int main(int argc, char **argv) { int devID; cudaDeviceProp deviceProps; printf("%s Starting...\n\n", sSDKname); printf("[%s] - [OpenGL/CUDA simulation] starting...\n", sSDKname); // 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 (false == initGL(&argc, argv)) { exit(EXIT_SUCCESS); } // use command-line specified CUDA device, otherwise use device with highest Gflops/s devID = findCudaGLDevice(argc, (const char **)argv); // get number of SMs on this GPU checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID)); printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount); // automated build testing harness if (checkCmdLineFlag(argc, (const char **)argv, "file")) { getCmdLineArgumentString(argc, (const char **)argv, "file", &ref_file); } // Allocate and initialize host data GLint bsize; sdkCreateTimer(&timer); sdkResetTimer(&timer); hvfield = (cData *)malloc(sizeof(cData) * DS); memset(hvfield, 0, sizeof(cData) * DS); // Allocate and initialize device data cudaMallocPitch((void **)&dvfield, &tPitch, sizeof(cData)*DIM, DIM); cudaMemcpy(dvfield, hvfield, sizeof(cData) * DS, cudaMemcpyHostToDevice); // Temporary complex velocity field data cudaMalloc((void **)&vxfield, sizeof(cData) * PDS); cudaMalloc((void **)&vyfield, sizeof(cData) * PDS); setupTexture(DIM, DIM); bindTexture(); // Create particle array particles = (cData *)malloc(sizeof(cData) * DS); memset(particles, 0, sizeof(cData) * DS); initParticles(particles, DIM, DIM); // Create CUFFT transform plan configuration cufftPlan2d(&planr2c, DIM, DIM, CUFFT_R2C); cufftPlan2d(&planc2r, DIM, DIM, CUFFT_C2R); // TODO: update kernels to use the new unpadded memory layout for perf // rather than the old FFTW-compatible layout cufftSetCompatibilityMode(planr2c, CUFFT_COMPATIBILITY_FFTW_PADDING); cufftSetCompatibilityMode(planc2r, CUFFT_COMPATIBILITY_FFTW_PADDING); glGenBuffersARB(1, &vbo); glBindBufferARB(GL_ARRAY_BUFFER_ARB, vbo); glBufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(cData) * DS, particles, GL_DYNAMIC_DRAW_ARB); glGetBufferParameterivARB(GL_ARRAY_BUFFER_ARB, GL_BUFFER_SIZE_ARB, &bsize); if (bsize != (sizeof(cData) * DS)) goto EXTERR; glBindBufferARB(GL_ARRAY_BUFFER_ARB, 0); checkCudaErrors(cudaGraphicsGLRegisterBuffer(&cuda_vbo_resource, vbo, cudaGraphicsMapFlagsNone)); getLastCudaError("cudaGraphicsGLRegisterBuffer failed"); if (ref_file) { autoTest(argv); cleanup(); cudaDeviceReset(); printf("[fluidsGL] - Test Results: %d Failures\n", g_TotalErrors); exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE); } else { atexit(cleanup); glutMainLoop(); } cudaDeviceReset(); if (!ref_file) { exit(EXIT_SUCCESS); } return 0; EXTERR: printf("Failed to initialize GL extensions.\n"); cudaDeviceReset(); exit(EXIT_FAILURE); }
int main(int argc, char **argv) { int devID; cudaDeviceProp deviceProps; printf("%s Starting...\n\n", sSDKname); printf("[%s] - [OpenGL/CUDA simulation] starting...\n", sSDKname); // 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 (false == initGL(&argc, argv)) { exit(EXIT_SUCCESS); } // use command-line specified CUDA device, otherwise use device with highest Gflops/s #ifndef OPTIMUS devID = findCudaGLDevice(argc, (const char **)argv); #else devID = gpuGetMaxGflopsDeviceId(); #endif // get number of SMs on this GPU checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID)); printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount); // automated build testing harness if (checkCmdLineFlag(argc, (const char **)argv, "file")) { getCmdLineArgumentString(argc, (const char **)argv, "file", &ref_file); } // Allocate and initialize host data GLint bsize; sdkCreateTimer(&timer); sdkResetTimer(&timer); hvfield = (cData *)malloc(sizeof(cData) * DS); memset(hvfield, 0, sizeof(cData) * DS); // Allocate and initialize device data cudaMallocPitch((void **)&dvfield, &tPitch, sizeof(cData)*DIM, DIM); cudaMemcpy(dvfield, hvfield, sizeof(cData) * DS, cudaMemcpyHostToDevice); // Temporary complex velocity field data cudaMalloc((void **)&vxfield, sizeof(cData) * PDS); cudaMalloc((void **)&vyfield, sizeof(cData) * PDS); setupTexture(DIM, DIM); bindTexture(); // Create particle array in host memory particles = (cData *)malloc(sizeof(cData) * DS); memset(particles, 0, sizeof(cData) * DS); #ifdef BROADCAST int step = 1; // Broadcasted visualization stepping. if (argc > 3) step = atoi(argv[3]); // Create additional space to store particle packets // for broadcasting. wstep = step; hstep = step; int npackets = sizeof(float) * (DIM / wstep) * (DIM / hstep) / UdpBroadcastServer::PacketSize; if (sizeof(float) * (DIM / wstep) * (DIM / hstep) % UdpBroadcastServer::PacketSize) npackets++; packets = (char*)malloc(npackets * (UdpBroadcastServer::PacketSize + sizeof(unsigned int))); #endif initParticles(particles, DIM, DIM); #if defined(OPTIMUS) || defined(BROADCAST) // Create particle array in device memory cudaMalloc((void **)&particles_gpu, sizeof(cData) * DS); cudaMemcpy(particles_gpu, particles, sizeof(cData) * DS, cudaMemcpyHostToDevice); #endif // Create CUFFT transform plan configuration cufftPlan2d(&planr2c, DIM, DIM, CUFFT_R2C); cufftPlan2d(&planc2r, DIM, DIM, CUFFT_C2R); // TODO: update kernels to use the new unpadded memory layout for perf // rather than the old FFTW-compatible layout cufftSetCompatibilityMode(planr2c, CUFFT_COMPATIBILITY_FFTW_PADDING); cufftSetCompatibilityMode(planc2r, CUFFT_COMPATIBILITY_FFTW_PADDING); glGenBuffersARB(1, &vbo); glBindBufferARB(GL_ARRAY_BUFFER_ARB, vbo); glBufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(cData) * DS, particles, GL_DYNAMIC_DRAW_ARB); glGetBufferParameterivARB(GL_ARRAY_BUFFER_ARB, GL_BUFFER_SIZE_ARB, &bsize); if (bsize != (sizeof(cData) * DS)) goto EXTERR; glBindBufferARB(GL_ARRAY_BUFFER_ARB, 0); #ifndef OPTIMUS checkCudaErrors(cudaGraphicsGLRegisterBuffer(&cuda_vbo_resource, vbo, cudaGraphicsMapFlagsNone)); getLastCudaError("cudaGraphicsGLRegisterBuffer failed"); #endif if (ref_file) { autoTest(argv); cleanup(); // cudaDeviceReset causes the driver to clean up all state. While // not mandatory in normal operation, it is good practice. It is also // needed to ensure correct operation when the application is being // profiled. Calling cudaDeviceReset causes all profile data to be // flushed before the application exits cudaDeviceReset(); printf("[fluidsGL] - Test Results: %d Failures\n", g_TotalErrors); exit(g_TotalErrors == 0 ? EXIT_SUCCESS : EXIT_FAILURE); } else { #ifdef BROADCAST const char *sv_addr = "127.0.0:9097"; const char *bc_addr = "127.255.255.2:9097"; // Server address if (argc > 2) sv_addr = argv[2]; // Broadcast address if (argc > 1) bc_addr = argv[1]; server.reset(new UdpBroadcastServer(sv_addr, bc_addr)); // Listen to clients' feedbacks in a separate thread. { pthread_t tid; pthread_create(&tid, NULL, &feedback_listener, &step); } // Broadcast the particles state in a separate thread. { pthread_t tid; pthread_create(&tid, NULL, &broadcaster, &step); } #endif #if defined (__APPLE__) || defined(MACOSX) atexit(cleanup); #else glutCloseFunc(cleanup); #endif glutMainLoop(); } // cudaDeviceReset causes the driver to clean up all state. While // not mandatory in normal operation, it is good practice. It is also // needed to ensure correct operation when the application is being // profiled. Calling cudaDeviceReset causes all profile data to be // flushed before the application exits cudaDeviceReset(); if (!ref_file) { exit(EXIT_SUCCESS); } return 0; EXTERR: printf("Failed to initialize GL extensions.\n"); // cudaDeviceReset causes the driver to clean up all state. While // not mandatory in normal operation, it is good practice. It is also // needed to ensure correct operation when the application is being // profiled. Calling cudaDeviceReset causes all profile data to be // flushed before the application exits cudaDeviceReset(); exit(EXIT_FAILURE); }