#include <dense.h>

Inheritance diagram for Beatmup::NNets::Dense:

Public Member Functions
	Dense (Context &context, const std::string &name, int numOutputDims, bool useBias)
	Creates a Dense operation. More...

Size	getOutputSize (int outputIndex=0) const
	Returns full size of a specific operation output. More...

int	getInputCount () const
	Returns number of operation inputs. More...

int	getOutputCount () const
	Returns number of operation outputs. More...

bool	acceptsStorageInput (int index=0) const
	Returns `true` if the operation can take a Storage::View at a specific input. More...

bool	acceptsVectorInput (int index=0) const
	Returns `true` if the operation can take a GL::Vector at a specific input. More...

bool	acceptsVectorOutput (int index=0) const
	Returns `true` if the operation can take a GL::Vector at a specific output. More...

void	getOutput (GL::Vector *&, int index=0)
	Returns a GL::Vector bound to a specific operation output. More...

void	setInput (Storage::View &&view, int index=0)

void	setInput (GL::Vector &vector, int index=0)

void	setOutput (GL::Vector &vector, int index=0)

GL::Vector::Format	getOutputVectorFormat () const

std::map< std::string, std::string >	serialize () const
	Returns a serialized representation of th operation;. More...

void	disconnect ()
	Assigns empty inputs and outputs. More...

unsigned long	countMultiplyAdds () const
	Counts (approximate) number of multiply-adds used by this operation. More...

unsigned long	countTexelFetches () const
	Counts (approximate) number of texels fetches. More...

Public Member Functions inherited from Beatmup::NNets::AbstractOperation
virtual	~AbstractOperation ()

virtual bool	usesGpu () const
	Returns `true` if the operation is run on GPU. More...

virtual bool	acceptsTextureInput (int index=0) const
	Returns `true` if the operation can take a GL::TextureHandler at a specific input. More...

virtual bool	acceptsStorageOutput (int index=0) const
	Returns `true` if the operation can take a Storage::View at a specific output. More...

virtual bool	acceptsTextureOutput (int index=0) const
	Returns `true` if the operation can take a GL::TextureHandler at a specific output. More...

virtual Storage::View	getOutput (int index=0)
	Returns a storage view bound to a specific operation output. More...

virtual void	getOutput (GL::TextureHandler *&vector, int index=0)
	Returns a GL::TextureHandler bound to a specific operation output. More...

virtual void	setOutput (Storage::View &&storage, int index=0)

virtual void	setInput (GL::TextureHandler &image, int index=0)

virtual void	setOutput (GL::TextureHandler &image, int index=0)

std::string	getName () const

Static Public Member Functions
static bool	initDeserializer ()
	Sets up deserialization of the operation. More...

Static Public Attributes
static const char *	MATRIX_CHUNK_SUFFIX = "/w"
	suffix added to the op name to get the matrix chunk id in the model data More...

static const char *	BIAS_CHUNK_SUFFIX = "/b"
	suffix added to the op name to get the bias chunk id in the model data More...

Private Member Functions
void	prepare (GraphicPipeline &gpu, ChunkCollection &data, GL::ProgramBank &bank)
	Compiles GLSL shaders. More...

void	execute (TaskThread &thread, GraphicPipeline &gpu)
	Executes the operation. More...

void	getSampledChannels (int index, int &min, int &max) const
	Retrieves range of input features channels sampled at the same time for a specific input. More...

Private Member Functions inherited from Beatmup::GL::LinearMapping
	LinearMapping (Context &context, bool forceFixed16=false)
	Instantiates LinearMapping. More...

	~LinearMapping ()

void	setMatrix (GraphicPipeline &gpu, const int width, const int height, const float *values)

void	setBias (GraphicPipeline &gpu, const int height, const float *values)

void	operator() (GraphicPipeline &gpu, TextureHandler &result, TextureHandler &input)

void	prepare (GraphicPipeline &gpu, TextureHandler &output, TextureHandler &input, ProgramBank *bank=nullptr)
	Prepares the mapping for application (builds its GPU programs). More...

void	process (GraphicPipeline &gpu, TextureHandler &output, TextureHandler &input)

Private Attributes
const int	numOutputDims
	output feature vector size / number of rows in `A` More...

const bool	useBias
	if `true`, the bias vector `b` is searched in the model data to add to the output More...

GL::Vector *	inputVector
	if not null, points the input vector `x`; otherwise the input is a storage view More...

GL::Vector *	outputVector
	pointer to the output vector More...

Storage::View	inputStorage
	if not empty, contains the input features; otherwise the input is a GL vector More...

Additional Inherited Members
Protected Member Functions inherited from Beatmup::NNets::AbstractOperation
	AbstractOperation (const std::string &name)

virtual void	execute (TaskThread &thread)
	Executes the operation within a specific CPU thread. More...

virtual int	getInputPadding (int index=0) const
	Retrieves minimum required size of zero padding for a given input. More...

Detailed Description

Dense (linear) layer.

Computes A*x + b for input feature vector x, a matrix A and an optional bias vector b. Accepts a GL::Vector or a flat Storage view on input, amd only a GL::Vector on output.

Constraints:

Number of input channels must be a multiple of 8.
No activation function is applied on output (not yet implemented).

The matrix and bias coefficients are searched in chunks. The chunk names consist of the operation name followed by Dense::MATRIX_CHUNK_SUFFIX and Dense::BIAS_CHUNK_SUFFIX respectively. The chunk contents is a single precision floating point array. The matrix coefficients are taken in row-major order.

Definition at line 42 of file dense.h.

Constructor & Destructor Documentation

◆ Dense()

Dense::Dense	(	Context &	context,
		const std::string &	name,
		int	numOutputDims,
		bool	useBias
	)

Creates a Dense operation.

Parameters

context	A context instance
name	Operation name
numOutputDims	Number of output dimensions
useBias	If `true`, the bias vector addition is enabled.

Definition at line 30 of file dense.cpp.

                                                                                           :
     AbstractOperation(name),
     LinearMapping(context, false),
     numOutputDims(numOutputDims), useBias(useBias), inputVector(nullptr), outputVector(nullptr)
 {
     OutOfRange::checkMin(numOutputDims, 1, "Positive number of output dimensions expected, %d got");
     Storage::checkChannelNumber(numOutputDims);
 }

Member Function Documentation

◆ prepare()

void Dense::prepare	(	GraphicPipeline &	gpu,
		ChunkCollection &	data,
		GL::ProgramBank &	bank
	)

privatevirtual

Compiles GLSL shaders.

Parameters

[in,out]	gpu	A graphic pipeline instance
[in,out]	data	Chunkfile containing operation data (e.g. weights and biases)
[in,out]	bank	A program bank with existing GLSL programs to be reused when possible. If a new program is built, it is added to the bank.

Implements Beatmup::NNets::AbstractOperation.

Definition at line 144 of file dense.cpp.

                                                                                     {
     RuntimeError::check(inputVector || inputStorage, "Input is not provided to Dense operation " + getName());
     RuntimeError::check(outputVector, "Output is not provided to Dense operation " + getName());
  
     const int numInputDims = inputVector ? inputVector->getSize() : inputStorage.getSize().volume();
  
     // set matrix
     const Chunk matrix(data, getName() + MATRIX_CHUNK_SUFFIX);
     if (matrix.size() != numInputDims * numOutputDims * sizeof(float))
         throw InconsistentModelData(this, "Matrix size mismatch");
     LinearMapping::setMatrix(gpu, numInputDims, numOutputDims, matrix.ptr<float>());
  
     // set bias
     if (useBias) {
         const Chunk bias(data, getName() + BIAS_CHUNK_SUFFIX);
         if (bias.size() != numOutputDims * sizeof(float))
             throw InconsistentModelData(this, "Bias size mismatch");
         LinearMapping::setBias(gpu, numOutputDims, bias.ptr<float>());
     }
  
     // prepare
     if (inputVector) {
         LinearMapping::prepare(gpu, *outputVector, *inputVector, &bank);
     }
     else {
         Storage::TextureHandler inputHandle(inputStorage, 0);
         LinearMapping::prepare(gpu, *outputVector, inputHandle, &bank);
     }
 }

◆ execute()

void Dense::execute	(	TaskThread &	thread,
		GraphicPipeline &	gpu
	)

privatevirtual

Executes the operation.

The operation should be prepared.

Parameters

[in,out]	thread	Calling CPU thread descriptor
[in,out]	gpu	A graphic pipeline instance

Implements Beatmup::NNets::AbstractOperation.

Definition at line 175 of file dense.cpp.

                                                             {
     if (inputVector) {
         LinearMapping::process(gpu, *outputVector, *inputVector);
     }
     else {
         Storage::TextureHandler inputHandle(inputStorage, 0);
         LinearMapping::process(gpu, *outputVector, inputHandle);
     }
 }

◆ getSampledChannels()

void Dense::getSampledChannels	(	int	index,
		int &	min,
		int &	max
	)		const

privatevirtual

Retrieves range of input features channels sampled at the same time for a specific input.

The operation would typically take the entire storage and sample it at once, if needed. If the number of textures in a storage exceeds the number of texture samplers that the GPU may use simultaneously, an exception occurs. This function provides the necessary information to limit the number of textures in the storage when allocating it. When the limit is reached, multiple channels are packed into a single texture in the storage.

Parameters

[in]	index	The input index. Expected to fall in the valid range, i.e. from zero to getInputCount() - 1 inclusive.
[out]	min	The minimum number of channels that can be sampled at once
[out]	max	The maximum number of channels that can be sampled at once

Implements Beatmup::NNets::AbstractOperation.

Definition at line 186 of file dense.cpp.

                                                                   {
     min = max = (index == 0 ? 4 : 0);
 }

◆ getOutputSize()

Size Beatmup::NNets::Dense::getOutputSize ( int outputIndex = 0 ) const

inlinevirtual

Returns full size of a specific operation output.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 67 of file dense.h.

                                                                  {
                 return Size(1, numOutputDims, 1);
             }

◆ getInputCount()

int Beatmup::NNets::Dense::getInputCount ( ) const

inlinevirtual

Returns number of operation inputs.

Inputs are then indexed from zero to the returned value minus one inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 71 of file dense.h.

71 { return 1; }

◆ getOutputCount()

int Beatmup::NNets::Dense::getOutputCount ( ) const

inlinevirtual

Returns number of operation outputs.

Outputs are then indexed from zero to the returned value minus one inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 72 of file dense.h.

72 { return 1; }

◆ acceptsStorageInput()

bool Beatmup::NNets::Dense::acceptsStorageInput ( int index = 0 ) const

inlinevirtual

Returns true if the operation can take a Storage::View at a specific input.

Neural network operations may accept different kinds of data containers on inputs and outputs, namely Storage::View, GL::Vector and textures. This function is used to check whether a given operation accepts a storage view on input.

Parameters

[in] index The input index. Expected to fall in the valid range, i.e. from zero to getInputCount() - 1 inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 74 of file dense.h.

74 { return index == 0; }

◆ acceptsVectorInput()

bool Beatmup::NNets::Dense::acceptsVectorInput ( int index = 0 ) const

inlinevirtual

Returns true if the operation can take a GL::Vector at a specific input.

Neural network operations may accept different kinds of data containers on inputs and outputs, namely Storage::View, GL::Vector and textures. This function is used to check whether a given operation accepts a vector on input.

Parameters

[in] index The input index. Expected to fall in the valid range, i.e. from zero to getInputCount() - 1 inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 75 of file dense.h.

75 { return index == 0; }

◆ acceptsVectorOutput()

bool Beatmup::NNets::Dense::acceptsVectorOutput ( int index = 0 ) const

inlinevirtual

Returns true if the operation can take a GL::Vector at a specific output.

Neural network operations may accept different kinds of data containers on outputs and outputs, namely Storage::View, GL::Vector and textures. This function is used to check whether a given operation accepts a vector on output.

Parameters

[in] index The output index. Expected to fall in the valid range, i.e. from zero to getOutputCount() - 1 inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 76 of file dense.h.

76 { return index == 0; }

◆ getOutput()

void Dense::getOutput	(	GL::Vector *&	vector,
		int	index = `0`
	)

virtual

Returns a GL::Vector bound to a specific operation output.

Parameters

[out]	vector	Pointer to the GL::Vector. If no vector is bound, becomes null.
[in]	index	The output index. Expected to fall in the valid range, i.e. from zero to getOutputCount() - 1 inclusive.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 40 of file dense.cpp.

                                                   {
     OutOfRange::check(index, 0, 1, "Output index out of range: %d");
     vector = this->outputVector;
 }

◆ setInput() [1/2]

void Dense::setInput	(	Storage::View &&	view,
		int	index = `0`
	)

virtual

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 46 of file dense.cpp.

                                                   {
     OutOfRange::check(index, 0, 1, "Input index out of range: %d");
     if (view) {
         RuntimeError::check(view.getStorage().getPadding() == 0, "Storages with padding are not supported");
         const auto& storage = view.getStorage();
         RuntimeError::check(view.getNumberOfTextures() == 1 && storage.getTextureWidth() == 1, "Input size mismatch: a column-like view is expected");
     }
     inputStorage = std::move(view);
     inputVector = nullptr;
 }

◆ setInput() [2/2]

void Dense::setInput	(	GL::Vector &	vector,
		int	index = `0`
	)

virtual

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 58 of file dense.cpp.

                                                 {
     OutOfRange::check(index, 0, 1, "Input index out of range: %d");
     inputStorage = Storage::View();
     inputVector = &vector;
 }

◆ setOutput()

void Dense::setOutput	(	GL::Vector &	vector,
		int	index = `0`
	)

virtual

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 65 of file dense.cpp.

                                                  {
     outputVector = &vector;
 }

◆ getOutputVectorFormat()

GL::Vector::Format Dense::getOutputVectorFormat ( ) const

Returns: a supported data format for GL::Vector on output.

Definition at line 70 of file dense.cpp.

                                                   {
 #ifdef BEATMUP_OPENGLVERSION_GLES20
     return GL::Vector::Format::FIXED16;
 #else
     return GL::Vector::Format::FLOAT;
 #endif
 }

◆ serialize()

std::map< std::string, std::string > Dense::serialize ( ) const

virtual

Returns a serialized representation of th operation;.

Implements Beatmup::NNets::AbstractOperation.

Definition at line 79 of file dense.cpp.

                                                     {
     return {
         { "_name",          getName() },
         { "_type",          "dense" },
         { "output_dims",    std::to_string(numOutputDims) },
         { "use_bias",       useBias ? "true" : "false" },
     };
 }

◆ disconnect()

void Dense::disconnect ( )

virtual

Assigns empty inputs and outputs.

Implements Beatmup::NNets::AbstractOperation.

Definition at line 114 of file dense.cpp.

                        {
     inputVector = nullptr;
     inputStorage = Storage::View();
     outputVector = nullptr;
 }

◆ countMultiplyAdds()

unsigned long Dense::countMultiplyAdds ( ) const

virtual

Counts (approximate) number of multiply-adds used by this operation.

A single multiply-add is one multiplication and one addition.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 121 of file dense.cpp.

                                              {
     const int numInputDims = inputVector ? inputVector->getSize() : inputStorage.getSize().volume();
     return numInputDims * numOutputDims;
 }

◆ countTexelFetches()

unsigned long Dense::countTexelFetches ( ) const

virtual

Counts (approximate) number of texels fetches.

Reimplemented from Beatmup::NNets::AbstractOperation.

Definition at line 127 of file dense.cpp.

                                              {
     const unsigned long numInputDims = inputVector ? inputVector->getSize() : inputStorage.getSize().volume();
     // sampling the matrix
     unsigned long count = numInputDims * numOutputDims / 4;
     // sampling the bias
     if (useBias)
         count += numOutputDims / 4;
 #ifdef BEATMUP_OPENGLVERSION_GLES20
     // using 16 bit fixed point format
     count *= 2;
 #endif
     // sampling the input vector
     count += inputVector ? inputVector->getHeight() : (inputStorage.getSize().volume() / 4);
     return count;
 }

◆ initDeserializer()

static bool Beatmup::NNets::Dense::initDeserializer ( )

static

Sets up deserialization of the operation.

Member Data Documentation

◆ numOutputDims

const int Beatmup::NNets::Dense::numOutputDims

private

output feature vector size / number of rows in A

Definition at line 44 of file dense.h.

◆ useBias

const bool Beatmup::NNets::Dense::useBias

private

if true, the bias vector b is searched in the model data to add to the output

Definition at line 45 of file dense.h.

◆ inputVector

GL::Vector* Beatmup::NNets::Dense::inputVector

private

if not null, points the input vector x; otherwise the input is a storage view

Definition at line 46 of file dense.h.

◆ outputVector

GL::Vector* Beatmup::NNets::Dense::outputVector

private

pointer to the output vector

Definition at line 47 of file dense.h.

◆ inputStorage

Storage::View Beatmup::NNets::Dense::inputStorage

private

if not empty, contains the input features; otherwise the input is a GL vector

Definition at line 48 of file dense.h.

◆ MATRIX_CHUNK_SUFFIX

const char * Dense::MATRIX_CHUNK_SUFFIX = "/w"

static

suffix added to the op name to get the matrix chunk id in the model data

Definition at line 55 of file dense.h.

◆ BIAS_CHUNK_SUFFIX

const char * Dense::BIAS_CHUNK_SUFFIX = "/b"

static

suffix added to the op name to get the bias chunk id in the model data

Definition at line 56 of file dense.h.

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

core/nnets/dense.h
core/nnets/dense.cpp

Public Member Functions

Static Public Member Functions

Static Public Attributes

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ Dense()

Member Function Documentation

◆ prepare()

◆ execute()

◆ getSampledChannels()

◆ getOutputSize()

◆ getInputCount()

◆ getOutputCount()

◆ acceptsStorageInput()

◆ acceptsVectorInput()

◆ acceptsVectorOutput()

◆ getOutput()

◆ setInput() [1/2]

◆ setInput() [2/2]

◆ setOutput()

◆ getOutputVectorFormat()

◆ serialize()

◆ disconnect()

◆ countMultiplyAdds()

◆ countTexelFetches()

◆ initDeserializer()

Member Data Documentation

◆ numOutputDims

◆ useBias

◆ inputVector

◆ outputVector

◆ inputStorage

◆ MATRIX_CHUNK_SUFFIX

◆ BIAS_CHUNK_SUFFIX