Mila::Dnn::Compute::Cpu::TransferOps Struct Reference

CPU specialization of TensorOps for transfer operations. More...

Inheritance diagram for Mila::Dnn::Compute::Cpu::TransferOps:

Static Public Member Functions
template<TensorDataType TSrcDataType, typename TSrcMemoryResource, TensorDataType TDstDataType, typename TDstMemoryResource> requires isValidTensor<TSrcDataType, TSrcMemoryResource>&& isValidTensor<TDstDataType, TDstMemoryResource>
static void	copy (const Tensor< TSrcDataType, TSrcMemoryResource > &src, Tensor< TDstDataType, TDstMemoryResource > &dst, IExecutionContext *exec_context=nullptr)
	Copy tensor data between pre-allocated tensors.
template<TensorDataType TDstDataType, typename TDstMemoryResource> requires isValidTensor<TDstDataType, TDstMemoryResource>
static void	copyFromBlob (const Serialization::ITensorBlob &blob, Tensor< TDstDataType, TDstMemoryResource > &dst, IExecutionContext *exec_context=nullptr)

Static Private Member Functions
template<TensorDataType TSrcDataType, TensorDataType TDstDataType>
static void	hostConvertImpl (const void *src_data, void *dst_data, size_t count)
	Private wrapper forwarding to Detail::copyHostToHostWithConversionImpl.
template<TensorDataType TDataType>
static void	hostCopyImpl (const void *src_data, void *dst_data, size_t count)
	Private wrapper forwarding to Detail::copyHostToHostImpl.

Detailed Description

CPU specialization of TensorOps for transfer operations.

This specialization provides CPU-specific implementations of tensor transfer operations for the DeviceType::Cpu device type. Handles both same-type and type-conversion transfers with host memory optimization.

Key features:

• Direct memcpy for same-type transfers (optimal performance)
• Element-wise conversion for different-type transfers
• Accepts ExecutionContext for API consistency (unused on CPU)
• All operations are synchronous (no stream management)

Member Function Documentation

copy()

template<TensorDataType TSrcDataType, typename TSrcMemoryResource, TensorDataType TDstDataType, typename TDstMemoryResource>
requires isValidTensor<TSrcDataType, TSrcMemoryResource>&& isValidTensor<TDstDataType, TDstMemoryResource>

void Mila::Dnn::Compute::Cpu::TransferOps::copy ( const Tensor< TSrcDataType, TSrcMemoryResource > & src, Tensor< TDstDataType, TDstMemoryResource > & dst, IExecutionContext * exec_context = nullptr )

inlinestatic

Copy tensor data between pre-allocated tensors.

Validates that src and dst have identical shapes, ensures non-zero element count, and dispatches to the appropriate same-type or conversion copy implementation.

Template Parameters

TSrcDataType	Abstract source tensor data type
TSrcMemoryResource	Source memory resource type (must be host-accessible)
TDstDataType	Abstract destination tensor data type
TDstMemoryResource	Destination memory resource type (must be host-accessible)

Parameters

src	Source tensor (pre-allocated)
dst	Destination tensor (pre-allocated)
exec_context	Optional execution context pointer (ignored for CPU)

Exceptions

std::invalid_argument	If tensor shapes do not match
std::runtime_error	If underlying data pointers are invalid

Note

This function is synchronous and does not use the execution context.

Here is the call graph for this function:

copyFromBlob()

template<TensorDataType TDstDataType, typename TDstMemoryResource>
requires isValidTensor<TDstDataType, TDstMemoryResource>

void Mila::Dnn::Compute::Cpu::TransferOps::copyFromBlob ( const Serialization::ITensorBlob & blob, Tensor< TDstDataType, TDstMemoryResource > & dst, IExecutionContext * exec_context = nullptr )

inlinestatic

Here is the call graph for this function:

hostConvertImpl()

template<TensorDataType TSrcDataType, TensorDataType TDstDataType>

void Mila::Dnn::Compute::Cpu::TransferOps::hostConvertImpl ( const void * src_data, void * dst_data, size_t count )

inlinestaticprivate

Private wrapper forwarding to Detail::copyHostToHostWithConversionImpl.

Thin forwarding wrapper used to encapsulate the internal conversion implementation.

Template Parameters

TSrcDataType	Abstract source tensor data type
TDstDataType	Abstract destination tensor data type

Parameters

src_data	Pointer to source memory
dst_data	Pointer to destination memory
count	Number of logical elements to convert and copy

Here is the call graph for this function:

Here is the caller graph for this function:

hostCopyImpl()

template<TensorDataType TDataType>

void Mila::Dnn::Compute::Cpu::TransferOps::hostCopyImpl ( const void * src_data, void * dst_data, size_t count )

inlinestaticprivate

Private wrapper forwarding to Detail::copyHostToHostImpl.

Thin forwarding wrapper used to encapsulate the internal implementation.

Template Parameters

TDataType

Abstract tensor data type

Parameters

src_data	Pointer to source memory
dst_data	Pointer to destination memory
count	Number of logical elements to copy

Here is the call graph for this function:

Here is the caller graph for this function:

---

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

• /__w/Mila/Mila/Mila/Src/Dnn/Compute/Devices/Cpu/Tensors/Operations/CpuTensorOps.Transfer.ixx