Task running inference of a Model. More...

#include <inference_task.h>

Inheritance diagram for Beatmup::NNets::InferenceTask:

Public Member Functions
	InferenceTask (Model &model, ChunkCollection &data)

void	connect (AbstractBitmap &image, AbstractOperation &operation, int inputIndex=0)
	Connects an image to a specific operation input. More...

void	connect (AbstractBitmap &image, const std::string &operation, int inputIndex=0)

Public Member Functions inherited from Beatmup::Object
virtual	~Object ()

Protected Attributes
ChunkCollection &	data

Model &	model

Private Member Functions
void	beforeProcessing (ThreadIndex threadCount, ProcessingTarget target, GraphicPipeline *gpu) override
	Instruction called before the task is executed. More...

void	afterProcessing (ThreadIndex threadCount, GraphicPipeline *gpu, bool aborted) override
	Instruction called after the task is executed. More...

bool	processOnGPU (GraphicPipeline &gpu, TaskThread &thread) override
	Executes the task on GPU. More...

bool	process (TaskThread &thread) override
	Executes the task on CPU within a given thread. More...

ThreadIndex	getMaxThreads () const override
	Gives the upper limint on the number of threads the task may be performed by. More...

Private Member Functions inherited from Beatmup::BitmapContentLock
	BitmapContentLock ()

	~BitmapContentLock ()

void	readLock (GraphicPipeline gpu, AbstractBitmap bitmap, ProcessingTarget target)
	Locks content of a bitmap for reading using a specific processing target device. More...

void	writeLock (GraphicPipeline gpu, AbstractBitmap bitmap, ProcessingTarget target)
	Locks content of a bitmap for writing using a specific processing target device. More...

void	unlock (AbstractBitmap *bitmap)
	Drops a lock to the bitmap. More...

void	unlockAll ()
	Unlocks all the locked bitmaps unconditionally. More...

template<const ProcessingTarget target>
void	lock (GraphicPipeline gpu, AbstractBitmap input, AbstractBitmap *output)

void	lock (GraphicPipeline gpu, ProcessingTarget target, AbstractBitmap input, AbstractBitmap *output)

template<const ProcessingTarget target>
void	lock (GraphicPipeline gpu, std::initializer_list< AbstractBitmap > read, std::initializer_list< AbstractBitmap * > write)

template<typename ... Args>
void	unlock (AbstractBitmap *first, Args ... others)

Private Attributes
std::map< std::pair< AbstractOperation , int >, AbstractBitmap >	inputImages

Additional Inherited Members
Public Types inherited from Beatmup::AbstractTask
enum class	TaskDeviceRequirement { CPU_ONLY , GPU_OR_CPU , GPU_ONLY }
	Specifies which device (CPU and/or GPU) is used to run the task. More...

Static Public Member Functions inherited from Beatmup::AbstractTask
static ThreadIndex	validThreadCount (int number)
	Valid thread count from a given integer value. More...

Detailed Description

Task running inference of a Model.

During the firs run of this task with a given model the shader programs are built and the memory is allocated. The subsequent runs are much faster.

Definition at line 33 of file inference_task.h.

Constructor & Destructor Documentation

◆ InferenceTask()

Beatmup::NNets::InferenceTask::InferenceTask	(	Model &	model,
		ChunkCollection &	data
	)

inline

Definition at line 48 of file inference_task.h.

48 : model(model), data(data) {}

Beatmup::NNets::InferenceTask::data

ChunkCollection & data

Definition: inference_task.h:44

Beatmup::NNets::InferenceTask::model

Model & model

Definition: inference_task.h:45

Member Function Documentation

◆ beforeProcessing()

void InferenceTask::beforeProcessing	(	ThreadIndex	threadCount,
		ProcessingTarget	target,
		GraphicPipeline *	gpu
	)

overrideprivatevirtual

Instruction called before the task is executed.

Parameters

threadCount	Number of threads used to perform the task
target	Device used to perform the task
gpu	A graphic pipeline instance; may be null.

Reimplemented from Beatmup::AbstractTask.

Definition at line 31 of file inference_task.cpp.

                                                                                                            {
     for (auto it : inputImages)
         readLock(gpu, it.second, ProcessingTarget::GPU);
     model.prepare(*gpu, data);
 }

◆ afterProcessing()

void InferenceTask::afterProcessing	(	ThreadIndex	threadCount,
		GraphicPipeline *	gpu,
		bool	aborted
	)

overrideprivatevirtual

Instruction called after the task is executed.

Parameters

threadCount	Number of threads used to perform the task
gpu	GPU to be used to execute the task; may be null.
aborted	`true` if the task was aborted

Reimplemented from Beatmup::AbstractTask.

Definition at line 38 of file inference_task.cpp.

                                                                                                {
     if (gpu)
         gpu->flush();
     unlockAll();
 }

◆ processOnGPU()

bool InferenceTask::processOnGPU	(	GraphicPipeline &	gpu,
		TaskThread &	thread
	)

overrideprivatevirtual

Executes the task on GPU.

Parameters

gpu	graphic pipeline instance
thread	associated task execution context

Returns: true if the execution is finished correctly, false otherwise

Reimplemented from Beatmup::AbstractTask.

Definition at line 45 of file inference_task.cpp.

                                                                          {
     model.execute(thread, &gpu);
     return true;
 }

◆ process()

bool InferenceTask::process ( TaskThread & thread )

overrideprivatevirtual

Executes the task on CPU within a given thread.

Generally called by multiple threads.

Parameters

thread associated task execution context

Returns: true if the execution is finished correctly, false otherwise

Reimplemented from Beatmup::GpuTask.

Definition at line 51 of file inference_task.cpp.

                                               {
     model.execute(thread, nullptr);
     return true;
 }

◆ getMaxThreads()

ThreadIndex Beatmup::NNets::InferenceTask::getMaxThreads ( ) const

inlineoverrideprivatevirtual

Gives the upper limint on the number of threads the task may be performed by.

The actual number of threads running a specific task may be less or equal to the returned value, depending on the number of workers in ThreadPool running the task.

Reimplemented from Beatmup::GpuTask.

Definition at line 41 of file inference_task.h.

41 { return MAX_THREAD_INDEX; }

Beatmup::MAX_THREAD_INDEX

static const ThreadIndex MAX_THREAD_INDEX

maximum possible thread index value

Definition: parallelism.h:71

◆ connect() [1/2]

void InferenceTask::connect	(	AbstractBitmap &	image,
		AbstractOperation &	operation,
		int	inputIndex = `0`
	)

Connects an image to a specific operation input.

Ensures the image content is up-to-date in GPU memory by the time the inference is run.

Parameters

[in]	image	The image
[in]	operation	The operation
[in]	inputIndex	The input index of the operation

Definition at line 25 of file inference_task.cpp.

                                                                                                {
     inputImages[std::make_pair(&operation, inputIndex)] = &image;
     operation.setInput(image, inputIndex);
 }

◆ connect() [2/2]

void Beatmup::NNets::InferenceTask::connect	(	AbstractBitmap &	image,
		const std::string &	operation,
		int	inputIndex = `0`
	)

inline

Definition at line 58 of file inference_task.h.

                                                                                                        {
                 connect(image, model.getOperation(operation), inputIndex);
             }

Member Data Documentation

◆ inputImages

std::map<std::pair<AbstractOperation*, int>, AbstractBitmap*> Beatmup::NNets::InferenceTask::inputImages

private

Definition at line 35 of file inference_task.h.

◆ data

ChunkCollection& Beatmup::NNets::InferenceTask::data

protected

Definition at line 44 of file inference_task.h.

◆ model

Model& Beatmup::NNets::InferenceTask::model

protected

Definition at line 45 of file inference_task.h.

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

core/nnets/inference_task.h
core/nnets/inference_task.cpp

Public Member Functions

Protected Attributes

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ InferenceTask()

Member Function Documentation

◆ beforeProcessing()

◆ afterProcessing()

◆ processOnGPU()

◆ process()

◆ getMaxThreads()

◆ connect() [1/2]

◆ connect() [2/2]

Member Data Documentation

◆ inputImages

◆ data

◆ model