tensor

package
v2.0.2 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Feb 17, 2022 License: MIT Imports: 10 Imported by: 5

Documentation

Overview

Package tensor provides GPU tensor and tensor operation implementations.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

This section is empty.

Types

type AvgPoolingKernelArgsBackward 

type AvgPoolingKernelArgsBackward struct {
	NumThreads uint64
	Top        driver.Ptr
	N          int32
	C          int32
	H          int32
	W          int32
	PooledH    int32
	PooledW    int32
	KernelH    int32
	KernelW    int32
	StrideH    int32
	StrideW    int32
	PadH       int32
	PadW       int32
	Bottom     driver.Ptr

	HiddenGlobalOffsetX int64
	HiddenGlobalOffsetY int64
	HiddenGlobalOffsetZ int64
}

AvgPoolingKernelArgsBackward defines forward kernel arguments

type AvgPoolingKernelArgsForward 

type AvgPoolingKernelArgsForward struct {
	NumThreads uint64
	Bottom     driver.Ptr
	N          int32
	C          int32
	H          int32
	W          int32
	PooledH    int32
	PooledW    int32
	KernelH    int32
	KernelW    int32
	StrideH    int32
	StrideW    int32
	PadH       int32
	PadW       int32
	Top        driver.Ptr

	HiddenGlobalOffsetX int64
	HiddenGlobalOffsetY int64
	HiddenGlobalOffsetZ int64
}

AvgPoolingKernelArgsForward defines forward kernel arguments

type GPUOperator 

type GPUOperator struct {
	// contains filtered or unexported fields
}

GPUOperator can perform operations on GPU tensors.

func NewGPUOperator 

func NewGPUOperator(
	gpuDriver *driver.Driver,
	ctx *driver.Context,
) *GPUOperator

NewGPUOperator creates a new GPU Operator.

func (*GPUOperator) Adam 

func (o *GPUOperator) Adam(
	params, gradient, vHistory, sHistory tensor.Tensor,
	smoothFactor1, smoothFactor2, learningRate float64,
)

Adam uses the Adam algorithm to update the parameters

func (*GPUOperator) AvgPoolingBackward 

func (o *GPUOperator) AvgPoolingBackward(
	forwardIn, backwardIn tensor.Tensor,
	kernelSize, padding, stride []int,
) tensor.Tensor

AvgPoolingBackward claculates the backward propagation of the average pooling layer.

func (*GPUOperator) AvgPoolingForward 

func (o *GPUOperator) AvgPoolingForward(
	t tensor.Tensor,
	kernelSize, padding, stride []int,
) tensor.Tensor

AvgPoolingForward calculates the forward propagation of the average pooling layer.

func (*GPUOperator) Clear 

func (o *GPUOperator) Clear(t tensor.Tensor)

Clear sets the content of the tensor to 0.

func (*GPUOperator) Clone 

func (o *GPUOperator) Clone(t tensor.Tensor) tensor.Tensor

Clone duplicates the input tensor.

func (*GPUOperator) Copy 

func (o *GPUOperator) Copy(dst tensor.Tensor, src tensor.Tensor)

Copy copies data from one tensor to another tensor. The src and dst tensor must have the same number of elements.

func (*GPUOperator) Create 

func (o *GPUOperator) Create(size []int) tensor.Tensor

Create creates a new GPU tensor

func (*GPUOperator) CreateWithData 

func (o *GPUOperator) CreateWithData(
	data []float64,
	size []int,
	descriptor string,
) tensor.Tensor

CreateWithData creates the tensor and copies the given data to the GPU memory.

func (*GPUOperator) CrossEntropy 

func (o *GPUOperator) CrossEntropy(t tensor.Tensor, label []int) float64

CrossEntropy calculates the cross entropy of the output.

func (*GPUOperator) CrossEntropyDerivative 

func (o *GPUOperator) CrossEntropyDerivative(
	t tensor.Tensor, label []int,
) tensor.Tensor

CrossEntropyDerivative calculates the derivative using cross entropies.

func (*GPUOperator) Dilate 

func (o *GPUOperator) Dilate(t tensor.Tensor, dilate []int) tensor.Tensor

Dilate adds 0s between rows and columns.

func (*GPUOperator) Dump 

func (o *GPUOperator) Dump(t tensor.Tensor) string

Dump writes the content of the tensor to a string.

func (*GPUOperator) ElementWiseMul 

func (o *GPUOperator) ElementWiseMul(
	a, b tensor.Tensor,
) tensor.Tensor

ElementWiseMul calculates the element multiplication of A and B.

func (*GPUOperator) EnableVerification 

func (o *GPUOperator) EnableVerification()

EnableVerification will run the same operations in a CPU operator and compare the results.

func (*GPUOperator) Free 

func (o *GPUOperator) Free(t tensor.Tensor)

Free releases the allocated GPU memory.

func (*GPUOperator) Gemm 

func (o *GPUOperator) Gemm(
	transA, transB bool,
	alpha, beta float64,
	a, b, c tensor.Tensor,
) tensor.Tensor

Gemm performs alpha * A * B + beta * C operation.

func (*GPUOperator) Im2Col 

func (o *GPUOperator) Im2Col(
	t tensor.Tensor,
	kernelSize, padding, stride, dilation []int,
) tensor.Tensor

Im2Col converts images to columns so that convolutional operations can be completed with GEMM.

func (*GPUOperator) Init 

func (o *GPUOperator) Init(t tensor.Tensor, data []float64)

Init sets the data of the tensor

func (*GPUOperator) MaxPoolingBackward 

func (o *GPUOperator) MaxPoolingBackward(
	forwardIn, backwardIn tensor.Tensor,
	mask tensor.Tensor,
	kernelSize, padding, stride []int,
) tensor.Tensor

MaxPoolingBackward calculates the backward propagation of the max pooling layer.

func (*GPUOperator) MaxPoolingForward 

func (o *GPUOperator) MaxPoolingForward(
	t tensor.Tensor,
	kernelSize, padding, stride []int,
) (out tensor.Tensor, mask tensor.Tensor)

MaxPoolingForward calculates the forward propagation of the max pooling layer.

func (*GPUOperator) RMSProp 

func (o *GPUOperator) RMSProp(
	params, gradient, sHistory tensor.Tensor,
	smoothFactor, learningRate float64,
)

RMSProp uses the RMSProp algorithm to update the parameters

func (*GPUOperator) ReluBackward 

func (o *GPUOperator) ReluBackward(
	forwardIn, backIn tensor.Tensor,
) tensor.Tensor

ReluBackward Implementation

func (*GPUOperator) ReluForward 

func (o *GPUOperator) ReluForward(
	in tensor.Tensor,
) tensor.Tensor

ReluForward Implementation

func (*GPUOperator) Repeat 

func (o *GPUOperator) Repeat(t tensor.Tensor, times int) tensor.Tensor

Repeat will create another tensor that duplicates the input tensor by n times.

func (*GPUOperator) ReportTime added in v2.0.1 

func (o *GPUOperator) ReportTime()

ReportTime lets the operator to report the execution time of each kernel.

func (*GPUOperator) Reshape 

func (o *GPUOperator) Reshape(t tensor.Tensor, newSize []int) tensor.Tensor

Reshape creates another tensor with the same elements but a different size.

func (*GPUOperator) Rotate180 

func (o *GPUOperator) Rotate180(t tensor.Tensor) tensor.Tensor

Rotate180 rotates the lowest two dimensions of the tensor by 180 degree.

func (*GPUOperator) ScaleAdd 

func (o *GPUOperator) ScaleAdd(
	alpha, beta float64,
	a, b tensor.Tensor,
) tensor.Tensor

ScaleAdd performs element-wide alpha*A + beta*B operation.

func (*GPUOperator) Slice 

func (o *GPUOperator) Slice(t tensor.Tensor, start int, end int) tensor.Tensor

Slice will create another tensor that shares part of the buffer with the input tensor.

func (*GPUOperator) Softmax 

func (o *GPUOperator) Softmax(t tensor.Tensor) tensor.Tensor

Softmax performs the softmax operation.

func (*GPUOperator) SoftmaxCrossEntropyDerivative 

func (o *GPUOperator) SoftmaxCrossEntropyDerivative(
	t tensor.Tensor,
	label []int,
) tensor.Tensor

SoftmaxCrossEntropyDerivative calculates the derivatives using both softmax /// and cross entropy algorithms.

func (*GPUOperator) Sum 

func (o *GPUOperator) Sum(t tensor.Tensor, axis []int) tensor.Tensor

Sum reduces the number of axes by summing the numbers on given axes.

func (*GPUOperator) Transpose 

func (o *GPUOperator) Transpose(t tensor.Tensor, order []int) tensor.Tensor

Transpose reorders the axises of the tensor.

func (*GPUOperator) Zeros 

func (o *GPUOperator) Zeros(size []int) tensor.Tensor

Zeros creates a tensor prefilled with zeros.

type Tensor 

type Tensor struct {
	// contains filtered or unexported fields
}

A Tensor is a multi-dementional array.

func (Tensor) Descriptor 

func (t Tensor) Descriptor() string

Descriptor returns the descriptor of the tensor.

func (Tensor) Dim 

func (t Tensor) Dim() int

Dim returns the number of dimensions that the tensor has.

func (Tensor) NumElement 

func (t Tensor) NumElement() int

NumElement returns the number of elements in the tensor.

func (*Tensor) Ptr 

func (t *Tensor) Ptr() driver.Ptr

Ptr returns the GPU pointer of the tensor.

func (*Tensor) SetDescriptor 

func (t *Tensor) SetDescriptor(d string)

SetDescriptor sets the descriptor of the tensor.

func (*Tensor) SetSize 

func (t *Tensor) SetSize(s []int)

SetSize sets the size of the tensor.

func (Tensor) Size 

func (t Tensor) Size() []int

Size returns the size of the tensor on each dimension.

func (*Tensor) Vector 

func (t *Tensor) Vector() []float64

Vector copies the data from the GPU to the simulator.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.