This stage transfers the input data element from device to host. More...

#include <H2D.h>

Collaboration diagram for risa::cuda::H2D:

Public Types
using	input_type = glados::Image< glados::cuda::HostMemoryManager< unsigned short, glados::cuda::async_copy_policy >>
	The input data type that needs to fit the output type of the previous stage. More...

using	output_type = glados::Image< glados::cuda::DeviceMemoryManager< unsigned short, glados::cuda::async_copy_policy >>
	The output data type that needs to fit the input type of the following stage. More...

using	deviceManagerType = glados::cuda::DeviceMemoryManager< unsigned short, glados::cuda::async_copy_policy >

Public Member Functions
	H2D (const std::string &configFile)
	Initializes everything, that needs to be done only once. More...

	~H2D ()
	Destroys everything that is not destroyed automatically. More...

auto	process (input_type &&sinogram) -> void
	Pushes the sinogram to the processor-threads. More...

auto	wait () -> output_type
	Takes one sinogram from the output queue results_ and transfers it to the neighbored stage. More...

Private Member Functions
auto	processor (int deviceID) -> void
	main data processing routine executed in its own thread for each CUDA device, that performs the data processing of this stage More...

auto	readConfig (const std::string &configFile) -> bool
	Read configuration values from configuration file. More...

Private Attributes
std::map< int, glados::Queue< input_type > >	sinograms_
	one separate input queue for each available CUDA device More...

glados::Queue< output_type >	results_
	the output queue in which the processed sinograms are stored More...

std::map< int, std::thread >	processorThreads_
	stores the processor()-threads More...

std::map< int, cudaStream_t >	streams_
	stores the cudaStreams that are created once More...

std::map< int, unsigned int >	memoryPoolIdxs_
	stores the indeces received when regisitering in MemoryPool More...

double	worstCaseTime_
	stores the worst case time between the arrival of two following sinograms More...

double	bestCaseTime_
	stores the besst case time between the arrival of two following sinograms More...

Timer	tmr_
	used to measure the timings More...

std::size_t	lastIndex_
	stores the index of the last sinogram. Used to analyze which percentage of the arrived sinograms could be reconstructed More...

std::size_t	lostSinos_
	stores the number of sinograms that could not be reconstructed More...

std::size_t	count_ { 0 }
	counts the total number of reconstructed sinograms More...

int	lastDevice_
	stores, to which device the last arrived sinogram was sent More...

int	numberOfDevices_
	the number of available CUDA devices in the system More...

int	numberOfDetectors_
	the number of detectors in the fan beam sinogram More...

int	numberOfProjections_
	the number of projections in the fan beam sinogram More...

int	memPoolSize_
	specifies, how many elements are allocated by memory pool More...

Detailed Description

This stage transfers the input data element from device to host.

This class transfers the data to be processed from host to device. Furthermore, it performs the scheduling between the available devices. Scheduling is done statically, so far. Heterogeneous can be used, by adapting the scheduling in a way, that the more powerful device receives more data inputs.

Definition at line 52 of file H2D.h.

Member Typedef Documentation

using risa::cuda::H2D::deviceManagerType = glados::cuda::DeviceMemoryManager<unsigned short, glados::cuda::async_copy_policy>

Definition at line 58 of file H2D.h.

using risa::cuda::H2D::input_type = glados::Image<glados::cuda::HostMemoryManager<unsigned short, glados::cuda::async_copy_policy>>

The input data type that needs to fit the output type of the previous stage.

Definition at line 55 of file H2D.h.

using risa::cuda::H2D::output_type = glados::Image<glados::cuda::DeviceMemoryManager<unsigned short, glados::cuda::async_copy_policy>>

The output data type that needs to fit the input type of the following stage.

Definition at line 57 of file H2D.h.

Constructor & Destructor Documentation

risa::cuda::H2D::H2D ( const std::string & configFile )

Initializes everything, that needs to be done only once.

Runs as many processor-thread as CUDA devices are available in the system. Allocates memory using the MemoryPool for all CUDA devices.

Parameters

[in] configFile path to configuration file

Definition at line 41 of file H2D.cu.

risa::cuda::H2D::~H2D ( )

Destroys everything that is not destroyed automatically.

Tells MemoryPool to free the allocated memory. Destroys the cudaStreams.

Definition at line 71 of file H2D.cu.

Member Function Documentation

auto risa::cuda::H2D::process ( input_type && sinogram ) -> void

Pushes the sinogram to the processor-threads.

The scheduling for multi-GPU usage is done in this function.

Parameters

[in] sinogram input data that arrived from previous stage

Definition at line 84 of file H2D.cu.

auto risa::cuda::H2D::processor ( int deviceID ) -> void

private

main data processing routine executed in its own thread for each CUDA device, that performs the data processing of this stage

This method takes one sinogram from the input queue sinograms_. The sinogram is transfered to the device using the asynchronous cudaMemcpyAsync()-operation. The resulting device strucutre is pushed back into the output queue results_.

Parameters

[in] deviceID specifies on which CUDA device to execute the device functions

Definition at line 134 of file H2D.cu.

auto risa::cuda::H2D::readConfig ( const std::string & configFile ) -> bool

private

Read configuration values from configuration file.

All values needed for setting up the class are read from the config file in this function.

Parameters

[in] configFile path to config file

Return values

true	configuration options were read successfully
false	configuration options could not be read successfully

Definition at line 171 of file H2D.cu.

auto risa::cuda::H2D::wait ( ) -> output_type

Takes one sinogram from the output queue results_ and transfers it to the neighbored stage.

Returns: the oldest sinogram in the output queue results_

Definition at line 130 of file H2D.cu.

Member Data Documentation

double risa::cuda::H2D::bestCaseTime_

private

stores the besst case time between the arrival of two following sinograms

Definition at line 113 of file H2D.h.

std::size_t risa::cuda::H2D::count_ { 0 }

private

counts the total number of reconstructed sinograms

Definition at line 119 of file H2D.h.

int risa::cuda::H2D::lastDevice_

private

stores, to which device the last arrived sinogram was sent

Definition at line 121 of file H2D.h.

std::size_t risa::cuda::H2D::lastIndex_

private

stores the index of the last sinogram. Used to analyze which percentage of the arrived sinograms could be reconstructed

Definition at line 116 of file H2D.h.

std::size_t risa::cuda::H2D::lostSinos_

private

stores the number of sinograms that could not be reconstructed

Definition at line 117 of file H2D.h.

std::map<int, unsigned int> risa::cuda::H2D::memoryPoolIdxs_

private

stores the indeces received when regisitering in MemoryPool

Definition at line 100 of file H2D.h.

int risa::cuda::H2D::memPoolSize_

private

specifies, how many elements are allocated by memory pool

Definition at line 128 of file H2D.h.

int risa::cuda::H2D::numberOfDetectors_

private

the number of detectors in the fan beam sinogram

Definition at line 125 of file H2D.h.

int risa::cuda::H2D::numberOfDevices_

private

the number of available CUDA devices in the system

Definition at line 123 of file H2D.h.

int risa::cuda::H2D::numberOfProjections_

private

the number of projections in the fan beam sinogram

Definition at line 126 of file H2D.h.

std::map<int, std::thread> risa::cuda::H2D::processorThreads_

private

stores the processor()-threads

Definition at line 98 of file H2D.h.

glados::Queue<output_type> risa::cuda::H2D::results_

private

the output queue in which the processed sinograms are stored

Definition at line 96 of file H2D.h.

std::map<int, glados::Queue<input_type> > risa::cuda::H2D::sinograms_

private

one separate input queue for each available CUDA device

Definition at line 95 of file H2D.h.

std::map<int, cudaStream_t> risa::cuda::H2D::streams_

private

stores the cudaStreams that are created once

Definition at line 99 of file H2D.h.

Timer risa::cuda::H2D::tmr_

private

used to measure the timings

Definition at line 114 of file H2D.h.

double risa::cuda::H2D::worstCaseTime_

private

stores the worst case time between the arrival of two following sinograms

Definition at line 112 of file H2D.h.

The documentation for this class was generated from the following files:

/home/frust45/RISA/RISA/risaLib/include/risa/Copy/H2D.h
/home/frust45/RISA/RISA/risaLib/src/Copy/H2D.cu

Public Types

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation