std::pair<std::vector<cl::Context>, std::vector<command_queue>>
queue_list(DevFilter &&filter, cl_command_queue_properties properties = 0) {
std::vector<cl::Context> context;
std::vector<command_queue> queue;
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
for(auto p = platforms.begin(); p != platforms.end(); p++) {
std::vector<cl::Device> device;
std::vector<cl::Device> dev_list;
p->getDevices(CL_DEVICE_TYPE_ALL, &dev_list);
for(auto d = dev_list.begin(); d != dev_list.end(); d++) {
if (!d->getInfo<CL_DEVICE_AVAILABLE>()) continue;
if (!filter(*d)) continue;
device.push_back(*d);
}
if (device.empty()) continue;
for(auto d = device.begin(); d != device.end(); d++)
try {
context.push_back(cl::Context(std::vector<cl::Device>(1, *d)));
queue.push_back(command_queue(context.back(), *d, properties));
} catch(const cl::Error&) {
// Something bad happened. Better skip this device.
}
}
return std::make_pair(context, queue);
}
void OpenCLInterface::printOCLDeviceInfo(const cl::Device &device) const {
std::string device_name;
std::string device_vendor;
cl_int error;
error = device.getInfo(CL_DEVICE_NAME, &device_name);
if (error != CL_SUCCESS) {
Logger::writeLine("OpenCLInterface::printOCLDeviceInfo(): Error during query for CL_DEVICE_NAME: " + std::to_string(error));
}
error = device.getInfo(CL_DEVICE_VENDOR, &device_vendor);
if (error != CL_SUCCESS) {
Logger::writeLine("OpenCLInterface::printOCLDeviceInfo(): Error during query for CL_DEVICE_VENDOR: " + std::to_string(error));
}
Logger::writeLine("OpenCLInterface::printOCLDeviceInfo(): OpenCL device info: " + device_name + ", vendor " + device_vendor + ".");
}
std::string Convolution3DCLBuffer::getDeviceInfo(cl::Device device, cl_device_info info)
{
std::string result;
status = device.getInfo(info,&result);
CHECK_ERROR(status, "cl::Device::getInfo");
return result;
}
std::string Convolution3DCLBuffer::getDeviceName(cl::Device device)
{
std::string result;
cl_int status = device.getInfo(CL_DEVICE_NAME,&result);
CHECK_ERROR(status, "cl::Device::getInfo");
return result;
}
/*
* Get the info from device dev
*/
void CLGL::CLGLGetDevicesInfo(cl::Device dev, cl_device_info name, std::string * info)
{
try{
dev.getInfo(name, info);
}
catch(cl::Error error){
std::cout << error.what() << " " << CLGLError::errToStr(error.err()) << std::endl;
}
return;
}
OGLSharedFramebuffer::OGLSharedFramebuffer(CL::Device& device,
const ivec2& size, int tile_size) :
Framebuffer(device, size, tile_size),
_tex_buffer(_act_size.x * _act_size.y * sizeof(vec4), GL_RGBA32F),
_shared(device.share_gl()),
_local(0)
{
if (_shared) {
_cl_buffer = new CL::Buffer(device, _tex_buffer.get_buffer());
} else {
_cl_buffer = new CL::Buffer(device, _tex_buffer.get_size(), CL_MEM_READ_WRITE);
_local = malloc(_tex_buffer.get_size());
}
}
std::vector<cl::Device> device_list(DevFilter&& filter) {
std::vector<cl::Device> device;
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
for(auto p = platforms.begin(); p != platforms.end(); p++) {
std::vector<cl::Device> dev_list;
p->getDevices(CL_DEVICE_TYPE_ALL, &dev_list);
for(auto d = dev_list.begin(); d != dev_list.end(); d++) {
if (!d->getInfo<CL_DEVICE_AVAILABLE>()) continue;
if (!filter(*d)) continue;
device.push_back(*d);
}
}
return device;
}
cl_ulong getLocalMemSize(const cl::Device &device)
{
cl_ulong localMemSize;
device.getInfo(CL_DEVICE_LOCAL_MEM_SIZE, &localMemSize);
return localMemSize;
}
size_t getMaxWorkGroupSize(const cl::Device &device)
{
size_t maxWorkGroupSize;
device.getInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE, &maxWorkGroupSize);
return maxWorkGroupSize;
}
cl_uint getMaxComputeUnits(const cl::Device &device)
{
cl_uint nComputeUnits;
device.getInfo(CL_DEVICE_MAX_COMPUTE_UNITS, &nComputeUnits);
return nComputeUnits;
}
cl_device_type getType(const cl::Device &device)
{
cl_device_type deviceType;
device.getInfo(CL_DEVICE_TYPE, &deviceType);
return deviceType;
}
namespace opencl {
typedef cl::Context context;
typedef cl::Device device;
typedef cl::CommandQueue command_queue;
typedef cl_command_queue_properties command_queue_properties;
typedef cl_device_id device_id;
typedef cl::NDRange ndrange;
/// Binds the specified context to the calling CPU thread.
/**
* With the OpenCL backend this is an empty stub provided for compatibility
* with the CUDA backend.
*/
inline void select_context(const command_queue&) {
}
/// Returns id of the device associated with the given queue.
inline device_id get_device_id(const command_queue &q) {
return q.getInfo<CL_QUEUE_DEVICE>()();
}
/// \cond INTERNAL
typedef cl_context context_id;
/// Returns raw context id for the given queue.
inline context_id get_context_id(const command_queue &q) {
return q.getInfo<CL_QUEUE_CONTEXT>()();
}
/// Returns context for the given queue.
inline context get_context(const command_queue &q) {
return q.getInfo<CL_QUEUE_CONTEXT>();
}
/// Compares contexts by raw ids.
struct compare_contexts {
bool operator()(const context &a, const context &b) const {
return a() < b();
}
};
/// Compares queues by raw ids.
struct compare_queues {
bool operator()(const command_queue &a, const command_queue &b) const {
return a() < b();
}
};
/// \endcond
/// Create command queue on the same context and device as the given one.
inline command_queue duplicate_queue(const command_queue &q) {
return command_queue(
q.getInfo<CL_QUEUE_CONTEXT>(), q.getInfo<CL_QUEUE_DEVICE>());
}
/// Checks if the compute device is CPU.
inline bool is_cpu(const command_queue &q) {
cl::Device d = q.getInfo<CL_QUEUE_DEVICE>();
#ifdef _MSC_VER
# pragma warning(push)
# pragma warning(disable: 4800)
#endif
return d.getInfo<CL_DEVICE_TYPE>() & CL_DEVICE_TYPE_CPU;
#ifdef _MSC_VER
# pragma warning(pop)
#endif
}
/// Select devices by given criteria.
/**
* \param filter Device filter functor. Functors may be combined with logical
* operators.
* \returns list of devices satisfying the provided filter.
*
* This example selects any GPU which supports double precision arithmetic:
\code
auto devices = device_list(
Filter::Type(CL_DEVICE_TYPE_GPU) && Filter::DoublePrecision
);
\endcode
*/
template<class DevFilter>
std::vector<cl::Device> device_list(DevFilter&& filter) {
std::vector<cl::Device> device;
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
for(auto p = platforms.begin(); p != platforms.end(); p++) {
std::vector<cl::Device> dev_list;
p->getDevices(CL_DEVICE_TYPE_ALL, &dev_list);
for(auto d = dev_list.begin(); d != dev_list.end(); d++) {
if (!d->getInfo<CL_DEVICE_AVAILABLE>()) continue;
if (!filter(*d)) continue;
device.push_back(*d);
}
}
return device;
}
/// Create command queues on devices by given criteria.
/**
* \param filter Device filter functor. Functors may be combined with logical
* operators.
* \param properties Command queue properties.
*
* \returns list of queues accociated with selected devices.
* \see device_list
*/
template<class DevFilter>
std::pair<std::vector<cl::Context>, std::vector<command_queue>>
queue_list(DevFilter &&filter, cl_command_queue_properties properties = 0) {
std::vector<cl::Context> context;
std::vector<command_queue> queue;
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
for(auto p = platforms.begin(); p != platforms.end(); p++) {
std::vector<cl::Device> device;
std::vector<cl::Device> dev_list;
p->getDevices(CL_DEVICE_TYPE_ALL, &dev_list);
for(auto d = dev_list.begin(); d != dev_list.end(); d++) {
if (!d->getInfo<CL_DEVICE_AVAILABLE>()) continue;
if (!filter(*d)) continue;
device.push_back(*d);
}
if (device.empty()) continue;
for(auto d = device.begin(); d != device.end(); d++)
try {
context.push_back(cl::Context(std::vector<cl::Device>(1, *d)));
queue.push_back(command_queue(context.back(), *d, properties));
} catch(const cl::Error&) {
// Something bad happened. Better skip this device.
}
}
return std::make_pair(context, queue);
}
} // namespace opencl
void printDeviceInfo(const cl::Device &device, cl_device_info info) {
if (!initialized) {
printf("not initialized. call initCLUtility().");
}
printf("%s: ", clDeviceInfoMetaInfos[info].name);
switch (clDeviceInfoMetaInfos[info].infoType) {
case CLDeviceInfoType_bool: {
cl_bool val;
device.getInfo(info, &val);
printf(val != 0 ? "YES" : "NO");
break;
}
case CLDeviceInfoType_uint: {
cl_uint val;
device.getInfo(info, &val);
printf("%u", val);
break;
}
case CLDeviceInfoType_ulong: {
cl_ulong val;
device.getInfo(info, &val);
printf("%llu", val);
break;
}
case CLDeviceInfoType_size_t: {
size_t val;
device.getInfo(info, &val);
printf("%lu", val);
break;
}
case CLDeviceInfoType_string: {
std::string val;
device.getInfo(info, &val);
printf("%s", val.c_str());
break;
}
case CLDeviceInfoType_size_t_vec: {
VECTOR_CLASS<size_t> val;
device.getInfo(info, &val);
for (uint32_t i = 0; i < val.size() - 1; ++i) {
printf("%lu, ", val[i]);
}
printf("%lu", val.back());
break;
}
case CLDeviceInfoType_device_id: {
cl_device_id val;
device.getInfo(info, &val);
printf("%#018llx", (uint64_t)val);
break;
}
case CLDeviceInfoType_platform_id: {
cl_platform_id val;
device.getInfo(info, &val);
printf("%#018llx", (uint64_t)val);
break;
}
case CLDeviceInfoType_device_type: {
cl_device_type val;
device.getInfo(info, &val);
switch (val) {
case CL_DEVICE_TYPE_CPU:
printf("CPU");
break;
case CL_DEVICE_TYPE_GPU:
printf("GPU");
break;
case CL_DEVICE_TYPE_ACCELERATOR:
printf("Accelerator");
break;
case CL_DEVICE_TYPE_CUSTOM:
printf("Custom");
break;
default:
break;
}
break;
}
case CLDeviceInfoType_device_fp_config: {
cl_device_fp_config val;
device.getInfo(info, &val);
if ((val & CL_FP_DENORM) != 0)
printf("CL_FP_DENORM ");
if ((val & CL_FP_INF_NAN) != 0)
printf("CL_FP_INF_NAN ");
if ((val & CL_FP_ROUND_TO_NEAREST) != 0)
printf("CL_FP_ROUND_TO_NEAREST ");
if ((val & CL_FP_ROUND_TO_ZERO) != 0)
printf("CL_FP_ROUND_TO_ZERO ");
if ((val & CL_FP_ROUND_TO_INF) != 0)
printf("CL_FP_ROUND_TO_INF ");
if ((val & CL_FP_FMA) != 0)
printf("CL_FP_FMA ");
if ((val & CL_FP_SOFT_FLOAT) != 0)
printf("CL_FP_SOFT_FLOAT ");
if ((val & CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT) != 0)
printf("CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT ");
break;
}
case CLDeviceInfoType_device_local_mem_type: {
cl_device_local_mem_type val;
device.getInfo(info, &val);
switch (val) {
case CL_LOCAL:
printf("Local");
break;
case CL_GLOBAL:
printf("Global");
default:
break;
}
break;
}
case CLDeviceInfoType_device_mem_cache_type: {
cl_device_mem_cache_type val;
device.getInfo(info, &val);
switch (val) {
case CL_NONE:
printf("None");
break;
case CL_READ_ONLY_CACHE:
printf("Read Only Cache");
break;
case CL_READ_WRITE_CACHE:
printf("Read Write Cache");
break;
default:
break;
}
break;
}
case CLDeviceInfoType_device_exec_capabilities: {
cl_device_exec_capabilities val;
device.getInfo(info, &val);
if ((val & CL_EXEC_KERNEL) != 0)
printf("CL_EXEC_KERNEL ");
if ((val & CL_EXEC_NATIVE_KERNEL) != 0)
printf("CL_EXEC_NATIVE_KERNEL ");
break;
}
case CLDeviceInfoType_command_queue_properties: {
cl_command_queue_properties val;
device.getInfo(info, &val);
if ((val & CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE) != 0)
printf("CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE ");
if ((val & CL_QUEUE_PROFILING_ENABLE) != 0)
printf("CL_QUEUE_PROFILING_ENABLE ");
break;
}
case CLDeviceInfoType_device_affinity_domain: {
cl_device_affinity_domain val;
device.getInfo(info, &val);
if ((val & CL_DEVICE_AFFINITY_DOMAIN_NUMA) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_NUMA ");
if ((val & CL_DEVICE_AFFINITY_DOMAIN_L4_CACHE) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_L4_CACHE ");
if ((val & CL_DEVICE_AFFINITY_DOMAIN_L3_CACHE) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_L3_CACHE ");
if ((val & CL_DEVICE_AFFINITY_DOMAIN_L2_CACHE) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_L2_CACHE ");
if ((val & CL_DEVICE_AFFINITY_DOMAIN_L1_CACHE) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_L1_CACHE ");
if ((val & CL_DEVICE_AFFINITY_DOMAIN_NEXT_PARTITIONABLE) != 0)
printf("CL_DEVICE_AFFINITY_DOMAIN_NEXT_PARTITIONABLE ");
break;
}
case CLDeviceInfoType_device_partition_property_vec: {
VECTOR_CLASS<cl_device_partition_property> val;
device.getInfo(info, &val);
for (uint32_t i = 0; i < val.size(); ++i) {
switch (val[i]) {
case CL_DEVICE_PARTITION_EQUALLY:
printf("CL_DEVICE_PARTITION_EQUALLY");
break;
case CL_DEVICE_PARTITION_BY_COUNTS:
printf("CL_DEVICE_PARTITION_BY_COUNTS");
break;
case CL_DEVICE_PARTITION_BY_COUNTS_LIST_END:
printf("CL_DEVICE_PARTITION_BY_COUNTS_LIST_END");
break;
case CL_DEVICE_PARTITION_BY_AFFINITY_DOMAIN:
printf("CL_DEVICE_PARTITION_BY_AFFINITY_DOMAIN");
break;
default:
break;
}
if (i < val.size() - 1)
printf(", ");
}
break;
}
default:
break;
}
printf("\n");
}