Mila::Dnn::Serialization::TensorBlob< MR > Struct Template Reference

Concrete tensor blob owning a TensorBuffer-backed raw byte buffer. More...

Inheritance diagram for Mila::Dnn::Serialization::TensorBlob< MR >:

Public Member Functions
	TensorBlob (const TensorBlob &)=delete
	TensorBlob (TensorBlob &&)=default
	TensorBlob (TensorMetadata meta, TensorBuffer< dtype_t::UINT8, MR > buf)
const void *	data () const noexcept override
const TensorMetadata &	getMetadata () const noexcept override
TensorBlob &	operator= (const TensorBlob &)=delete
TensorBlob &	operator= (TensorBlob &&)=default
size_t	sizeBytes () const noexcept override
Public Member Functions inherited from Mila::Dnn::Serialization::ITensorBlob
	ITensorBlob (const ITensorBlob &)=delete
	ITensorBlob (ITensorBlob &&)=default
ITensorBlob &	operator= (const ITensorBlob &)=delete
ITensorBlob &	operator= (ITensorBlob &&)=default

Public Attributes
TensorBuffer< dtype_t::UINT8, MR >	data_buffer
TensorMetadata	metadata

Detailed Description

template<typename MR = Compute::CpuMemoryResource>
requires isValidTensor<dtype_t::UINT8, MR>
struct Mila::Dnn::Serialization::TensorBlob< MR >

Concrete tensor blob owning a TensorBuffer-backed raw byte buffer.

The memory resource controls the allocation strategy:

TensorBlob<CpuMemoryResource> – pageable host memory (default) TensorBlob<CudaPinnedMemoryResource> – pinned host memory; copyFromBlob issues a direct DMA to device with no hidden driver staging

Template Parameters

MR	Memory resource type. Must satisfy isValidTensor<UINT8, MR>.

Member Function Documentation

data()

template<typename MR = Compute::CpuMemoryResource>

const void * Mila::Dnn::Serialization::TensorBlob< MR >::data ( ) const

inlineoverridevirtualnoexcept

Implements Mila::Dnn::Serialization::ITensorBlob.

getMetadata()

template<typename MR = Compute::CpuMemoryResource>

const TensorMetadata & Mila::Dnn::Serialization::TensorBlob< MR >::getMetadata ( ) const

inlineoverridevirtualnoexcept

Implements Mila::Dnn::Serialization::ITensorBlob.

sizeBytes()

template<typename MR = Compute::CpuMemoryResource>

size_t Mila::Dnn::Serialization::TensorBlob< MR >::sizeBytes ( ) const

inlineoverridevirtualnoexcept

Implements Mila::Dnn::Serialization::ITensorBlob.

---

The documentation for this struct was generated from the following file:

• Mila/Src/Dnn/Tensors/Tensor.Serialization.ixx