void initGL(int *argc, char **argv)
{
// 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");
if (bFullscreen)
{
glutFullScreen();
}
GLenum err = glewInit();
if (GLEW_OK != err)
{
printf("GLEW Error: %s\n", glewGetErrorString(err));
cudaDeviceReset();
exit(EXIT_FAILURE);
}
else if (!glewIsSupported("GL_VERSION_2_0 "
"GL_VERSION_1_5 "
"GL_ARB_multitexture "
"GL_ARB_vertex_buffer_object"))
{
fprintf(stderr, "Required OpenGL extensions missing.");
exit(EXIT_FAILURE);
}
else
{
#if defined(WIN32)
wglSwapIntervalEXT(0);
#elif defined(LINUX)
glxSwapIntervalSGI(0);
#endif
}
glEnable(GL_DEPTH_TEST);
glClearColor(0.0, 0.0, 0.0, 1.0);
checkGLErrors("initGL");
}
//////////////////////////////////////////////////////////////////////////////
// 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;
}
