//! Returns true if NVIDIA is the device vendor
bool cl_deviceIsNVIDIA(cl_device_id dev) {
bool retval = false;
char* vendor = cl_getDeviceVendor(dev);
if(strncmp(vendor, "NVIDIA", 6) == 0) {
retval = true;
}
free(vendor);
return retval;
}
//! Returns true if AMD is the device vendor
bool cl_deviceIsAMD(cl_device_id dev) {
bool retval = false;
char* vendor = cl_getDeviceVendor(dev);
if(strncmp(vendor, "Advanced", 8) == 0) {
retval = true;
}
free(vendor);
return retval;
}
/*!
\brief Initialize OpenCl environment on one device
Init function for one device. Looks for supported devices and creates a context
\return returns a context initialized
*/
cl_context cl_init(char devicePreference)
{
cl_int status;
// Allocate the event table
events = new EventList();
// Discover and populate the platforms
status = clGetPlatformIDs(0, NULL, &numPlatforms);
cl_errChk(status, "Getting platform IDs", true);
if (numPlatforms > 0)
{
// Get all the platforms
platforms = (cl_platform_id*)alloc(numPlatforms *
sizeof(cl_platform_id));
status = clGetPlatformIDs(numPlatforms, platforms, NULL);
cl_errChk(status, "Getting platform IDs", true);
}
else
{
// If no platforms are available, we shouldn't continue
printf("No OpenCL platforms found\n");
exit(-1);
}
// Allocate space for the device lists and lengths
numDevices = (cl_uint*)alloc(sizeof(cl_uint)*numPlatforms);
devices = (cl_device_id**)alloc(sizeof(cl_device_id*)*numPlatforms);
// If a device preference was supplied, we'll limit the search of devices
// based on type
cl_device_type deviceType = CL_DEVICE_TYPE_ALL;
if(devicePreference == 'c') {
deviceType = CL_DEVICE_TYPE_CPU;
}
if(devicePreference == 'g') {
deviceType = CL_DEVICE_TYPE_GPU;
}
// Traverse the platforms array printing information and
// populating devices
for(unsigned int i = 0; i < numPlatforms ; i++)
{
// Print out some basic info about the platform
char* platformName = NULL;
char* platformVendor = NULL;
platformName = cl_getPlatformName(platforms[i]);
platformVendor = cl_getPlatformVendor(platforms[i]);
status = clGetDeviceIDs(platforms[i], deviceType, 0, NULL, &numDevices[i]);
cl_errChk(status, "Getting device IDs", false);
if(status != CL_SUCCESS) {
printf("This is a known NVIDIA bug (if platform == AMD then die)\n");
printf("Setting number of devices to 0 and continuing\n");
numDevices[i] = 0;
}
printf("Platform %d (%d devices):\n", i, numDevices[i]);
printf("\tName: %s\n", platformName);
printf("\tVendor: %s\n", platformVendor);
free(platformName);
free(platformVendor);
// Populate OpenCL devices if any exist
if(numDevices[i] != 0)
{
// Allocate an array of devices of size "numDevices"
devices[i] = (cl_device_id*)alloc(sizeof(cl_device_id)*numDevices[i]);
// Populate Arrray with devices
status = clGetDeviceIDs(platforms[i], deviceType, numDevices[i],
devices[i], NULL);
cl_errChk(status, "Getting device IDs", true);
}
// Print some information about each device
for( unsigned int j = 0; j < numDevices[i]; j++)
{
char* deviceName = NULL;
char* deviceVendor = NULL;
printf("\tDevice %d:\n", j);
deviceName = cl_getDeviceName(devices[i][j]);
deviceVendor = cl_getDeviceVendor(devices[i][j]);
printf("\t\tName: %s\n", deviceName);
printf("\t\tVendor: %s\n", deviceVendor);
free(deviceName);
free(deviceVendor);
}
}
// Hard-code in the platform/device to use, or uncomment 'scanf'
// to decide at runtime
cl_uint chosen_platform, chosen_device;
// UNCOMMENT the following two lines to manually select device each time
//printf("Enter Platform and Device No (Seperated by Space) \n");
//scanf("%d %d", &chosen_platform, &chosen_device);
chosen_platform = 0;
chosen_device = 0;
printf("Using Platform %d, Device %d \n", chosen_platform, chosen_device);
// Do a sanity check of platform/device selection
if(chosen_platform >= numPlatforms ||
chosen_device >= numDevices[chosen_platform]) {
printf("Invalid platform/device combination\n");
exit(-1);
}
// Set the selected platform and device
platform = platforms[chosen_platform];
device = devices[chosen_platform][chosen_device];
// Create the context
cl_context_properties cps[3] = {CL_CONTEXT_PLATFORM,
(cl_context_properties)(platform), 0};
context = clCreateContext(cps, 1, &device, NULL, NULL, &status);
cl_errChk(status, "Creating context", true);
// Create the command queue
commandQueueProf = clCreateCommandQueue(context, device,
CL_QUEUE_PROFILING_ENABLE, &status);
cl_errChk(status, "creating command queue", true);
commandQueueNoProf = clCreateCommandQueue(context, device, 0, &status);
cl_errChk(status, "creating command queue", true);
if(eventsEnabled) {
printf("Profiling enabled\n");
commandQueue = commandQueueProf;
}
else {
printf("Profiling disabled\n");
commandQueue = commandQueueNoProf;
}
return context;
}
//! Dump a CSV file with event information
void EventList::dumpTraceCSV(char* path)
{
char* fullpath = NULL;
FILE* fp = NULL;
// Construct a filename based on the current time
char* filename = this->createFilenameWithTimestamp();
fullpath = smartStrcat(path, filename);
// Try to open the file for writing
fp = fopen(fullpath, "w");
if(fp == NULL) {
printf("Error opening %s\n", fullpath);
exit(-1);
}
// Write some information out about the environment
char* tmp;
char* tmp2;
// Write the device name
tmp = cl_getDeviceName();
if(isUsingImages()) {
tmp2 = smartStrcat(tmp, " (images)");
}
else {
tmp2 = smartStrcat(tmp, " (buffers)");
}
fprintf(fp, "Info;\t%s\n", tmp2);
free(tmp);
free(tmp2);
// Write the vendor name
tmp = cl_getDeviceVendor();
fprintf(fp, "Info;\t%s\n", tmp);
free(tmp);
// Write the driver version
tmp = cl_getDeviceDriverVersion();
fprintf(fp, "Info;\tDriver version %s\n", tmp);
free(tmp);
// Write the device version
tmp = cl_getDeviceVersion();
fprintf(fp, "Info;\t%s\n", tmp);
free(tmp);
// Write the hostname
#ifdef _WIN32
WSADATA wsaData;
WSAStartup(MAKEWORD(2,2), &wsaData);
#endif
char hostname[50];
if(gethostname(hostname, 50) != 0) {
printf("Error getting hostname\n");
}
else {
fprintf(fp, "Info;\tHost %s\n", hostname);
}
int kernelEventSize = this->kernel_events.size();
cl_ulong timer;
cl_int status;
for(int i = 0; i < kernelEventSize; i++)
{
fprintf(fp, "Kernel\t%s", this->kernel_events[i].second);
// cl_computeExecTime(this->kernel_events[i].first));
status = clGetEventProfilingInfo(this->kernel_events[i].first,
CL_PROFILING_COMMAND_QUEUED, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->kernel_events[i].first,
CL_PROFILING_COMMAND_SUBMIT, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->kernel_events[i].first,
CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->kernel_events[i].first,
CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu\n", timer);
}
int ioEventSize = this->io_events.size();
for(int i = 0; i < ioEventSize; i++)
{
fprintf(fp, "IO\t%s", this->io_events[i].second);
//,cl_computeExecTime(this->io_events[i].first));
status = clGetEventProfilingInfo(this->io_events[i].first,
CL_PROFILING_COMMAND_QUEUED, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->io_events[i].first,
CL_PROFILING_COMMAND_SUBMIT, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->io_events[i].first,
CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu", timer);
status = clGetEventProfilingInfo(this->io_events[i].first,
CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &timer, NULL);
cl_errChk(status, "profiling", true);
fprintf(fp, "\t%lu\n", timer);
}
/*
int compileEventSize = this->compile_events.size();
for(int i = 0; i < compileEventSize; i++)
{
fprintf(fp, "Compile;\t%s;\t%3.3f\n", this->compile_events[i].second,
compile_events[i].first);
}
int userEventSize = this->user_events.size();
for(int i = 0; i < userEventSize; i++) {
fprintf(fp, "User;\t%s;\t%3.3f\n", this->user_events[i].second,
user_events[i].first);
}
*/
fclose(fp);
free(filename);
free(fullpath);
}
