std::vector<DeviceEntry> PlatformList::getDevices (cl_device_type type) const {
std::vector<DeviceEntry> result;
for (size_t i = 0; i < platforms ().size (); i++) {
std::vector<DeviceEntry> d = platforms ()[i]->getDevices (type);
result.insert (result.end (), d.begin (), d.end ());
}
return result;
}
//---------------------------------------------------------------------------
int
OpenCLData::initDeviceList()
{
cl_uint numPlatforms;
cl_uint numDevices;
cl_int status;
char sTemp[128];
// Get platform count
status = clGetPlatformIDs(0, NULL, &numPlatforms);
CL_STATUS(status, "clGetPlatformIDs");
if(numPlatforms <= 0)
return FAILURE;
// Get platform IDs
std::unique_ptr<cl_platform_id[]> platforms(new cl_platform_id[numPlatforms]);
status = clGetPlatformIDs(numPlatforms, platforms.get(), NULL);
CL_STATUS(status, "clGetPlatformIDs");
for(cl_uint i = 0; i < numPlatforms; i++)
{
// Get device count
status = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, 0, NULL, &numDevices);
CL_STATUS(status, "clGetDeviceIDs");
if(numDevices <= 0)
continue;
// Get device IDs
std::unique_ptr<cl_device_id[]> devices(new cl_device_id[numDevices]);
status = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, numDevices, devices.get(), NULL);
CL_STATUS(status, "clGetDeviceIDs");
for(cl_uint j = 0; j < numDevices; j++)
{
SDeviceInfo info;
status = clGetDeviceInfo(devices[j], CL_DEVICE_TYPE, sizeof(cl_device_type), &info.deviceType, NULL);
status |= clGetDeviceInfo(devices[j], CL_DEVICE_VENDOR, 128, sTemp, NULL); info.vendor = QString(sTemp).trimmed();;
status |= clGetDeviceInfo(devices[j], CL_DEVICE_NAME, 128, sTemp, NULL); info.name = QString(sTemp).trimmed();
status |= clGetDeviceInfo(devices[j], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(cl_uint), &info.computeUnits, NULL);
status |= clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), (void*)&info.maxWorkGroupSize, NULL);
status |= clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(cl_uint), (void*)&info.maxDims, NULL);
status |= clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(size_t)*3, (void*)info.maxWorkItemSizes, NULL);
status |= clGetDeviceInfo(devices[j], CL_DEVICE_MAX_CLOCK_FREQUENCY, sizeof(cl_uint), &info.maxClockFreq, NULL);
CL_STATUS(status, "clGetDeviceInfo");
m_deviceInfo.append(info);
m_deviceList.append(info.name);
}
}
return SUCCESS;
}
error*
get_platforms(clobj_t **_platforms, uint32_t *num_platforms)
{
return c_handle_error([&] {
*num_platforms = 0;
pyopencl_call_guarded(clGetPlatformIDs, 0, nullptr,
buf_arg(*num_platforms));
pyopencl_buf<cl_platform_id> platforms(*num_platforms);
pyopencl_call_guarded(clGetPlatformIDs, platforms,
buf_arg(*num_platforms));
*_platforms = buf_to_base<platform>(platforms).release();
});
}
/// Main function
int main()
{
clxx::platforms platforms(clxx::get_platforms());
clxx::context context(clxx::make_context_properties(platforms[0]),
clxx::device_type_t::all);
float array[ARRAY_SIZE];
clxx::mem buffer(context, clxx::mem_flags_t::read_write, sizeof(array), array);
std::cout << "buffer.get_type(): " << buffer.get_type() << std::endl;
std::cout << "buffer.get_flags(): " << buffer.get_flags() << std::endl;
std::cout << "buffer.get_size(): " << buffer.get_size() << std::endl;
std::cout << "buffer.get_map_count(): " << buffer.get_map_count() << std::endl;
std::cout << "buffer.get_reference_count(): " << buffer.get_reference_count() << std::endl;
return 0;
}
SEXP getPlatformIDs(){
//returns a list of platform ids
cl_uint num_platforms = 0;
clGetPlatformIDs(0, 0, &num_platforms);
std::vector<cl_platform_id> platforms(num_platforms);
clGetPlatformIDs(num_platforms, platforms.empty() ? NULL : &platforms.front(), &num_platforms);
//for each platform in platforms add its pointer to the return list
Rcpp::List result(platforms.size());
for (int i=0; i<platforms.size(); i++){
cl_platform_id *tempPlatformID = new cl_platform_id;
*tempPlatformID = platforms[i];
Rcpp::XPtr<cl_platform_id> tempXPtr(tempPlatformID);
result[i] = tempXPtr;
}
return result;
}
void Platform::queryTotalPlatforms()
{
cl_int errorCode;
cl_uint numberOfPlatforms;
errorCode = clGetPlatformIDs( 0, 0, &numberOfPlatforms );
errorCheck( errorCode );
std::cout << "Found " << numberOfPlatforms << " platform(s)" << std::endl;
std::vector<cl_platform_id> platforms(numberOfPlatforms);
errorCheck( clGetPlatformIDs( numberOfPlatforms, platforms.data(), 0 ) );
for( unsigned int i = 0; i < platforms.size(); ++i )
{
std::cout << "Platform " << i << std::endl;
getPlatformInfo( platforms.at(0) );
std::cout << std::endl;
}
}
//------------------------------------------------------------------------------
void print_platforms() {
cl_uint numPlatforms = 0;
cl_platform_id platform = 0;
cl_int status = clGetPlatformIDs(0, 0, &numPlatforms);
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformIDs()" << std::endl;
exit(EXIT_FAILURE);
}
if(numPlatforms < 1) {
std::cout << "No OpenCL platform detected" << std::endl;
exit(EXIT_SUCCESS);
}
typedef std::vector< cl_platform_id > PlatformIds;
PlatformIds platforms(numPlatforms);
status = clGetPlatformIDs(platforms.size(), &platforms[0], 0);
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformIDs()" << std::endl;
exit(EXIT_FAILURE);
}
std::vector< char > buf(0x10000, char(0));
int p = 0;
std::cout << "\n***************************************************\n";
std::cout << "Number of platforms: " << platforms.size() << std::endl;
for(PlatformIds::const_iterator i = platforms.begin();
i != platforms.end(); ++i, ++p) {
std::cout << "\n-----------\n";
std::cout << "Platform " << p << std::endl;
std::cout << "-----------\n";
status = ::clGetPlatformInfo(*i, CL_PLATFORM_VENDOR,
buf.size(), &buf[ 0 ], 0 );
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformInfo(): " << std::endl;
exit(EXIT_FAILURE);
}
std::cout << "Vendor: " << &buf[ 0 ] << '\n';
status = ::clGetPlatformInfo(*i, CL_PLATFORM_PROFILE,
buf.size(), &buf[ 0 ], 0 );
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformInfo(): " << std::endl;
exit(EXIT_FAILURE);
}
std::cout << "Profile: " << &buf[ 0 ] << '\n';
status = ::clGetPlatformInfo(*i, CL_PLATFORM_VERSION,
buf.size(), &buf[ 0 ], 0 );
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformInfo(): " << std::endl;
exit(EXIT_FAILURE);
}
std::cout << "Version: " << &buf[ 0 ] << '\n';
status = ::clGetPlatformInfo(*i, CL_PLATFORM_NAME,
buf.size(), &buf[ 0 ], 0 );
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformInfo(): " << std::endl;
exit(EXIT_FAILURE);
}
std::cout << "Name: " << &buf[ 0 ] << '\n';
status = ::clGetPlatformInfo(*i, CL_PLATFORM_EXTENSIONS,
buf.size(), &buf[ 0 ], 0 );
if(status != CL_SUCCESS) {
std::cerr << "ERROR - clGetPlatformInfo(): " << std::endl;
exit(EXIT_FAILURE);
}
std::cout << "Extensions: " << &buf[ 0 ] << '\n';
print_devices(*i);
std::cout << "\n===================================================\n";
}
}
Context& initializeContextFromVA(VADisplay display)
{
(void)display;
#if !defined(HAVE_VAAPI)
NO_VAAPI_SUPPORT_ERROR;
#elif !defined(HAVE_OPENCL)
NO_OPENCL_SUPPORT_ERROR;
#else
contextInitialized = false;
cl_uint numPlatforms;
cl_int status = clGetPlatformIDs(0, NULL, &numPlatforms);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLInitError, "OpenCL: Can't get number of platforms");
if (numPlatforms == 0)
CV_Error(cv::Error::OpenCLInitError, "OpenCL: No available platforms");
std::vector<cl_platform_id> platforms(numPlatforms);
status = clGetPlatformIDs(numPlatforms, &platforms[0], NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLInitError, "OpenCL: Can't get platform Id list");
// For CL-VA interop, we must find platform/device with "cl_intel_va_api_media_sharing" extension.
// With standard initialization procedure, we should examine platform extension string for that.
// But in practice, the platform ext string doesn't contain it, while device ext string does.
// Follow Intel procedure (see tutorial), we should obtain device IDs by extension call.
// Note that we must obtain function pointers using specific platform ID, and can't provide pointers in advance.
// So, we iterate and select the first platform, for which we got non-NULL pointers, device, and CL context.
int found = -1;
cl_context context = 0;
cl_device_id device = 0;
for (int i = 0; i < (int)numPlatforms; ++i)
{
// Get extension function pointers
clGetDeviceIDsFromVA_APIMediaAdapterINTEL = (clGetDeviceIDsFromVA_APIMediaAdapterINTEL_fn)
clGetExtensionFunctionAddressForPlatform(platforms[i], "clGetDeviceIDsFromVA_APIMediaAdapterINTEL");
clCreateFromVA_APIMediaSurfaceINTEL = (clCreateFromVA_APIMediaSurfaceINTEL_fn)
clGetExtensionFunctionAddressForPlatform(platforms[i], "clCreateFromVA_APIMediaSurfaceINTEL");
clEnqueueAcquireVA_APIMediaSurfacesINTEL = (clEnqueueAcquireVA_APIMediaSurfacesINTEL_fn)
clGetExtensionFunctionAddressForPlatform(platforms[i], "clEnqueueAcquireVA_APIMediaSurfacesINTEL");
clEnqueueReleaseVA_APIMediaSurfacesINTEL = (clEnqueueReleaseVA_APIMediaSurfacesINTEL_fn)
clGetExtensionFunctionAddressForPlatform(platforms[i], "clEnqueueReleaseVA_APIMediaSurfacesINTEL");
if (((void*)clGetDeviceIDsFromVA_APIMediaAdapterINTEL == NULL) ||
((void*)clCreateFromVA_APIMediaSurfaceINTEL == NULL) ||
((void*)clEnqueueAcquireVA_APIMediaSurfacesINTEL == NULL) ||
((void*)clEnqueueReleaseVA_APIMediaSurfacesINTEL == NULL))
{
continue;
}
// Query device list
cl_uint numDevices = 0;
status = clGetDeviceIDsFromVA_APIMediaAdapterINTEL(platforms[i], CL_VA_API_DISPLAY_INTEL, display,
CL_PREFERRED_DEVICES_FOR_VA_API_INTEL, 0, NULL, &numDevices);
if ((status != CL_SUCCESS) || !(numDevices > 0))
continue;
numDevices = 1; // initializeContextFromHandle() expects only 1 device
status = clGetDeviceIDsFromVA_APIMediaAdapterINTEL(platforms[i], CL_VA_API_DISPLAY_INTEL, display,
CL_PREFERRED_DEVICES_FOR_VA_API_INTEL, numDevices, &device, NULL);
if (status != CL_SUCCESS)
continue;
// Creating CL-VA media sharing OpenCL context
cl_context_properties props[] = {
CL_CONTEXT_VA_API_DISPLAY_INTEL, (cl_context_properties) display,
CL_CONTEXT_INTEROP_USER_SYNC, CL_FALSE, // no explicit sync required
0
};
context = clCreateContext(props, numDevices, &device, NULL, NULL, &status);
if (status != CL_SUCCESS)
{
clReleaseDevice(device);
}
else
{
found = i;
break;
}
}
if (found < 0)
CV_Error(cv::Error::OpenCLInitError, "OpenCL: Can't create context for VA-API interop");
Context& ctx = Context::getDefault(false);
initializeContextFromHandle(ctx, platforms[found], context, device);
contextInitialized = true;
return ctx;
#endif
}
void opencl_metainfo::initialize() {
// get number of available platforms
auto num_platforms = v1get<cl_uint>(CAF_CLF(clGetPlatformIDs));
// get platform ids
std::vector<cl_platform_id> platforms(num_platforms);
v2callcl(CAF_CLF(clGetPlatformIDs), num_platforms, platforms.data());
if (platforms.empty()) {
throw std::runtime_error("no OpenCL platform found");
}
// support multiple platforms -> "for (auto platform : platforms)"?
auto platform = platforms.front();
// detect how many devices we got
cl_uint num_devs = 0;
cl_device_type dev_type = CL_DEVICE_TYPE_GPU;
// try get some GPU devices and try falling back to CPU devices on error
try {
num_devs = v1get<cl_uint>(CAF_CLF(clGetDeviceIDs), platform, dev_type);
}
catch (std::runtime_error&) {
dev_type = CL_DEVICE_TYPE_CPU;
num_devs = v1get<cl_uint>(CAF_CLF(clGetDeviceIDs), platform, dev_type);
}
// get available devices
std::vector<cl_device_id> ds(num_devs);
v2callcl(CAF_CLF(clGetDeviceIDs), platform, dev_type, num_devs, ds.data());
std::vector<device_ptr> devices(num_devs);
// lift raw pointer as returned by OpenCL to C++ smart pointers
auto lift = [](cl_device_id ptr) { return device_ptr{ptr, false}; };
std::transform(ds.begin(), ds.end(), devices.begin(), lift);
// create a context
context_.reset(v2get(CAF_CLF(clCreateContext), nullptr, num_devs, ds.data(),
pfn_notify, nullptr),
false);
for (auto& device : devices) {
CAF_LOG_DEBUG("creating command queue for device(s)");
command_queue_ptr cmd_queue;
try {
cmd_queue.reset(v2get(CAF_CLF(clCreateCommandQueue),
context_.get(), device.get(),
unsigned{CL_QUEUE_PROFILING_ENABLE}),
false);
}
catch (std::runtime_error&) {
CAF_LOG_DEBUG("unable to create command queue for device");
}
if (cmd_queue) {
auto max_wgs = v3get<size_t>(CAF_CLF(clGetDeviceInfo), device.get(),
unsigned{CL_DEVICE_MAX_WORK_GROUP_SIZE});
auto max_wid = v3get<cl_uint>(CAF_CLF(clGetDeviceInfo), device.get(),
unsigned{CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS});
dim_vec max_wi_per_dim(max_wid);
v2callcl(CAF_CLF(clGetDeviceInfo), device.get(),
unsigned{CL_DEVICE_MAX_WORK_ITEM_SIZES},
sizeof(size_t) * max_wid,
max_wi_per_dim.data());
devices_.push_back(device_info{std::move(device), std::move(cmd_queue),
max_wgs, max_wid, max_wi_per_dim});
}
}
if (devices_.empty()) {
std::string errstr = "could not create a command queue for any device";
CAF_LOG_ERROR(errstr);
throw std::runtime_error(std::move(errstr));
}
}
/*
* Find and return the preferred OpenCL platform
* In case that preferredPlatform is NULL, the ID of the first discovered platform will be returned
*/
cl_platform_id FindOpenCLPlatform(const char* preferredPlatform, cl_device_type deviceType)
{
cl_uint numPlatforms = 0;
cl_int err = CL_SUCCESS;
// Get (in numPlatforms) the number of OpenCL platforms available
// No platform ID will be return, since platforms is NULL
err = clGetPlatformIDs(0, NULL, &numPlatforms);
if (CL_SUCCESS != err)
{
LogError("Error: clGetplatform_ids() to get num platforms returned %s.\n", TranslateOpenCLError(err));
return NULL;
}
LogInfo("Number of available platforms: %u\n", numPlatforms);
if (0 == numPlatforms)
{
LogError("Error: No platforms found!\n");
return NULL;
}
std::vector<cl_platform_id> platforms(numPlatforms);
// Now, obtains a list of numPlatforms OpenCL platforms available
// The list of platforms available will be returned in platforms
err = clGetPlatformIDs(numPlatforms, &platforms[0], NULL);
if (CL_SUCCESS != err)
{
LogError("Error: clGetplatform_ids() to get platforms returned %s.\n", TranslateOpenCLError(err));
return NULL;
}
// Check if one of the available platform matches the preferred requirements
for (cl_uint i = 0; i < numPlatforms; i++)
{
bool match = true;
cl_uint numDevices = 0;
// If the preferredPlatform is not NULL then check if platforms[i] is the required one
// Otherwise, continue the check with platforms[i]
if ((NULL != preferredPlatform) && (strlen(preferredPlatform) > 0))
{
// In case we're looking for a specific platform
match = CheckPreferredPlatformMatch(platforms[i], preferredPlatform);
}
// match is true if the platform's name is the required one or don't care (NULL)
if (match)
{
// Obtains the number of deviceType devices available on platform
// When the function failed we expect numDevices to be zero.
// We ignore the function return value since a non-zero error code
// could happen if this platform doesn't support the specified device type.
err = clGetDeviceIDs(platforms[i], deviceType, 0, NULL, &numDevices);
if (CL_SUCCESS != err)
{
LogError("clGetDeviceIDs() returned %s.\n", TranslateOpenCLError(err));
}
if (0 != numDevices)
{
// There is at list one device that answer the requirements
return platforms[i];
}
}
}
return NULL;
}
int
main()
try
{
sge::opencl::system opencl_system;
fcppt::io::cout()
<< FCPPT_TEXT("Querying the number of available platforms...\n");
sge::opencl::platform::object_sequence &platforms(
opencl_system.platforms());
if(platforms.empty())
{
fcppt::io::cerr()
<< FCPPT_TEXT("Couldn't find any OpenCL platforms on your system.\n");
return EXIT_FAILURE;
}
fcppt::io::cout()
<< FCPPT_TEXT("Number of OpenCL platforms: ") << platforms.size() << FCPPT_TEXT("\n")
<< FCPPT_TEXT("Platform listing begin:\n")
<< FCPPT_TEXT("-----------------------\n");
sge::opencl::platform::object_sequence::size_type platform_index = 0;
for(
sge::opencl::platform::object const ¤t_platform
:
platforms
)
{
fcppt::io::cout()
<< FCPPT_TEXT("\tPlatform ")
<< platform_index++
<< FCPPT_TEXT(":\n")
<< FCPPT_TEXT("\tName: ")
<< fcppt::from_std_string(
current_platform.name()
)
<< FCPPT_TEXT("\n")
<< FCPPT_TEXT("\tVendor: ")
<< fcppt::from_std_string(
current_platform.vendor()
)
<< FCPPT_TEXT("\n")
<< FCPPT_TEXT("Profile type: ")
<<
(current_platform.profile() == sge::opencl::platform::profile_type::full
?
fcppt::string(FCPPT_TEXT("full"))
:
fcppt::string(FCPPT_TEXT("embedded")))
<< FCPPT_TEXT("\n")
<< FCPPT_TEXT("\tVersion: ")
<< current_platform.version().major_part()
<< FCPPT_TEXT(".")
<< current_platform.version().minor_part()
<< FCPPT_TEXT("\n");
if(!current_platform.version().platform_specific().empty())
fcppt::io::cout()
<< FCPPT_TEXT("\tPlatform specific version info: ")
<< fcppt::from_std_string(current_platform.version().platform_specific())
<< FCPPT_TEXT("\n");
fcppt::io::cout()
<< FCPPT_TEXT("\tExtension list begin:\n")
<< FCPPT_TEXT("\t*********************\n")
<< FCPPT_TEXT("\t\t")
<<
fcppt::from_std_string(
boost::algorithm::join(
current_platform.extensions(),
std::string("\n\t\t")))
<< FCPPT_TEXT("\r\t*********************\n")
<< FCPPT_TEXT("-----------------------\n");
}
fcppt::io::cout()
<< FCPPT_TEXT("Platform listing end\n");
sge::opencl::platform::object_sequence::size_type chosen_platform_index;
if(platforms.size() == 1)
{
chosen_platform_index = 0;
}
else
{
fcppt::io::cout()
<< FCPPT_TEXT("Your choice: ");
do
chosen_platform_index =
query_value_from_user<sge::opencl::platform::object_sequence::size_type>(
fcppt::io::cin());
while(chosen_platform_index >= platforms.size());
}
fcppt::io::cout()
<< FCPPT_TEXT("List devices with properties? [y/n] ");
fcppt::char_type list_devices;
do
list_devices =
query_value_from_user<fcppt::char_type>(
fcppt::io::cin());
while(list_devices != FCPPT_TEXT('y') && list_devices != FCPPT_TEXT('n'));
sge::opencl::platform::object &chosen_platform =
platforms[chosen_platform_index];
if(list_devices == FCPPT_TEXT('y'))
{
fcppt::io::cout()
<< FCPPT_TEXT("Number of devices on this platform: ")
<< chosen_platform.devices().size()
<< FCPPT_TEXT('\n')
<< FCPPT_TEXT("Device listing begin:\n")
<< FCPPT_TEXT("-----------------------\n");
for(
sge::opencl::device::object const ¤t_device
:
chosen_platform.devices()
)
{
current_device.output_info(
std::cout);
fcppt::io::cout()
<< FCPPT_TEXT("-----------------------\n");
}
fcppt::io::cout()
<< FCPPT_TEXT("-----------------------\n")
<< FCPPT_TEXT("Device listing end\n");
}
fcppt::io::cout()
<< FCPPT_TEXT("Creating sge::systems object...\n");
sge::window::dim const window_dim{
1024u,
768u
};
sge::systems::instance<
brigand::list<
sge::systems::with_window,
sge::systems::with_renderer<
sge::systems::renderer_caps::core
>
>
> const sys(
sge::systems::make_list
(
// FIXME: Move this higher?
sge::systems::config()
.log_settings(
sge::systems::log_settings{
sge::log::option_container{
sge::log::option{
sge::opencl::log_location(),
fcppt::log::level::verbose
}
}
}
)
)
(
sge::systems::window(
sge::systems::window_source(
sge::systems::original_window(
sge::window::title(
FCPPT_TEXT("Simple OpenCL example")
)
)
)
)
.dont_show()
)
(
sge::systems::renderer(
sge::renderer::pixel_format::object(
sge::renderer::pixel_format::color::depth32,
sge::renderer::pixel_format::depth_stencil::off,
sge::renderer::pixel_format::optional_multi_samples(),
sge::renderer::pixel_format::srgb::no
),
sge::renderer::display_mode::parameters(
sge::renderer::display_mode::vsync::on,
sge::renderer::display_mode::optional_object()
),
sge::viewport::optional_resize_callback{
sge::viewport::center_on_resize(
window_dim
)
}
)
)
);
fcppt::io::cout()
<< FCPPT_TEXT("Done. Creating a context with all devices on this platform...\n");
sge::opencl::device::object_ref_sequence const device_refs(
fcppt::algorithm::map<
sge::opencl::device::object_ref_sequence
>(
chosen_platform.devices(),
[](
sge::opencl::device::object &_device
)
{
return
fcppt::make_ref(
_device
);
}
)
);
sge::opencl::context::object main_context(
sge::opencl::context::parameters(
chosen_platform,
device_refs
)
.share_with(
sys.renderer_device_core())
.error_callback(
sge::opencl::context::error_callback{
&opencl_error_callback
}
)
);
fcppt::io::cout()
<< FCPPT_TEXT("Context created, listing available planar image formats (read/write)\n");
sge::opencl::memory_object::image::format_sequence const planar_image_formats =
main_context.supported_planar_image_formats(
CL_MEM_READ_WRITE);
for(
auto const format
:
planar_image_formats
)
{
sge::opencl::memory_object::image::format_output(
std::cout,
format
);
std::cout << '\n';
}
fcppt::io::cout()
<< FCPPT_TEXT("Listing available volume image formats (read/write)...\n");
sge::opencl::memory_object::image::format_sequence const volume_image_formats =
main_context.supported_volume_image_formats(
CL_MEM_READ_WRITE);
for(
auto const format
:
volume_image_formats
)
{
sge::opencl::memory_object::image::format_output(
std::cout,
format
);
std::cout << '\n';
}
fcppt::io::cout()
<< FCPPT_TEXT("Done, now creating a program...");
sge::opencl::program::object main_program(
sys.log_context(),
main_context,
sge::opencl::program::source_string_sequence{
std::string(
"__kernel void hello_kernel("
"float const multiplier,"
"__global float *input)"
"{"
"int gid = get_global_id(0);"
"int lid = get_local_id(0);"
"input[gid] = lid * multiplier;"
"}"
)
},
sge::opencl::program::optional_build_parameters());
fcppt::io::cout()
<< FCPPT_TEXT("Program created, building the program...\n");
volatile bool build_finished =
false;
main_program.build(
sge::opencl::program::build_parameters()
.notification_callback(
sge::opencl::program::notification_callback{
std::bind(
&program_build_finished,
std::ref(
build_finished
)
)
}
)
);
std::cout << "Waiting for build completion\n";
while(!build_finished)
std::cout << "Build not finished yet\n";
fcppt::io::cout()
<< FCPPT_TEXT("Program built, now creating a kernel...\n");
sge::opencl::kernel::object main_kernel(
main_program,
sge::opencl::kernel::name(
"hello_kernel"));
fcppt::io::cout()
<< FCPPT_TEXT("Kernel created, now creating a vertex buffer...\n");
sge::renderer::vertex::declaration_unique_ptr const vertex_declaration(
sys.renderer_device_core().create_vertex_declaration(
sge::renderer::vertex::declaration_parameters(
sge::renderer::vf::dynamic::make_format<vf::format>())));
sge::renderer::vertex::buffer_unique_ptr const vb(
sys.renderer_device_core().create_vertex_buffer(
sge::renderer::vertex::buffer_parameters(
*vertex_declaration,
sge::renderer::vf::dynamic::make_part_index<
vf::format,
vf::part
>(),
sge::renderer::vertex::count(
6u),
sge::renderer::resource_flags_field{
sge::renderer::resource_flags::readable})));
fcppt::io::cout()
<< FCPPT_TEXT("Done, now creating OpenCL buffer from it\n");
sge::opencl::memory_object::buffer cl_vb(
main_context,
*vb,
sge::opencl::memory_object::renderer_buffer_lock_mode::write_only);
main_kernel.argument(
sge::opencl::kernel::argument_index(
1u),
cl_vb);
main_kernel.argument(
sge::opencl::kernel::argument_index(
0u),
sge::opencl::kernel::numeric_type(
static_cast<cl_float>(
2.0
)
)
);
fcppt::io::cout()
<< FCPPT_TEXT("Done, now creating a command queue\n");
sge::opencl::command_queue::object main_queue(
device_refs[0].get(),
main_context,
sge::opencl::command_queue::execution_mode::out_of_order,
sge::opencl::command_queue::profiling_mode::disabled);
fcppt::io::cout()
<< FCPPT_TEXT("Done, now enqueueing kernel and running it\n");
{
sge::opencl::memory_object::scoped_objects scoped_vb(
main_queue,
sge::opencl::memory_object::base_ref_sequence{
fcppt::reference_to_base<
sge::opencl::memory_object::base
>(
fcppt::make_ref(
cl_vb
)
)
}
);
sge::opencl::command_queue::enqueue_kernel(
main_queue,
main_kernel,
sge::opencl::command_queue::global_dim1(
sge::opencl::dim1(
vb->linear_size()
* 2u
)
),
sge::opencl::command_queue::local_dim1(
sge::opencl::dim1(
2u
)
),
sge::opencl::event::sequence()
);
}
fcppt::io::cout()
<< FCPPT_TEXT("Now locking the vb for reading and printing the values\n");
{
sge::renderer::vertex::scoped_lock const scoped_vb(
*vb,
sge::renderer::lock_mode::readwrite);
typedef
sge::renderer::vf::view<vf::part>
vertex_view;
vertex_view const vertices(
scoped_vb.value());
for(
auto const vertex
:
vertices
)
{
fcppt::io::cout()
<< vertex.get<vf::scalar_quantity>()
<< FCPPT_TEXT('\n');
}
}
fcppt::io::cout()
<< FCPPT_TEXT("Done\n");
}
catch(
fcppt::exception const &_error
)
{
fcppt::io::cerr()
<< _error.string()
<< FCPPT_TEXT('\n');
return EXIT_FAILURE;
}
catch(
std::exception const &_error
)
{
std::cerr
<< _error.what()
<< '\n';
return EXIT_FAILURE;
}
int ocl_t::init()
{
std::cout << "Query available compute devices ...\n";
cl_int err;
cl_uint num;
err = clGetPlatformIDs(0, 0, &num);
if (err != CL_SUCCESS) {
std::cerr << "Unable to get platforms\n";
return 0;
}
std::vector<cl_platform_id> platforms(num);
err = clGetPlatformIDs(num, &platforms[0], &num);
if (err != CL_SUCCESS) {
std::cerr << "Unable to get platform ID\n";
return 0;
}
int device_counter = 0;
for (size_t platform_id = 0; platform_id < num; platform_id++){
size_t dev_c, info_c;
clGetPlatformInfo(platforms[platform_id], CL_PLATFORM_NAME, 0, NULL, &info_c);
std::string platname;
platname.resize(info_c);
clGetPlatformInfo(platforms[platform_id], CL_PLATFORM_NAME, info_c, &platname[0], 0);
std::cout << "Platform :" << platname << "\n";
cl_context_properties prop[] = { CL_CONTEXT_PLATFORM, reinterpret_cast<cl_context_properties>(platforms[platform_id]), 0 };
context = clCreateContextFromType(prop, CL_DEVICE_TYPE_ALL, NULL, NULL, NULL);
if (context == 0) {
std::cerr << "Can't create OpenCL context\n";
return 0;
}
clGetContextInfo(context, CL_CONTEXT_DEVICES, 0, NULL, &dev_c);
std::vector<cl_device_id> devices(dev_c / sizeof(cl_device_id));
clGetContextInfo(context, CL_CONTEXT_DEVICES, dev_c, &devices[0], 0);
for (auto i = devices.begin(); i != devices.end(); i++){
clGetDeviceInfo(*i, CL_DEVICE_NAME, 0, NULL, &info_c);
std::string devname;
devname.resize(info_c);
clGetDeviceInfo(*i, CL_DEVICE_NAME, info_c, &devname[0], 0);
std::cout << "\tDevice " << device_counter++ << ": " << devname.c_str() << "\n";
pdpair_t pd;
pd.device_id = i - devices.begin();
pd.platform_id = platform_id;
ocl_device_list.push_back(pd);
}
clReleaseContext(context);
}
if (list_available_devices) return 0;
cl_context_properties prop[] = { CL_CONTEXT_PLATFORM, reinterpret_cast<cl_context_properties>(platforms[ocl_device_list[opencl_device_id].platform_id]), 0 };
context = clCreateContextFromType(prop, CL_DEVICE_TYPE_ALL, NULL, NULL, NULL);
if (context == 0) {
std::cerr << "Can't create OpenCL context\n";
return 0;
}
size_t dev_c, info_c;
clGetContextInfo(context, CL_CONTEXT_DEVICES, 0, NULL, &dev_c);
std::vector<cl_device_id> devices(dev_c / sizeof(cl_device_id));
clGetContextInfo(context, CL_CONTEXT_DEVICES, dev_c, &devices[0], 0);
device_used = devices[ocl_device_list[opencl_device_id].device_id];
clGetDeviceInfo(device_used, CL_DEVICE_NAME, 0, NULL, &info_c);
std::string devname;
devname.resize(info_c);
clGetDeviceInfo(device_used, CL_DEVICE_NAME, info_c, &devname[0], 0);
std::cout << "Execute on Device " << opencl_device_id << ": " << devname << std::endl;
std::cout << "OK!\n";
queue = clCreateCommandQueue(context, device_used, 0, 0);
if (queue == 0) {
std::cerr << "Can't create command queue\n";
return 0;
}
cl_res = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(cl_float)* iterations, NULL, NULL);
if (cl_res == 0) {
std::cerr << "Can't create OpenCL buffer\n";
return 0;
}
FILE* f = fopen("kernel.c", "rb");
fseek(f, 0, SEEK_END);
size_t tell = ftell(f);
rewind(f);
ocl_src_char = (char*)calloc(tell + 1, 1);
fread(ocl_src_char, tell, 1, f);
initialized = 1;
return 0;
}
std::string PlatformList::list () const {
std::stringstream str;
for (size_t i = 0; i < platforms ().size (); i++)
str << platforms ()[i]->list ();
return str.str ();
}
int Parallel::setup() {
/**
* OpenCL initialization.
*/
cl_int status = Simulator::setup();
CheckStatus(status, "Simulator::setup() failed.");
cl_uint numPlatforms;
status = clGetPlatformIDs(0, NULL, &numPlatforms);
CheckStatus(status, "clGetPlatformIDs, fetching number");
DEBUG_STDOUT("Number of platforms: " << numPlatforms);
cl_platform_id platform = NULL;
if (numPlatforms > 0) {
std::unique_ptr<cl_platform_id[]> platforms (new cl_platform_id[numPlatforms]);
status = clGetPlatformIDs(numPlatforms, platforms.get(), NULL);
CheckStatus(status, "clGetPlatformIDs, fetching platforms");
for (unsigned i = 0; i < numPlatforms; ++i) {
char pbuf[100];
status = clGetPlatformInfo(platforms[i],
CL_PLATFORM_VENDOR,
sizeof(pbuf),
pbuf,
NULL);
CheckStatus(status, "clGetPlatformInfo");
}
// Just grab the first platform.
platform = platforms[0];
}
CheckConditional(platform != NULL, "platform == NULL");
cl_uint numDevices;
status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 0, NULL, &numDevices);
CheckStatus(status, "clGetDeviceIDs: fetching number");
DEBUG_STDOUT("Number of devices: " << numDevices);
cl_device_id *devices = new cl_device_id[numDevices];
status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, numDevices, devices, NULL);
CheckStatus(status, "clGetDeviceIDs: fetching devices");
int deviceIndex = 0;
for (unsigned i = 0; i < numDevices; ++i) {
char pbuf[100];
status = clGetDeviceInfo(devices[i],
CL_DEVICE_NAME,
sizeof(pbuf),
pbuf,
NULL);
if (!strncmp(pbuf, "ATI", 3)) {
deviceIndex = i;
}
}
/* Create the context. */
context = clCreateContext(0, numDevices, devices, NULL, NULL, &status);
CheckConditional(context != NULL, "clCreateContextFromType");
/* Create command queue */
cl_command_queue_properties prop = CL_QUEUE_PROFILING_ENABLE;
commandQueue = clCreateCommandQueue(context, devices[deviceIndex], prop, &status);
CheckStatus(status, "clCreateCommandQueue");
/* Create a CL program using the kernel source */
SDKFile kernelFile;
std::string kernelPath = getenv("HOME") + std::string("/md-simulator/src/TestKernel.cl");
if(!kernelFile.open(kernelPath.c_str())) {
DEBUG_STDERR("Failed to load kernel file : " << kernelPath);
return MD_FAILURE;
}
const char *source = kernelFile.source().c_str();
size_t sourceSize[] = {strlen(source)};
program = clCreateProgramWithSource(context,
1,
&source,
sourceSize,
&status);
CheckStatus(status, "clCreateProgramWithSource");
/* Create a cl program executable for all the devices specified */
status = clBuildProgram(program,
numDevices,
devices,
NULL,
NULL,
NULL);
if (status != CL_SUCCESS) {
if (status == CL_BUILD_PROGRAM_FAILURE) {
cl_int logStatus;
std::unique_ptr<char[]> buildLog (nullptr);
//char *buildLog = NULL;
size_t buildLogSize = 0;
logStatus = clGetProgramBuildInfo(program,
devices[deviceIndex],
CL_PROGRAM_BUILD_LOG,
buildLogSize,
buildLog.get(),
&buildLogSize);
CheckStatus(logStatus, "clGetProgramBuildInfo");
buildLog = std::unique_ptr<char[]>(new char[buildLogSize]);
if(!buildLog) {
return MD_FAILURE;
}
std::fill_n(buildLog.get(), buildLogSize, 0);
logStatus = clGetProgramBuildInfo(program,
devices[deviceIndex],
CL_PROGRAM_BUILD_LOG,
buildLogSize,
buildLog.get(),
NULL);
CheckStatus(logStatus, "clGetProgramBuildInfo (2)");
DEBUG_STDERR("\n\t\t\tBUILD LOG\n");
DEBUG_STDERR("************************************************\n");
DEBUG_STDERR(buildLog.get());
DEBUG_STDERR("************************************************\n");
}
}
CheckStatus(status, "clBuildProgram");
/* Get a kernel object handle for a kernel with the given name */
kernel = clCreateKernel(program, "computeAccelerations", &status);
CheckStatus(status, "clCreateKernel");
/* Check group size against group size returned by kernel */
status = clGetKernelWorkGroupInfo(kernel,
devices[deviceIndex],
CL_KERNEL_WORK_GROUP_SIZE,
sizeof(size_t),
&kernelWorkGroupSize,
0);
CheckStatus(status, "clGetKernelWorkGroupInfo");
DEBUG_STDOUT("kernelWorkGroupSize: " << kernelWorkGroupSize);
/**
* Initialize some simulator data structures.
*/
global = particleCount * particleCount;
local = particleCount;
if (global * local > kernelWorkGroupSize) {
DEBUG_STDERR("WARNING - global * local > kernelWorkGroupSize; global: " << global << ", local: " << local << ", kernelWorkGroupSize: " << kernelWorkGroupSize);
return MD_FAILURE;
}
// Data holds the molecule positions.
data = std::unique_ptr<float[]> (new float[particleCount * 3]);
// Constants holds simulator constants.
constants = std::unique_ptr<float[]> (new float[NUM_CONSTANTS]);
// Copy constants to buffer;
constants[0] = epsilon;
constants[1] = sigma;
constants[2] = negForceCutoffMinusHalf;
constants[3] = forceCutoffMinusHalf;
constants[4] = wallStiffness;
// Results holds pairwise forces.
results = std::unique_ptr<float[]> (new float[particleCount * particleCount * 3]);
return MD_SUCCESS;
}