static CUresult
initCuda(CUcontext _cuContext, char* executablePath, CUfunction *mathop,
int argc, char** argv, const char* cubin_name, const char* kernel_name)
{
CUdevice cuDevice;
CUT_DEVICE_INIT_DRV(cuDevice, argc, argv);
print_GetProperties(cuDevice);
CUresult status = cuCtxCreate( &_cuContext, 0, cuDevice );
if ( CUDA_SUCCESS != status ) {
Error(_cuContext, status);
}
else printf("(1) context creation successful\n");
char* module_path = cutFindFilePath(cubin_name, executablePath);
printf ("\t cubin:%s, path:%s, mmp_ptr:%lu\n", cubin_name, executablePath, module_path);
if(module_path != NULL)
printf ("\t cubin:%s, path:%s, module_path:%c%c%c%c\n", cubin_name, executablePath, *module_path, *(module_path+1), *(module_path+2), *(module_path+3));
char* data_path = "./data/";
size_t len_path = strlen(data_path);
size_t len_fn = strlen(cubin_name);
// printf ("Sizes: data:%lu, cubinname:%lu\n", len_path, len_fn);
char* module_path_new = (char*)malloc(sizeof(char) * (len_path + len_fn));
strcpy(module_path_new, data_path);
strcat(module_path_new, cubin_name);
strcat(module_path_new, "\0");
if (module_path_new == 0) {
status = CUDA_ERROR_NOT_FOUND;
Error(_cuContext, status);
}
FILE *fp = fopen(module_path_new,"r");
if( fp ) {
printf("(2) cubin_File found in modulepath:%s\n", module_path_new);
fclose(fp);
} else {
printf("(2) cubin file not exist: %s\n", module_path_new);
}
CUmodule cuModule;
status = cuModuleLoad(&cuModule, module_path_new);
cutFree(module_path_new);
if ( CUDA_SUCCESS != status ) {
Error(_cuContext, status);
}
else printf ("(3) module Load successful\n");
CUfunction cuFunction = 0;
status = cuModuleGetFunction(&cuFunction, cuModule, kernel_name);
if ( CUDA_SUCCESS != status) {
Error(_cuContext, status);
}
else printf ("(4) getFunction successful w/cuFunction\n");
*mathop = cuFunction;
return CUDA_SUCCESS;
}
////////////////////////////////////////////////////////////////////////////////
//! Initialize GL
////////////////////////////////////////////////////////////////////////////////
CUTBoolean initGL(int *argc, char **argv)
{
// Create GL context
glutInit(argc, argv);
glutInitDisplayMode(GLUT_RGBA | GLUT_DOUBLE | GLUT_DEPTH);
glutInitWindowSize(windowW, windowH);
glutCreateWindow("CUDA FFT Ocean Simulation");
vertShaderPath = cutFindFilePath("ocean.vert", argv[0]);
fragShaderPath = cutFindFilePath("ocean.frag", argv[0]);
if (vertShaderPath == 0 || fragShaderPath == 0) {
fprintf(stderr, "Error finding shader files!\n");
cudaThreadExit();
exit(EXIT_FAILURE);
}
// initialize necessary OpenGL extensions
glewInit();
if (! glewIsSupported("GL_VERSION_2_0 "
)) {
fprintf(stderr, "ERROR: Support for necessary OpenGL extensions missing.");
fflush(stderr);
return CUTFalse;
}
if (!glewIsSupported( "GL_VERSION_1_5 GL_ARB_vertex_buffer_object GL_ARB_pixel_buffer_object" )) {
fprintf(stderr, "Error: failed to get minimal extensions for demo\n");
fprintf(stderr, "This sample requires:\n");
fprintf(stderr, " OpenGL version 1.5\n");
fprintf(stderr, " GL_ARB_vertex_buffer_object\n");
fprintf(stderr, " GL_ARB_pixel_buffer_object\n");
cleanup();
exit(-1);
}
// default initialization
glClearColor(0.0, 0.0, 0.0, 1.0);
glEnable(GL_DEPTH_TEST);
// load shader
shaderProg = loadGLSLProgram(vertShaderPath, fragShaderPath);
CUT_CHECK_ERROR_GL();
return CUTTrue;
}
void loadDefaultImage( int argc, char ** argv ) {
printf("Reading image: lena.pgm\n");
const char* image_filename = "lena.pgm";
char* image_path = cutFindFilePath(image_filename, argv[0] );
if (image_path == 0) {
printf( "Reading image failed.\n");
exit(EXIT_FAILURE);
}
initializeData( image_path, argc, argv );
cutFree( image_path );
}
void loadDefaultImage( int argc, char ** argv ) {
printf("Loading input image: lena.pgm\n");
const char* image_filename = "lena.pgm";
char* image_path = cutFindFilePath(image_filename, argv[0] );
if (image_path == NULL) {
printf( "FunctionPointers unable to find and load image file <%s>.\nExiting...\n", image_filename);
shrQAFinishExit(argc, (const char **)argv, QA_FAILED);
}
initializeData( image_path, argc, argv );
cutFree( image_path );
}
void runAutoTest(int argc, char **argv)
{
int devID = 0;
printf("[%s] - (automated testing w/ readback)\n", sSDKsample);
devID = cutilChooseCudaDevice(argc, argv);
// First load the image, so we know what the size of the image (imageW and imageH)
printf("Allocating host and CUDA memory and loading image file...\n");
const char *image_path = cutFindFilePath("portrait_noise.bmp", argv[0]);
if (image_path == NULL) {
printf( "imageDenoisingGL was unable to find and load image file <portrait_noise.bmp>.\nExiting...\n");
shrQAFinishExit(argc, (const char **)argv, QA_FAILED);
}
LoadBMPFile(&h_Src, &imageW, &imageH, image_path);
printf("Data init done.\n");
cutilSafeCall( CUDA_MallocArray(&h_Src, imageW, imageH) );
g_CheckRender = new CheckBackBuffer(imageW, imageH, sizeof(TColor), false);
g_CheckRender->setExecPath(argv[0]);
TColor *d_dst = NULL;
cutilSafeCall( cudaMalloc( (void **)&d_dst, imageW*imageH*sizeof(TColor)) );
while (g_Kernel <= 3) {
printf("[AutoTest]: %s <%s>\n", sSDKsample, filterMode[g_Kernel]);
cutilSafeCall( CUDA_Bind2TextureArray() );
runImageFilters(d_dst);
cutilSafeCall( CUDA_UnbindTexture() );
cutilSafeCall( cutilDeviceSynchronize() );
cudaMemcpy(g_CheckRender->imageData(), d_dst, imageW*imageH*sizeof(TColor), cudaMemcpyDeviceToHost);
g_CheckRender->savePPM(sOriginal[g_Kernel], true, NULL);
if (!g_CheckRender->PPMvsPPM(sOriginal[g_Kernel], sReference[g_Kernel], MAX_EPSILON_ERROR, 0.15f)) {
g_TotalErrors++;
}
g_Kernel++;
}
cutilSafeCall( CUDA_FreeArray() );
free(h_Src);
cutilSafeCall( cudaFree( d_dst ) );
delete g_CheckRender;
printf("\n[%s] -> Test Results: %d errors\n", sSDKsample, g_TotalErrors);
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, (!g_TotalErrors ? QA_PASSED : QA_FAILED));
}
bool CheckRenderD3D10::PPMvsPPM( const char *src_file, const char *ref_file, const char *exec_path,
const float epsilon, const float threshold )
{
char *ref_file_path = cutFindFilePath(ref_file, exec_path);
if (ref_file_path == NULL) {
printf("CheckRenderD3D10::PPMvsPPM unable to find <%s> in <%s> Aborting comparison!\n", ref_file, exec_path);
printf(">>> Check info.xml and [project//data] folder <%s> <<<\n", ref_file);
printf("Aborting comparison!\n");
printf(" FAILURE!\n");
return false;
}
return cutComparePPM(src_file,ref_file_path,epsilon,threshold,true) == CUTTrue;
}
void loadDefaultImage(char* loc_exec)
{
printf("Reading image: lena.pgm\n");
const char* image_filename = "lena.pgm";
char* image_path = cutFindFilePath(image_filename, loc_exec);
if (image_path == 0) {
printf( "Reading image: lena.pgm failed.\n");
exit(EXIT_FAILURE);
}
initializeData(image_path);
cutFree(image_path);
}
bool
CheckRender::PGMvsPGM( const char *src_file, const char *ref_file, const float epsilon, const float threshold )
{
unsigned char *src_data = NULL, *ref_data = NULL;
unsigned long error_count = 0;
unsigned int width, height;
char *ref_file_path = cutFindFilePath(ref_file, m_ExecPath);
if (ref_file_path == NULL) {
printf("CheckRender::PGMvsPGM unable to find <%s> in <%s> Aborting comparison!\n", ref_file, m_ExecPath);
printf(">>> Check info.xml and [project//data] folder <%s> <<<\n", ref_file);
printf("Aborting comparison!\n");
printf(" FAILED\n");
error_count++;
} else {
if (src_file == NULL || ref_file_path == NULL) {
printf("PGMvsPGM: Aborting comparison\n");
return false;
}
printf(" src_file <%s>\n", src_file);
printf(" ref_file <%s>\n", ref_file_path);
if (cutLoadPGMub(ref_file_path, &ref_data, &width, &height) != CUTTrue) {
printf("PGMvsPGM: unable to load ref image file: %s\n", ref_file_path);
return false;
}
if (cutLoadPGMub(src_file, &src_data, &width, &height) != CUTTrue) {
printf("PGMvsPGM: unable to load src image file: %s\n", src_file);
return false;
}
printf("PGMvsPGM: comparing images size (%d,%d) epsilon(%2.4f), threshold(%4.2f%%)\n", m_Height, m_Width, epsilon, threshold*100);
if (cutCompareubt( ref_data, src_data, m_Height*m_Width, epsilon, threshold ) == CUTFalse) {
error_count = 1;
}
}
if (error_count == 0) {
printf(" OK\n");
} else {
printf(" FAILURE: %d errors...\n", (unsigned int)error_count);
}
return (error_count == 0); // returns true if all pixels pass
}
// initialize OpenGL
void initGL(int *argc, char **argv)
{
glutInit(argc, argv);
glutInitDisplayMode(GLUT_RGB | GLUT_DEPTH | GLUT_DOUBLE);
glutInitWindowSize(winWidth, winHeight);
glutCreateWindow("CUDA Smoke Particles");
glewInit();
if (!glewIsSupported("GL_VERSION_2_0 GL_VERSION_1_5")) {
fprintf(stderr, "The following required OpenGL extensions missing:\n\tGL_VERSION_2_0\n\tGL_VERSION_1_5\n");
fprintf(stderr, " PASSED\n");
cutilExit(*argc, argv);
exit(-1);
}
if (!glewIsSupported("GL_ARB_multitexture GL_ARB_vertex_buffer_object GL_EXT_geometry_shader4")) {
fprintf(stderr, "The following required OpenGL extensions missing:\n\tGL_ARB_multitexture\n\tGL_ARB_vertex_buffer_object\n\tGL_EXT_geometry_shader4.\n");
fprintf(stderr, " PASSED\n");
cutilExit(*argc, argv);
exit(-1);
}
#if defined (_WIN32)
if (wglewIsSupported("WGL_EXT_swap_control")) {
// disable vertical sync
wglSwapIntervalEXT(0);
}
#endif
glEnable(GL_DEPTH_TEST);
// load floor texture
char* imagePath = cutFindFilePath("floortile.ppm", argv[0]);
if (imagePath == 0) {
fprintf(stderr, "Error finding floor image file\n");
fprintf(stderr, " FAILED\n");
exit(EXIT_FAILURE);
}
floorTex = loadTexture(imagePath);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_ANISOTROPY_EXT, 16.0f);
floorProg = new GLSLProgram(floorVS, floorPS);
glutReportErrors();
}
bool
CheckRender::PPMvsPPM( const char *src_file, const char *ref_file, const float epsilon, const float threshold )
{
unsigned long error_count = 0;
char *ref_file_path = cutFindFilePath(ref_file, m_ExecPath);
if (ref_file_path == NULL) {
printf("CheckRender::PPMvsPPM unable to find <%s> in <%s> Aborting comparison!\n", ref_file, m_ExecPath);
printf(">>> Check info.xml and [project//data] folder <%s> <<<\n", ref_file);
printf("Aborting comparison!\n");
printf(" FAILED\n");
error_count++;
}
if (src_file == NULL || ref_file_path == NULL) {
printf("PPMvsPPM: Aborting comparison\n");
return false;
}
printf(" src_file <%s>\n", src_file);
printf(" ref_file <%s>\n", ref_file_path);
return (cutComparePPM( src_file, ref_file_path, epsilon, threshold, true ) == CUTTrue ? true : false);
}
void loadImage(
unsigned char **img,
unsigned int *width,
unsigned int *height,
const char* image_filename,
const char* exec_path)
{
const char* image_path = cutFindFilePath(image_filename, exec_path);
if (image_path == 0) {
fprintf(stderr, "Error finding image file '%s'\n", image_filename);
exit(-1);
}
cutilCheckError(cutLoadPGMub(image_path, img, width, height));
if (!*img) {
fprintf(stderr, "Error opening file '%s'\n", image_path);
exit(-1);
}
printf("Loaded '%s', %d x %d pixels\n", image_path, *width, *height);
}
////////////////////////////////////////////////////////////////////////////////
// initialize marching cubes
////////////////////////////////////////////////////////////////////////////////
void
initMC(int argc, char** argv)
{
// parse command line arguments
int n;
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "grid", &n)) {
gridSizeLog2.x = gridSizeLog2.y = gridSizeLog2.z = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "gridx", &n)) {
gridSizeLog2.x = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "gridy", &n)) {
gridSizeLog2.y = n;
}
if (cutGetCmdLineArgumenti( argc, (const char**) argv, "gridz", &n)) {
gridSizeLog2.z = n;
}
char *filename;
if (cutGetCmdLineArgumentstr( argc, (const char**) argv, "file", &filename)) {
volumeFilename = filename;
}
gridSize = make_uint3(1<<gridSizeLog2.x, 1<<gridSizeLog2.y, 1<<gridSizeLog2.z);
gridSizeMask = make_uint3(gridSize.x-1, gridSize.y-1, gridSize.z-1);
gridSizeShift = make_uint3(0, gridSizeLog2.x, gridSizeLog2.x+gridSizeLog2.y);
numVoxels = gridSize.x*gridSize.y*gridSize.z;
voxelSize = make_float3(2.0f / gridSize.x, 2.0f / gridSize.y, 2.0f / gridSize.z);
maxVerts = gridSize.x*gridSize.y*100;
printf("grid: %d x %d x %d = %d voxels\n", gridSize.x, gridSize.y, gridSize.z, numVoxels);
printf("max verts = %d\n", maxVerts);
#if SAMPLE_VOLUME
// load volume data
char* path = cutFindFilePath(volumeFilename, argv[0]);
if (path == 0) {
fprintf(stderr, "Error finding file '%s'\n", volumeFilename);
cudaThreadExit();
exit(EXIT_FAILURE);
}
int size = gridSize.x*gridSize.y*gridSize.z*sizeof(uchar);
uchar *volume = loadRawFile(path, size);
cutilSafeCall(cudaMalloc((void**) &d_volume, size));
cutilSafeCall(cudaMemcpy(d_volume, volume, size, cudaMemcpyHostToDevice) );
free(volume);
bindVolumeTexture(d_volume);
#endif
if (g_bQAReadback) {
cudaMalloc((void **)&(d_pos), maxVerts*sizeof(float)*4);
cudaMalloc((void **)&(d_normal), maxVerts*sizeof(float)*4);
} else {
// create VBOs
createVBO(&posVbo, maxVerts*sizeof(float)*4);
// DEPRECATED: cutilSafeCall( cudaGLRegisterBufferObject(posVbo) );
cutilSafeCall(cudaGraphicsGLRegisterBuffer(&cuda_posvbo_resource, posVbo,
cudaGraphicsMapFlagsWriteDiscard));
createVBO(&normalVbo, maxVerts*sizeof(float)*4);
// DEPRECATED: cutilSafeCall(cudaGLRegisterBufferObject(normalVbo));
cutilSafeCall(cudaGraphicsGLRegisterBuffer(&cuda_normalvbo_resource, normalVbo,
cudaGraphicsMapFlagsWriteDiscard));
}
// allocate textures
allocateTextures( &d_edgeTable, &d_triTable, &d_numVertsTable );
// allocate device memory
unsigned int memSize = sizeof(uint) * numVoxels;
cutilSafeCall(cudaMalloc((void**) &d_voxelVerts, memSize));
cutilSafeCall(cudaMalloc((void**) &d_voxelVertsScan, memSize));
cutilSafeCall(cudaMalloc((void**) &d_voxelOccupied, memSize));
cutilSafeCall(cudaMalloc((void**) &d_voxelOccupiedScan, memSize));
cutilSafeCall(cudaMalloc((void**) &d_compVoxelArray, memSize));
// initialize CUDPP scan
CUDPPConfiguration config;
config.algorithm = CUDPP_SCAN;
config.datatype = CUDPP_UINT;
config.op = CUDPP_ADD;
config.options = CUDPP_OPTION_FORWARD | CUDPP_OPTION_EXCLUSIVE;
cudppPlan(&scanplan, config, numVoxels, 1, 0);
}
bool CheckRender::compareBin2BinFloat(const char *src_file, const char *ref_file, unsigned int nelements, const float epsilon, const float threshold)
{
float *src_buffer, *ref_buffer;
FILE *src_fp = NULL, *ref_fp = NULL;
size_t fsize = 0;
unsigned long error_count = 0;
if ((src_fp = fopen(src_file, "rb")) == NULL) {
printf("compareBin2Bin <float> unable to open src_file: %s\n", src_file);
error_count = 1;
}
char *ref_file_path = cutFindFilePath(ref_file, m_ExecPath);
if (ref_file_path == NULL) {
printf("compareBin2Bin <float> unable to find <%s> in <%s>\n", ref_file, m_ExecPath);
printf(">>> Check info.xml and [project//data] folder <%s> <<<\n", m_ExecPath);
printf("Aborting comparison!\n");
printf(" FAILED\n");
error_count++;
if (src_fp) fclose(src_fp);
if (ref_fp) fclose(ref_fp);
}
else
{
if ((ref_fp = fopen(ref_file_path, "rb")) == NULL) {
printf("compareBin2Bin <float> unable to open ref_file: %s\n", ref_file_path);
error_count = 1;
}
if (src_fp && ref_fp) {
src_buffer = (float *)malloc(nelements*sizeof(float));
ref_buffer = (float *)malloc(nelements*sizeof(float));
fsize = fread(src_buffer, nelements, sizeof(float), src_fp);
fsize = fread(ref_buffer, nelements, sizeof(float), ref_fp);
printf("> compareBin2Bin <float> nelements=%d, epsilon=%4.2f, threshold=%4.2f\n", nelements, epsilon, threshold);
printf(" src_file <%s>\n", src_file);
printf(" ref_file <%s>\n", ref_file_path);
if (!cutComparefet( ref_buffer, src_buffer, nelements, epsilon, threshold)) {
error_count++;
}
fclose(src_fp);
fclose(ref_fp);
free(src_buffer);
free(ref_buffer);
} else {
if (src_fp) fclose(src_fp);
if (ref_fp) fclose(ref_fp);
}
}
if (error_count == 0) {
printf(" OK\n");
} else {
printf(" FAILURE: %d errors...\n", (unsigned int)error_count);
}
return (error_count == 0); // returns true if all pixels pass
}
int main(int argc, char **argv)
{
shrQAStart(argc, argv);
if (argc > 1) {
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_bOpenGLQA = true;
g_bFBODisplay = false;
fpsLimit = frameCheckNumber;
}
if (cutCheckCmdLineFlag(argc, (const char **)argv, "fbo")) {
g_bFBODisplay = true;
fpsLimit = frameCheckNumber;
}
}
if (g_bQAReadback) {
runAutoTest(argc, argv);
} else {
printf("[%s] ", sSDKsample);
if (g_bFBODisplay) printf("[FBO Display] ");
if (g_bOpenGLQA) printf("[OpenGL Readback Comparisons] ");
printf("\n");
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
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 load the image, so we know what the size of the image (imageW and imageH)
printf("Allocating host and CUDA memory and loading image file...\n");
const char *image_path = cutFindFilePath("portrait_noise.bmp", argv[0]);
if (image_path == NULL) {
printf( "imageDenoisingGL was unable to find and load image file <portrait_noise.bmp>.\nExiting...\n");
shrQAFinishExit(argc, (const char **)argv, QA_FAILED);
}
LoadBMPFile(&h_Src, &imageW, &imageH, image_path);
printf("Data init done.\n");
// 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 );
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
cutilSafeCall( CUDA_MallocArray(&h_Src, imageW, imageH) );
initOpenGLBuffers();
// Creating the Auto-Validation Code
if (g_bOpenGLQA) {
if (g_bFBODisplay) {
g_CheckRender = new CheckFBO(imageW, imageH, 4);
} else {
g_CheckRender = new CheckBackBuffer(imageW, imageH, 4);
}
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(g_bOpenGLQA);
}
}
printf("Starting GLUT main loop...\n");
printf("Press [1] to view noisy image\n");
printf("Press [2] to view image restored with knn filter\n");
printf("Press [3] to view image restored with nlm filter\n");
printf("Press [4] to view image restored with modified nlm filter\n");
printf("Press [ ] to view smooth/edgy areas [RED/BLUE] Ct's\n");
printf("Press [f] to print frame rate\n");
printf("Press [?] to print Noise and Lerp Ct's\n");
printf("Press [q] to exit\n");
glutDisplayFunc(displayFunc);
glutKeyboardFunc(shutDown);
cutilCheckError( cutCreateTimer(&hTimer) );
cutilCheckError( cutStartTimer(hTimer) );
glutTimerFunc(REFRESH_DELAY, timerEvent,0);
glutMainLoop();
cutilDeviceReset();
shrQAFinishExit(argc, (const char **)argv, QA_PASSED);
}
////////////////////////////////////////////////////////////////////////////////
// Program main
////////////////////////////////////////////////////////////////////////////////
int
main( int argc, char** argv)
{
// use command-line specified CUDA device, otherwise use device with highest Gflops/s
if (!cutCheckCmdLineFlag(argc, (const char **)argv, "noqatest") ||
cutCheckCmdLineFlag(argc, (const char **)argv, "noprompt"))
{
g_bQAReadback = true;
fpsLimit = frameCheckNumber;
}
if (argc > 1) {
if (cutCheckCmdLineFlag(argc, (const char **)argv, "glverify")) {
g_bOpenGLQA = true;
fpsLimit = frameCheckNumber;
}
}
printf("[%s] ", sSDKsample);
if (g_bQAReadback) printf("(Automated Testing)\n");
if (g_bOpenGLQA) printf("(OpenGL Readback)\n");
// Get the path of the filename
char *filename;
if (cutGetCmdLineArgumentstr(argc, (const char**) argv, "image", &filename)) {
image_filename = filename;
}
// load image
char* image_path = cutFindFilePath(image_filename, argv[0]);
if (image_path == 0) {
fprintf(stderr, "Error finding image file '%s'\n", image_filename);
cudaThreadExit();
exit(EXIT_FAILURE);
}
cutilCheckError( cutLoadPPM4ub(image_path, (unsigned char **) &h_img, &width, &height));
if (!h_img) {
printf("Error opening file '%s'\n", image_path);
cudaThreadExit();
exit(-1);
}
printf("Loaded '%s', %d x %d pixels\n", image_path, width, height);
cutGetCmdLineArgumenti(argc, (const char**) argv, "threads", &nthreads);
cutGetCmdLineArgumentf(argc, (const char**) argv, "sigma", &sigma);
runBenchmark = cutCheckCmdLineFlag(argc, (const char**) argv, "bench");
int device;
struct cudaDeviceProp prop;
cudaGetDevice( &device );
cudaGetDeviceProperties( &prop, device );
if( !strncmp( "Tesla", prop.name, 5 ) ) {
printf("Tesla card detected, running the test in benchmark mode (no OpenGL display)\n");
// runBenchmark = CUTTrue;
g_bQAReadback = true;
}
// Benchmark or AutoTest mode detected, no OpenGL
if (runBenchmark == CUTTrue || g_bQAReadback) {
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);
if( cutCheckCmdLineFlag( argc, (const char **)argv, "device" ) )
cutilGLDeviceInit( argc, argv );
else
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
}
initCudaBuffers();
if (g_bOpenGLQA) {
g_CheckRender = new CheckBackBuffer(width, height, 4);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
}
if (g_bQAReadback) {
// This is the automated testing path
g_CheckRender = new CheckBackBuffer(width, height, 4, false);
g_CheckRender->setPixelFormat(GL_RGBA);
g_CheckRender->setExecPath(argv[0]);
g_CheckRender->EnableQAReadback(true);
runAutoTest(argc, argv);
cleanup();
cudaThreadExit();
cutilExit(argc, argv);
}
if (runBenchmark) {
benchmark(100);
cleanup();
cudaThreadExit();
exit(0);
}
initGLBuffers();
atexit(cleanup);
glutMainLoop();
cudaThreadExit();
cutilExit(argc, argv);
}