
#if defined(WIN32)
wglSwapIntervalEXT(0);
#elif defined(LINUX)
glxSwapIntervalSGI(0);
#endif
}
initGL();
initParameters();
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilGLDeviceInit(argc, argv);
} else {
devID = cutGetMaxGflopsDeviceId();
cudaGLSetGLDevice( devID );
}
}
// get number of SMs on this GPU
cutilSafeCall(cudaGetDevice(&devID));
cutilSafeCall(cudaGetDeviceProperties(&props, devID));
numIterations = 0;
int p = 256;
int q = 1;
cutGetCmdLineArgumenti(argc, (const char**) argv, "i", &numIterations);
cutGetCmdLineArgumenti(argc, (const char**) argv, "p", &p);
cutGetCmdLineArgumenti(argc, (const char**) argv, "q", &q);
// default number of bodies is #SMs * 4 * CTA size
numBodies = compareToCPU ? 4096 : p*q*4*props.multiProcessorCount;
cutGetCmdLineArgumenti(argc, (const char**) argv, "n", &numBodies);
switch (numBodies)
{
case 1024:
activeParams.m_clusterScale = 1.52f;
activeParams.m_velocityScale = 2.f;
break;
case 2048:
activeParams.m_clusterScale = 1.56f;
activeParams.m_velocityScale = 2.64f;
break;
case 4096:
activeParams.m_clusterScale = 1.68f;
activeParams.m_velocityScale = 2.98f;
break;
case 8192:
activeParams.m_clusterScale = 1.98f;
activeParams.m_velocityScale = 2.9f;
break;
default:
case 16384:
activeParams.m_clusterScale = 1.54f;
activeParams.m_velocityScale = 8.f;
break;
case 32768:
activeParams.m_clusterScale = 1.44f;
activeParams.m_velocityScale = 11.f;
break;
}
if (q * p > 256)
{
p = 256 / q;
printf("Setting p=%d, q=%d to maintain %d threads per block\n", p, q, 256);
}
if (q == 1 && numBodies < p)
{
p = numBodies;
}
init(numBodies, p, q, !(benchmark || compareToCPU));
reset(nbody, numBodies, NBODY_CONFIG_SHELL, !(benchmark || compareToCPU));
if (benchmark)
{
if (numIterations <= 0)
numIterations = 100;
runBenchmark(numIterations);
}
else if (compareToCPU || regression)
{
compareResults(regression, numBodies);
}
else
{
glutDisplayFunc(display);
glutReshapeFunc(reshape);
glutMouseFunc(mouse);
glutMotionFunc(motion);
glutKeyboardFunc(key);
glutSpecialFunc(special);
glutIdleFunc(idle);
cutilSafeCall(cudaEventRecord(startEvent, 0));
glutMainLoop();
}
if (nbodyCPU)
delete nbodyCPU;
if (nbodyCUDA)
delete nbodyCUDA;
if (hPos)
delete [] hPos;
if (hVel)
delete [] hVel;
if (hColor)
delete [] hColor;
cutilSafeCall(cudaEventDestroy(startEvent));
cutilSafeCall(cudaEventDestroy(stopEvent));
cutilCheckError(cutDeleteTimer(demoTimer));
return 0;
}
int main(int argc, char **argv)
{
uchar *h_Data;
uint *h_HistogramCPU, *h_HistogramGPU;
uchar *d_Data;
uint *d_Histogram;
uint hTimer;
int PassFailFlag = 1;
uint byteCount = 64 * 1048576;
uint uiSizeMult = 1;
cudaDeviceProp deviceProp;
deviceProp.major = 0;
deviceProp.minor = 0;
int dev;
shrQAStart(argc, argv);
// set logfile name and start logs
shrSetLogFileName ("histogram.txt");
//Use command-line specified CUDA device, otherwise use device with highest Gflops/s
if( shrCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
dev = cutilDeviceInit(argc, argv);
if (dev < 0) {
printf("No CUDA Capable Devices found, exiting...\n");
shrQAFinishExit(argc, (const char **)argv, QA_WAIVED);
}
} else {
cudaSetDevice( dev = cutGetMaxGflopsDeviceId() );
cutilSafeCall( cudaChooseDevice(&dev, &deviceProp) );
}
cutilSafeCall( cudaGetDeviceProperties(&deviceProp, dev) );
printf("CUDA device [%s] has %d Multi-Processors, Compute %d.%d\n",
deviceProp.name, deviceProp.multiProcessorCount, deviceProp.major, deviceProp.minor);
int version = deviceProp.major * 0x10 + deviceProp.minor;
if(version < 0x11)
{
printf("There is no device supporting a minimum of CUDA compute capability 1.1 for this SDK sample\n");
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_WAIVED);
}
cutilCheckError(cutCreateTimer(&hTimer));
// Optional Command-line multiplier to increase size of array to histogram
if (shrGetCmdLineArgumentu(argc, (const char**)argv, "sizemult", &uiSizeMult))
{
uiSizeMult = CLAMP(uiSizeMult, 1, 10);
byteCount *= uiSizeMult;
}
shrLog("Initializing data...\n");
shrLog("...allocating CPU memory.\n");
h_Data = (uchar *)malloc(byteCount);
h_HistogramCPU = (uint *)malloc(HISTOGRAM256_BIN_COUNT * sizeof(uint));
h_HistogramGPU = (uint *)malloc(HISTOGRAM256_BIN_COUNT * sizeof(uint));
shrLog("...generating input data\n");
srand(2009);
for(uint i = 0; i < byteCount; i++)
h_Data[i] = rand() % 256;
shrLog("...allocating GPU memory and copying input data\n\n");
cutilSafeCall( cudaMalloc((void **)&d_Data, byteCount ) );
cutilSafeCall( cudaMalloc((void **)&d_Histogram, HISTOGRAM256_BIN_COUNT * sizeof(uint) ) );
cutilSafeCall( cudaMemcpy(d_Data, h_Data, byteCount, cudaMemcpyHostToDevice) );
{
shrLog("Starting up 64-bin histogram...\n\n");
initHistogram64();
shrLog("Running 64-bin GPU histogram for %u bytes (%u runs)...\n\n", byteCount, numRuns);
for(int iter = -1; iter < numRuns; iter++){
//iter == -1 -- warmup iteration
if(iter == 0){
cutilSafeCall( cutilDeviceSynchronize() );
cutilCheckError( cutResetTimer(hTimer) );
cutilCheckError( cutStartTimer(hTimer) );
}
histogram64(d_Histogram, d_Data, byteCount);
}
cutilSafeCall( cutilDeviceSynchronize() );
cutilCheckError( cutStopTimer(hTimer));
double dAvgSecs = 1.0e-3 * (double)cutGetTimerValue(hTimer) / (double)numRuns;
shrLog("histogram64() time (average) : %.5f sec, %.4f MB/sec\n\n", dAvgSecs, ((double)byteCount * 1.0e-6) / dAvgSecs);
shrLogEx(LOGBOTH | MASTER, 0, "histogram64, Throughput = %.4f MB/s, Time = %.5f s, Size = %u Bytes, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * (double)byteCount / dAvgSecs), dAvgSecs, byteCount, 1, HISTOGRAM64_THREADBLOCK_SIZE);
shrLog("\nValidating GPU results...\n");
shrLog(" ...reading back GPU results\n");
cutilSafeCall( cudaMemcpy(h_HistogramGPU, d_Histogram, HISTOGRAM64_BIN_COUNT * sizeof(uint), cudaMemcpyDeviceToHost) );
shrLog(" ...histogram64CPU()\n");
histogram64CPU(
h_HistogramCPU,
h_Data,
byteCount
);
shrLog(" ...comparing the results...\n");
for(uint i = 0; i < HISTOGRAM64_BIN_COUNT; i++)
if(h_HistogramGPU[i] != h_HistogramCPU[i]) PassFailFlag = 0;
shrLog(PassFailFlag ? " ...64-bin histograms match\n\n" : " ***64-bin histograms do not match!!!***\n\n" );
shrLog("Shutting down 64-bin histogram...\n\n\n");
closeHistogram64();
}
{
shrLog("Initializing 256-bin histogram...\n");
initHistogram256();
shrLog("Running 256-bin GPU histogram for %u bytes (%u runs)...\n\n", byteCount, numRuns);
for(int iter = -1; iter < numRuns; iter++){
//iter == -1 -- warmup iteration
if(iter == 0){
cutilSafeCall( cutilDeviceSynchronize() );
cutilCheckError( cutResetTimer(hTimer) );
cutilCheckError( cutStartTimer(hTimer) );
}
histogram256(d_Histogram, d_Data, byteCount);
}
cutilSafeCall( cutilDeviceSynchronize() );
cutilCheckError( cutStopTimer(hTimer));
double dAvgSecs = 1.0e-3 * (double)cutGetTimerValue(hTimer) / (double)numRuns;
shrLog("histogram256() time (average) : %.5f sec, %.4f MB/sec\n\n", dAvgSecs, ((double)byteCount * 1.0e-6) / dAvgSecs);
shrLogEx(LOGBOTH | MASTER, 0, "histogram256, Throughput = %.4f MB/s, Time = %.5f s, Size = %u Bytes, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * (double)byteCount / dAvgSecs), dAvgSecs, byteCount, 1, HISTOGRAM256_THREADBLOCK_SIZE);
shrLog("\nValidating GPU results...\n");
shrLog(" ...reading back GPU results\n");
cutilSafeCall( cudaMemcpy(h_HistogramGPU, d_Histogram, HISTOGRAM256_BIN_COUNT * sizeof(uint), cudaMemcpyDeviceToHost) );
shrLog(" ...histogram256CPU()\n");
histogram256CPU(
h_HistogramCPU,
h_Data,
byteCount
);
shrLog(" ...comparing the results\n");
for(uint i = 0; i < HISTOGRAM256_BIN_COUNT; i++)
if(h_HistogramGPU[i] != h_HistogramCPU[i]) PassFailFlag = 0;
shrLog(PassFailFlag ? " ...256-bin histograms match\n\n" : " ***256-bin histograms do not match!!!***\n\n" );
shrLog("Shutting down 256-bin histogram...\n\n\n");
closeHistogram256();
}
shrLog("Shutting down...\n");
cutilCheckError(cutDeleteTimer(hTimer));
cutilSafeCall( cudaFree(d_Histogram) );
cutilSafeCall( cudaFree(d_Data) );
free(h_HistogramGPU);
free(h_HistogramCPU);
free(h_Data);
cutilDeviceReset();
shrLog("%s - Test Summary\n", sSDKsample);
// pass or fail (for both 64 bit and 256 bit histograms)
shrQAFinishExit(argc, (const char **)argv, (PassFailFlag ? QA_PASSED : QA_FAILED));
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int
main(int argc, char** argv)
{
int retVal = 0;
retVal = xnInit( argc, argv );
printf("[ %s ]\n", sSDKsample);
if (argc > 1) {
cutGetCmdLineArgumenti( argc, (const char**) argv, "n", (int *) &numParticles);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "qatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt")
)
{
g_bQAReadback = true;
}
if (cutCheckCmdLineFlag(argc, (const char **)argv, "glverify"))
{
g_bQAGLVerify = true;
}
}
if (g_bQAReadback) {
// For Automated testing, we do not use OpenGL/CUDA interop
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
cutilDeviceInit (argc, argv);
} else {
cudaSetDevice (cutGetMaxGflopsDeviceId() );
}
g_CheckRender = new CheckBackBuffer(winWidth, winHeight, 4, false);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
// This code path is used for Automated Testing
initParticles(numParticles, false, false);
initParams();
if (emitterOn) {
runEmitter();
}
SimParams ¶ms = psystem->getParams();
params.cursorPos = make_float3(cursorPosLag.x, cursorPosLag.y, cursorPosLag.z);
psystem->step(timestep);
float4 *pos = NULL, *vel = NULL;
int g_TotalErrors = 0;
psystem->dumpBin(&pos, &vel);
g_CheckRender->dumpBin(pos, numParticles*sizeof(float4), "smokeParticles_pos.bin");
g_CheckRender->dumpBin(vel, numParticles*sizeof(float4), "smokeParticles_vel.bin");
if (!g_CheckRender->compareBin2BinFloat("smokeParticles_pos.bin", sRefBin[0], numParticles*sizeof(float4), MAX_EPSILON_ERROR, THRESHOLD))
g_TotalErrors++;
if (!g_CheckRender->compareBin2BinFloat("smokeParticles_vel.bin", sRefBin[1], numParticles*sizeof(float4), MAX_EPSILON_ERROR, THRESHOLD))
g_TotalErrors++;
delete psystem;
delete g_CheckRender;
printf("%s\n", (g_TotalErrors > 0) ? "FAILED" : "PASSED");
cudaThreadExit();
} else {
// Normal smokeParticles rendering path
// 1st initialize OpenGL context, so we can properly set the GL for CUDA.
// This is needed to achieve optimal performance with OpenGL/CUDA interop.
initGL( &argc, argv );
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
cutilGLDeviceInit (argc, argv);
} else {
cudaGLSetGLDevice (cutGetMaxGflopsDeviceId() );
}
if (g_bQAGLVerify) {
g_CheckRender = new CheckBackBuffer(winWidth, winHeight, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
// This is the normal code path for SmokeParticles
initParticles(numParticles, true, true);
initParams();
initMenus();
glutDisplayFunc(display);
glutReshapeFunc(reshape);
glutMouseFunc(mouse);
glutMotionFunc(motion);
glutKeyboardFunc(key);
glutKeyboardUpFunc(keyUp);
glutSpecialFunc(special);
glutIdleFunc(idle);
glutMainLoop();
}
cutilExit(argc, argv);
return retVal;
}
//////////////////////////////////////////////////////////////////////////////
// Program main
//////////////////////////////////////////////////////////////////////////////
int
main( int argc, char** argv)
{
bool bTestResults = true;
shrQAStart(argc, argv);
if( cutCheckCmdLineFlag(argc, (const char**)argv, "help") ) {
showHelp();
return 0;
}
shrLog("Run \"nbody -benchmark [-n=<numBodies>]\" to measure perfomance.\n");
shrLog("\t-fullscreen (run n-body simulation in fullscreen mode)\n");
shrLog("\t-fp64 (use double precision floating point values for simulation)\n");
shrLog("\t-numdevices=N (use first N CUDA devices for simulation)\n");
// shrLog("\t-hostmem (stores simulation data in host memory)\n");
// shrLog("\t-cpu (performs simulation on the host)\n");
shrLog("\n");
bFullscreen = (cutCheckCmdLineFlag(argc, (const char**) argv, "fullscreen") != 0);
if (bFullscreen)
bShowSliders = false;
benchmark = (cutCheckCmdLineFlag(argc, (const char**) argv, "benchmark") != 0);
compareToCPU = ((cutCheckCmdLineFlag(argc, (const char**) argv, "compare") != 0) ||
(cutCheckCmdLineFlag(argc, (const char**) argv, "qatest") != 0));
QATest = (cutCheckCmdLineFlag(argc, (const char**) argv, "qatest") != 0);
useHostMem = (cutCheckCmdLineFlag(argc, (const char**) argv, "hostmem") != 0);
fp64 = (cutCheckCmdLineFlag(argc, (const char**) argv, "fp64") != 0);
flopsPerInteraction = fp64 ? 30 : 20;
useCpu = (cutCheckCmdLineFlag(argc, (const char**) argv, "cpu") != 0);
cutGetCmdLineArgumenti(argc, (const char**) argv, "numdevices", &numDevsRequested);
// for multi-device we currently require using host memory -- the devices share
// data via the host
if (numDevsRequested > 1)
useHostMem = true;
int numDevsAvailable = 0;
bool customGPU = false;
cudaGetDeviceCount(&numDevsAvailable);
if (numDevsAvailable < numDevsRequested) {
shrLog("Error: only %d Devices available, %d requested. Exiting.\n", numDevsAvailable, numDevsRequested);
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
}
shrLog("> %s mode\n", bFullscreen ? "Fullscreen" : "Windowed");
shrLog("> Simulation data stored in %s memory\n", useHostMem ? "system" : "video" );
shrLog("> %s precision floating point simulation\n", fp64 ? "Double" : "Single");
shrLog("> %d Devices used for simulation\n", numDevsRequested);
int devID;
cudaDeviceProp props;
// Initialize GL and GLUT if necessary
if (!benchmark && !compareToCPU) {
initGL(&argc, argv);
initParameters();
}
if (useCpu) {
useHostMem = true;
compareToCPU = false;
bSupportDouble = true;
#ifdef OPENMP
shrLog("> Simulation with CPU using OpenMP\n");
#else
shrLog("> Simulation with CPU\n");
#endif
}
else
{
// Now choose the CUDA Device
// Either without GL interop:
if (benchmark || compareToCPU || useHostMem)
{
// Note if we are using host memory for the body system, we
// don't use CUDA-GL interop.
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") )
{
devID = cutilDeviceInit(argc, argv);
if (devID < 0) {
printf("exiting...\n");
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
}
customGPU = true;
}
else
{
devID = cutGetMaxGflopsDeviceId();
cudaSetDevice( devID );
}
}
else // or with GL interop:
{
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilGLDeviceInit(argc, argv);
customGPU = true;
} else {
devID = cutGetMaxGflopsDeviceId();
cudaGLSetGLDevice( devID );
}
}
cutilSafeCall(cudaGetDevice(&devID));
cutilSafeCall(cudaGetDeviceProperties(&props, devID));
bSupportDouble = true;
#if CUDART_VERSION < 4000
if (numDevsRequested > 1)
{
shrLog("MultiGPU n-body requires CUDA 4.0 or later\n");
cutilDeviceReset();
shrQAFinishExit(argc, (const char**)argv, QA_PASSED);
}
#endif
// Initialize devices
if(numDevsRequested > 1 && customGPU)
{
printf("You can't use --numdevices and --device at the same time.\n");
shrQAFinishExit(argc, (const char**)argv, QA_PASSED);
}
if(customGPU) {
cudaDeviceProp props;
cutilSafeCall(cudaGetDeviceProperties(&props, devID));
shrLog("> Compute %d.%d CUDA device: [%s]\n", props.major, props.minor, props.name);
}
else
{
for (int i = 0; i < numDevsRequested; i++)
{
cudaDeviceProp props;
cutilSafeCall(cudaGetDeviceProperties(&props, i));
shrLog("> Compute %d.%d CUDA device: [%s]\n", props.major, props.minor, props.name);
if (useHostMem)
{
#if CUDART_VERSION >= 2020
if(!props.canMapHostMemory)
{
fprintf(stderr, "Device %d cannot map host memory!\n", devID);
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
}
if (numDevsRequested > 1)
cutilSafeCall(cudaSetDevice(i));
cutilSafeCall(cudaSetDeviceFlags(cudaDeviceMapHost));
#else
fprintf(stderr, "This CUDART version does not support <cudaDeviceProp.canMapHostMemory> field\n");
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
#endif
}
}
// CC 1.2 and earlier do not support double precision
if (props.major*10 + props.minor <= 12)
bSupportDouble = false;
}
//if(numDevsRequested > 1)
// cutilSafeCall(cudaSetDevice(devID));
if (fp64 && !bSupportDouble) {
fprintf(stderr, "One or more of the requested devices does not support double precision floating-point\n");
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
}
}
numIterations = 0;
p = 0;
q = 1;
cutGetCmdLineArgumenti(argc, (const char**) argv, "i", &numIterations);
cutGetCmdLineArgumenti(argc, (const char**) argv, "p", &p);
cutGetCmdLineArgumenti(argc, (const char**) argv, "q", &q);
if (p == 0) // p not set on command line
{
p = 256;
if (q * p > 256)
{
p = 256 / q;
shrLog("Setting p=%d, q=%d to maintain %d threads per block\n", p, q, 256);
}
}
// default number of bodies is #SMs * 4 * CTA size
if (useCpu)
#ifdef OPENMP
numBodies = 8192;
#else
numBodies = 4096;
#endif
else if (numDevsRequested == 1)
//------------------------------------------------------------------------------
int main(int argc, char ** argv) {
glutInit(&argc, argv);
glutInitDisplayMode(GLUT_RGBA |GLUT_DOUBLE | GLUT_DEPTH);
glutInitWindowSize(1024, 1024);
glutCreateWindow("OpenSubdiv test");
std::string str;
if (argc > 1) {
std::ifstream ifs(argv[1]);
if (ifs) {
std::stringstream ss;
ss << ifs.rdbuf();
ifs.close();
str = ss.str();
g_defaultShapes.push_back(SimpleShape(str.c_str(), argv[1], kCatmark));
}
}
initializeShapes();
int smenu = glutCreateMenu(modelMenu);
for(int i = 0; i < (int)g_defaultShapes.size(); ++i){
glutAddMenuEntry( g_defaultShapes[i].name.c_str(), i);
}
int lmenu = glutCreateMenu(levelMenu);
for(int i = 1; i < 8; ++i){
char level[16];
sprintf(level, "Level %d\n", i);
glutAddMenuEntry(level, i);
}
// Register Osd compute kernels
OpenSubdiv::OsdCpuKernelDispatcher::Register();
OpenSubdiv::OsdGlslKernelDispatcher::Register();
#if OPENSUBDIV_HAS_OPENCL
OpenSubdiv::OsdClKernelDispatcher::Register();
#endif
#if OPENSUBDIV_HAS_CUDA
OpenSubdiv::OsdCudaKernelDispatcher::Register();
// Note: This function randomly crashes with linux 5.0-dev driver.
// cudaGetDeviceProperties overrun stack..?
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
#endif
int kmenu = glutCreateMenu(kernelMenu);
int nKernels = OpenSubdiv::OsdKernelDispatcher::kMAX;
for(int i = 0; i < nKernels; ++i)
if(OpenSubdiv::OsdKernelDispatcher::HasKernelType(
OpenSubdiv::OsdKernelDispatcher::KernelType(i)))
glutAddMenuEntry(getKernelName(i), i);
glutCreateMenu(menu);
glutAddSubMenu("Level", lmenu);
glutAddSubMenu("Model", smenu);
glutAddSubMenu("Kernel", kmenu);
glutAttachMenu(GLUT_RIGHT_BUTTON);
glutDisplayFunc(display);
glutReshapeFunc(reshape);
glutMouseFunc(mouse);
glutKeyboardFunc(keyboard);
glutMotionFunc(motion);
glewInit();
initGL();
const char *filename = NULL;
for (int i = 1; i < argc; ++i) {
if (!strcmp(argv[i], "-d"))
g_level = atoi(argv[++i]);
else if (!strcmp(argv[i], "-c"))
g_repeatCount = atoi(argv[++i]);
else
filename = argv[i];
}
modelMenu(0);
glutIdleFunc(idle);
glutMainLoop();
quit();
}
//------------------------------------------------------------------------------
int main(int argc, char ** argv) {
glutInit(&argc, argv);
glutInitDisplayMode(GLUT_RGBA |GLUT_DOUBLE | GLUT_DEPTH);
glutInitWindowSize(1024, 1024);
glutCreateWindow("OpenSubdiv ptexViewer");
int lmenu = glutCreateMenu(levelMenu);
for(int i = 1; i < 8; ++i){
char level[16];
sprintf(level, "Level %d\n", i);
glutAddMenuEntry(level, i);
}
int smenu = glutCreateMenu(schemeMenu);
glutAddMenuEntry("Catmark", 0);
glutAddMenuEntry("Bilinear", 1);
// Register Osd compute kernels
OpenSubdiv::OsdCpuKernelDispatcher::Register();
#if OPENSUBDIV_HAS_GLSL
OpenSubdiv::OsdGlslKernelDispatcher::Register();
#endif
#if OPENSUBDIV_HAS_OPENCL
OpenSubdiv::OsdClKernelDispatcher::Register();
#endif
#if OPENSUBDIV_HAS_CUDA
OpenSubdiv::OsdCudaKernelDispatcher::Register();
// Note: This function randomly crashes with linux 5.0-dev driver.
// cudaGetDeviceProperties overrun stack..?
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
#endif
int kmenu = glutCreateMenu(kernelMenu);
int nKernels = OpenSubdiv::OsdKernelDispatcher::kMAX;
for(int i = 0; i < nKernels; ++i)
if(OpenSubdiv::OsdKernelDispatcher::HasKernelType(
OpenSubdiv::OsdKernelDispatcher::KernelType(i)))
glutAddMenuEntry(getKernelName(i), i);
glutCreateMenu(menu);
glutAddSubMenu("Level", lmenu);
glutAddSubMenu("Scheme", smenu);
glutAddSubMenu("Kernel", kmenu);
glutAttachMenu(GLUT_RIGHT_BUTTON);
glutDisplayFunc(display);
glutReshapeFunc(reshape);
glutMouseFunc(mouse);
glutKeyboardFunc(keyboard);
glutMotionFunc(motion);
glewInit();
initGL();
for (int i = 1; i < argc; ++i) {
if (!strcmp(argv[i], "-d"))
g_level = atoi(argv[++i]);
else if (!strcmp(argv[i], "-c"))
g_repeatCount = atoi(argv[++i]);
else if (g_ptexColorFile == NULL)
g_ptexColorFile = argv[i];
else if (g_ptexDisplacementFile == NULL)
g_ptexDisplacementFile = argv[i];
else if (g_ptexOcclusionFile == NULL)
g_ptexOcclusionFile = argv[i];
}
if (g_ptexColorFile == NULL) {
printf("Usage: %s <color.ptx> [<displacement.ptx>] [<occlusion.ptx>] \n", argv[0]);
return 1;
}
createOsdMesh(g_level, g_kernel);
fitFrame();
glutIdleFunc(idle);
glutMainLoop();
quit();
}
int main(int argc, char** argv)
{
ModelParameters model_params;
fillCalculationParameters(model_params);
fillDerivedParameters(model_params, params);
if (CUTFalse == initGL(argc, argv, params))
return CUTFalse;
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if(cutCheckCmdLineFlag(argc, (const char**)argv, "device"))
cutilDeviceInit(argc, argv);
else
cudaSetDevice(cutGetMaxGflopsDeviceId());
// initialize calculations
initConstants(params);
timeval init_start, init_stop;
// calculate steady state
value_pair *steady_state = new value_pair[params.cells];
initSpectre();
initWaveVectors(params);
gettimeofday(&init_start, NULL);
calculateSteadyState(steady_state, params);
gettimeofday(&init_stop, NULL);
printf("Steady state calculation: %.3f s\n", time_diff(init_start, init_stop));
/*
FILE *f = fopen("plot_gs_mu.txt", "w");
int shift = (params.nvz / 2) * params.nvx * params.nvy + (params.nvy / 2) * params.nvx;
for(int i = 0; i < params.nvx; i++)
{
value_pair val = steady_state[shift + i];
fprintf(f, "%f %f\n", (-params.xmax + params.dx * i) * 1000000, (val.x * val.x + val.y * val.y));
}
fclose(f);
*/
gettimeofday(&init_start, NULL);
state.init(params);
initEvolution(steady_state, params, state);
gettimeofday(&init_stop, NULL);
printf("Evolution init: %.3f s\n", time_diff(init_start, init_stop));
delete[] steady_state;
// measure propagation time, for testing purposes
calculateEvolution(params, state, 0.0); // warm-up
gettimeofday(&init_start, NULL);
calculateEvolution(params, state, 0.0); // zero time step - because we are just measuring speed here
gettimeofday(&init_stop, NULL);
printf("Propagation time: %.3f ms\n", time_diff(init_start, init_stop) * 1000.0f);
// prepare textures
a_xy.init(params.nvx, params.nvy);
b_xy.init(params.nvx, params.nvy);
a_zy.init(params.nvz, params.nvy);
b_zy.init(params.nvz, params.nvy);
// remember starting time
gettimeofday(&time_start, NULL);
// start main application cycle
atexit(cleanup);
glutMainLoop();
return 0;
}
int main()
{
int width = 512;
int height = 512;
// Creation du device
cutilSafeCall( cudaSetDevice( cutGetMaxGflopsDeviceId() ) );
// Creation des buffers sur CPU
int * a = new int[width*height];
int * b = new int[width*height];
int * res = new int[width*height];
for(int i = 0; i < width*height; i++)
{
a[i] = (int)ceil(((double)rand()/ (double)RAND_MAX)*100);
b[i] = (int)ceil(((double)rand()/ (double)RAND_MAX)*100);
}
// Allocation des objects on the device
// *** data
unsigned int size = width * height * sizeof(int);
std::cout << "Allocation d'un buffer de taille " << width*height << "\n";
int* d_a = NULL;
int* d_b = NULL;
int* d_res = NULL;
cutilSafeCall( cudaMalloc( (void**) &d_a, size));
cutilSafeCall( cudaMalloc( (void**) &d_b, size));
cutilSafeCall( cudaMalloc( (void**) &d_res, size));
// Copy des donnees
cutilSafeCall( cudaMemcpy(d_a, a, size, cudaMemcpyHostToDevice));
cutilSafeCall( cudaMemcpy(d_b, b, size, cudaMemcpyHostToDevice));
// Lancer le calcul
std::cout << "Lancer le kernel ... \n";
runKernel(d_a, d_b, d_res, width*height);
cutilSafeCall( cutilDeviceSynchronize() );
// Copie DeviceHost
cutilSafeCall( cudaMemcpy(res, d_res, size, cudaMemcpyDeviceToHost));
// Verification du test
int i = 0;
for(;i < width*height;i++)
if(res[i] != a[i] + b[i])
std::cout << "Error : [" << i << "] " << res[i] << " != " << a[i] << " + " << b[i] << std::endl;
// Liberation des ressources
// *** Device
cutilSafeCall(cudaFree(d_a));
cutilSafeCall(cudaFree(d_b));
cutilSafeCall(cudaFree(d_res));
// *** CPU
delete[] res;
delete[] a;
delete[] b;
// Close device
cutilDeviceReset();
std::cout << "Test result : " << ((i == width*height) ? "Succes" : "Error" ) << std::endl;
std::cout.flush();
return 0;
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int main( int argc, char** argv)
{
shrQAStart(argc, argv);
// start logs
shrSetLogFileName ("bilateralFilter.txt");
shrLog("%s Starting...\n\n", argv[0]);
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
cutGetCmdLineArgumenti( argc, (const char**) argv, "threads", &nthreads );
cutGetCmdLineArgumenti( argc, (const char**) argv, "radius", &filter_radius);
// load image to process
loadImageData(argc, argv);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "qatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
// Running CUDA kernel (bilateralFilter) without visualization (QA Testing/Verification)
runAutoTest(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else if (cutCheckCmdLineFlag(argc, (const char **)argv, "benchmark"))
{
// Running CUDA kernel (bilateralFilter) in Benchmarking Mode
runBenchmark(argc, argv);
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
else
{
// Running CUDA kernel (bilateralFilter) in CUDA + OpenGL Visualization Mode
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
printf("[%s]\n", argv[0]);
printf(" Does not explicitly support -device=n in OpenGL mode\n");
printf(" To use -device=n, the sample must be running w/o OpenGL\n\n");
printf(" > %s -device=n -qatest\n", argv[0]);
printf("exiting...\n");
exit(0);
}
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL( argc, argv );
if ( cutCheckCmdLineFlag(argc, (const char **)argv, "device")) {
cutilGLDeviceInit(argc, argv);
} else {
cudaGLSetGLDevice (cutGetMaxGflopsDeviceId() );
}
initCuda();
initOpenGL();
}
atexit(cleanup);
printf("Running Standard Demonstration with GLUT loop...\n\n");
printf("Press '+' and '-' to change number of iterations\n"
"Press LEFT and RIGHT change euclidean delta\n"
"Press UP and DOWN to change gaussian delta\n"
"Press '1' to show original image\n"
"Press '2' to show result\n\n");
glutMainLoop();
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, (g_TotalErrors == 0 ? QA_PASSED : QA_FAILED));
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int
main( int argc, char** argv)
{
//start logs
shrSetLogFileName ("volumeRender.txt");
shrLog("%s Starting...\n\n", argv[0]);
if (cutCheckCmdLineFlag(argc, (const char **)argv, "qatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
g_bQAReadback = true;
fpsLimit = frameCheckNumber;
}
if (cutCheckCmdLineFlag(argc, (const char **)argv, "glverify"))
{
g_bQAGLVerify = true;
fpsLimit = frameCheckNumber;
}
if (g_bQAReadback) {
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilDeviceInit(argc, argv);
} else {
cudaSetDevice( cutGetMaxGflopsDeviceId() );
}
} else {
// First initialize OpenGL context, so we can properly set the GL for CUDA.
// This is necessary in order to achieve optimal performance with OpenGL/CUDA interop.
initGL( &argc, argv );
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") ) {
cutilGLDeviceInit(argc, argv);
} else {
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
}
/*
int device;
struct cudaDeviceProp prop;
cudaGetDevice( &device );
cudaGetDeviceProperties( &prop, device );
if( !strncmp( "Tesla", prop.name, 5 ) ) {
shrLog("This sample needs a card capable of OpenGL and display.\n");
shrLog("Please choose a different device with the -device=x argument.\n");
cutilExit(argc, argv);
}
*/
}
// parse arguments
char *filename;
if (cutGetCmdLineArgumentstr( argc, (const char**) argv, "file", &filename)) {
volumeFilename = filename;
}
int n;
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "size", &n)) {
volumeSize.width = volumeSize.height = volumeSize.depth = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "xsize", &n)) {
volumeSize.width = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "ysize", &n)) {
volumeSize.height = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "zsize", &n)) {
volumeSize.depth = n;
}
// load volume data
char* path = shrFindFilePath(volumeFilename, argv[0]);
if (path == 0) {
shrLog("Error finding file '%s'\n", volumeFilename);
exit(EXIT_FAILURE);
}
size_t size = volumeSize.width*volumeSize.height*volumeSize.depth*sizeof(VolumeType);
void *h_volume = loadRawFile(path, size);
initCuda(h_volume, volumeSize);
free(h_volume);
cutilCheckError( cutCreateTimer( &timer));
shrLog("Press '=' and '-' to change density\n"
" ']' and '[' to change brightness\n"
" ';' and ''' to modify transfer function offset\n"
" '.' and ',' to modify transfer function scale\n\n");
// calculate new grid size
gridSize = dim3(iDivUp(width, blockSize.x), iDivUp(height, blockSize.y));
if (g_bQAReadback) {
g_CheckRender = new CheckBackBuffer(width, height, 4, false);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
uint *d_output;
cutilSafeCall(cudaMalloc((void**)&d_output, width*height*sizeof(uint)));
cutilSafeCall(cudaMemset(d_output, 0, width*height*sizeof(uint)));
float modelView[16] =
{
1.0f, 0.0f, 0.0f, 0.0f,
0.0f, 1.0f, 0.0f, 0.0f,
0.0f, 0.0f, 1.0f, 0.0f,
0.0f, 0.0f, 4.0f, 1.0f
};
invViewMatrix[0] = modelView[0]; invViewMatrix[1] = modelView[4]; invViewMatrix[2] = modelView[8]; invViewMatrix[3] = modelView[12];
invViewMatrix[4] = modelView[1]; invViewMatrix[5] = modelView[5]; invViewMatrix[6] = modelView[9]; invViewMatrix[7] = modelView[13];
invViewMatrix[8] = modelView[2]; invViewMatrix[9] = modelView[6]; invViewMatrix[10] = modelView[10]; invViewMatrix[11] = modelView[14];
// call CUDA kernel, writing results to PBO
copyInvViewMatrix(invViewMatrix, sizeof(float4)*3);
// Start timer 0 and process n loops on the GPU
int nIter = 10;
for (int i = -1; i < nIter; i++)
{
if( i == 0 ) {
cudaThreadSynchronize();
cutStartTimer(timer);
}
render_kernel(gridSize, blockSize, d_output, width, height, density, brightness, transferOffset, transferScale);
}
cudaThreadSynchronize();
cutStopTimer(timer);
// Get elapsed time and throughput, then log to sample and master logs
double dAvgTime = cutGetTimerValue(timer)/(nIter * 1000.0);
shrLogEx(LOGBOTH | MASTER, 0, "volumeRender, Throughput = %.4f MTexels/s, Time = %.5f s, Size = %u Texels, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * width * height)/dAvgTime, dAvgTime, (width * height), 1, blockSize.x * blockSize.y);
cutilCheckMsg("Error: render_kernel() execution FAILED");
cutilSafeCall( cudaThreadSynchronize() );
cutilSafeCall( cudaMemcpy(g_CheckRender->imageData(), d_output, width*height*4, cudaMemcpyDeviceToHost) );
g_CheckRender->savePPM(sOriginal[g_Index], true, NULL);
if (!g_CheckRender->PPMvsPPM(sOriginal[g_Index], sReference[g_Index], MAX_EPSILON_ERROR, THRESHOLD)) {
shrLog("\nFAILED\n\n");
} else {
shrLog("\nPASSED\n\n");
}
cudaFree(d_output);
freeCudaBuffers();
if (g_CheckRender) {
delete g_CheckRender; g_CheckRender = NULL;
}
} else {
// This is the normal rendering path for VolumeRender
glutDisplayFunc(display);
glutKeyboardFunc(keyboard);
glutMouseFunc(mouse);
glutMotionFunc(motion);
glutReshapeFunc(reshape);
glutIdleFunc(idle);
initPixelBuffer();
if (g_bQAGLVerify) {
g_CheckRender = new CheckBackBuffer(width, height, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
atexit(cleanup);
glutMainLoop();
}
cudaThreadExit();
shrEXIT(argc, (const char**)argv);
}
int main(int argc, char **argv)
{
GpuProfiling::initProf();
// Start logs
shrSetLogFileName ("scan.txt");
shrLog("%s Starting...\n\n", argv[0]);
//Use command-line specified CUDA device, otherwise use device with highest Gflops/s
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") )
cutilDeviceInit(argc, argv);
else
cudaSetDevice( cutGetMaxGflopsDeviceId() );
uint *d_Input, *d_Output;
uint *h_Input, *h_OutputCPU, *h_OutputGPU;
uint hTimer;
const uint N = 13 * 1048576 / 2;
shrLog("Allocating and initializing host arrays...\n");
cutCreateTimer(&hTimer);
h_Input = (uint *)malloc(N * sizeof(uint));
h_OutputCPU = (uint *)malloc(N * sizeof(uint));
h_OutputGPU = (uint *)malloc(N * sizeof(uint));
srand(2009);
for(uint i = 0; i < N; i++)
h_Input[i] = rand();
shrLog("Allocating and initializing CUDA arrays...\n");
cutilSafeCall( cudaMalloc((void **)&d_Input, N * sizeof(uint)) );
cutilSafeCall( cudaMalloc((void **)&d_Output, N * sizeof(uint)) );
cutilSafeCall( cudaMemcpy(d_Input, h_Input, N * sizeof(uint), cudaMemcpyHostToDevice) );
shrLog("Initializing CUDA-C scan...\n\n");
initScan();
int globalFlag = 1;
size_t szWorkgroup;
const int iCycles = 100;
shrLog("*** Running GPU scan for short arrays (%d identical iterations)...\n\n", iCycles);
for(uint arrayLength = MIN_SHORT_ARRAY_SIZE; arrayLength <= MAX_SHORT_ARRAY_SIZE; arrayLength <<= 1){
shrLog("Running scan for %u elements (%u arrays)...\n", arrayLength, N / arrayLength);
cutilSafeCall( cudaThreadSynchronize() );
cutResetTimer(hTimer);
cutStartTimer(hTimer);
for(int i = 0; i < iCycles; i++)
{
szWorkgroup = scanExclusiveShort(d_Output, d_Input, N / arrayLength, arrayLength);
}
cutilSafeCall( cudaThreadSynchronize());
cutStopTimer(hTimer);
double timerValue = 1.0e-3 * cutGetTimerValue(hTimer) / iCycles;
shrLog("Validating the results...\n");
shrLog("...reading back GPU results\n");
cutilSafeCall( cudaMemcpy(h_OutputGPU, d_Output, N * sizeof(uint), cudaMemcpyDeviceToHost) );
shrLog(" ...scanExclusiveHost()\n");
scanExclusiveHost(h_OutputCPU, h_Input, N / arrayLength, arrayLength);
// Compare GPU results with CPU results and accumulate error for this test
shrLog(" ...comparing the results\n");
int localFlag = 1;
for(uint i = 0; i < N; i++)
{
if(h_OutputCPU[i] != h_OutputGPU[i])
{
localFlag = 0;
break;
}
}
// Log message on individual test result, then accumulate to global flag
shrLog(" ...Results %s\n\n", (localFlag == 1) ? "Match" : "DON'T Match !!!");
globalFlag = globalFlag && localFlag;
// Data log
if (arrayLength == MAX_SHORT_ARRAY_SIZE)
{
shrLog("\n");
shrLogEx(LOGBOTH | MASTER, 0, "scan-Short, Throughput = %.4f MElements/s, Time = %.5f s, Size = %u Elements, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * (double)arrayLength/timerValue), timerValue, arrayLength, 1, szWorkgroup);
shrLog("\n");
}
}
shrLog("***Running GPU scan for large arrays (%u identical iterations)...\n\n", iCycles);
for(uint arrayLength = MIN_LARGE_ARRAY_SIZE; arrayLength <= MAX_LARGE_ARRAY_SIZE; arrayLength <<= 1){
shrLog("Running scan for %u elements (%u arrays)...\n", arrayLength, N / arrayLength);
cutilSafeCall( cudaThreadSynchronize() );
cutResetTimer(hTimer);
cutStartTimer(hTimer);
for(int i = 0; i < iCycles; i++)
{
szWorkgroup = scanExclusiveLarge(d_Output, d_Input, N / arrayLength, arrayLength);
}
cutilSafeCall( cudaThreadSynchronize() );
cutStopTimer(hTimer);
double timerValue = 1.0e-3 * cutGetTimerValue(hTimer) / iCycles;
shrLog("Validating the results...\n");
shrLog("...reading back GPU results\n");
cutilSafeCall( cudaMemcpy(h_OutputGPU, d_Output, N * sizeof(uint), cudaMemcpyDeviceToHost) );
shrLog("...scanExclusiveHost()\n");
scanExclusiveHost(h_OutputCPU, h_Input, N / arrayLength, arrayLength);
// Compare GPU results with CPU results and accumulate error for this test
shrLog(" ...comparing the results\n");
int localFlag = 1;
for(uint i = 0; i < N; i++)
{
if(h_OutputCPU[i] != h_OutputGPU[i])
{
localFlag = 0;
break;
}
}
// Log message on individual test result, then accumulate to global flag
shrLog(" ...Results %s\n\n", (localFlag == 1) ? "Match" : "DON'T Match !!!");
globalFlag = globalFlag && localFlag;
// Data log
if (arrayLength == MAX_LARGE_ARRAY_SIZE)
{
shrLog("\n");
shrLogEx(LOGBOTH | MASTER, 0, "scan-Large, Throughput = %.4f MElements/s, Time = %.5f s, Size = %u Elements, NumDevsUsed = %u, Workgroup = %u\n",
(1.0e-6 * (double)arrayLength/timerValue), timerValue, arrayLength, 1, szWorkgroup);
shrLog("\n");
}
}
// pass or fail (cumulative... all tests in the loop)
shrLog(globalFlag ? "PASSED\n\n" : "FAILED\n\n");
GpuProfiling::printResults();
shrLog("Shutting down...\n");
closeScan();
cutilSafeCall( cudaFree(d_Output));
cutilSafeCall( cudaFree(d_Input));
cutilCheckError( cutDeleteTimer(hTimer) );
cudaThreadExit();
exit(0);
shrEXIT(argc, (const char**)argv);
}
