Mila::Dnn::detail Namespace Reference

Functions
template<typename MR >
constexpr size_t	get_alignment ()
	Determines the appropriate memory alignment based on the memory resource type.

Variables
constexpr size_t	CPU_SIMD_ALIGN = 64
	AVX-512 alignment for CPU operations.
constexpr size_t	CUDA_WARP_SIZE = 32
	CUDA warp size alignment (32 threads)

Function Documentation

get_alignment()

template<typename MR >

constexpr size_t Mila::Dnn::detail::get_alignment ( )

constexpr

Determines the appropriate memory alignment based on the memory resource type.

For GPU memory resources, alignment is based on CUDA warp size. For CPU memory resources, alignment is based on AVX-512 requirements.

Template Parameters

MR	The memory resource type

Returns

constexpr size_t The required alignment in bytes

Variable Documentation

CPU_SIMD_ALIGN

constexpr size_t Mila::Dnn::detail::CPU_SIMD_ALIGN = 64

constexpr

AVX-512 alignment for CPU operations.

This constant defines the alignment required for optimal SIMD operations using AVX-512 instructions on CPU.

CUDA_WARP_SIZE

constexpr size_t Mila::Dnn::detail::CUDA_WARP_SIZE = 32

constexpr

CUDA warp size alignment (32 threads)

This constant defines the alignment required for optimal memory access patterns when using CUDA warp-level operations.