TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK > Struct Template Reference

• ck_tile
• TensorDescriptorUtils

Static Public Member Functions | List of all members

ck_tile::TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK > Struct Template Reference

Utility class for creating tensor descriptors in batched contraction operations. More...

#include <tensor_descriptor_utils.hpp>

Static Public Member Functions
static CK_TILE_HOST constexpr auto	Make_A_GridDescriptor_M_K (const std::vector< ck_tile::index_t > &A_dims={}, const std::vector< ck_tile::index_t > &A_strides={})
	Creates a tensor descriptor for input tensor A with batch dimensions removed.
static CK_TILE_HOST constexpr auto	Make_B_GridDescriptor_N_K (const std::vector< ck_tile::index_t > &B_dims={}, const std::vector< ck_tile::index_t > &B_strides={})
	Creates a tensor descriptor for input tensor B with batch dimensions removed.
static CK_TILE_HOST constexpr auto	Make_E_GridDescriptor_M_N (const std::vector< ck_tile::index_t > &E_dims={}, const std::vector< ck_tile::index_t > &E_strides={})
	Creates a tensor descriptor for output tensor E with batch dimensions removed.

Detailed Description

template<ck_tile::index_t NumDimG, ck_tile::index_t NumDimM, ck_tile::index_t NumDimN, ck_tile::index_t NumDimK>
struct ck_tile::TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK >

Utility class for creating tensor descriptors in batched contraction operations.

Template Parameters

NumDimG	Number of batch dimensions
NumDimM	Number of M (output row) dimensions
NumDimN	Number of N (output column) dimensions
NumDimK	Number of K (contraction) dimensions

Member Function Documentation

◆ Make_A_GridDescriptor_M_K()

template<ck_tile::index_t NumDimG, ck_tile::index_t NumDimM, ck_tile::index_t NumDimN, ck_tile::index_t NumDimK>

CK_TILE_HOST constexpr auto ck_tile::TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK >::Make_A_GridDescriptor_M_K ( const std::vector< ck_tile::index_t > & A_dims = {}, const std::vector< ck_tile::index_t > & A_strides = {} )

inlinestaticconstexpr

Creates a tensor descriptor for input tensor A with batch dimensions removed.

Parameters

A_dims	Dimension vector for tensor A: [G0, G1, ..., M0, M1, M2, ..., K0, K1, K2, ...]
A_strides	Stride vector for tensor A: [G0, G1, ..., M0, M1, M2, ..., K0, K1, K2, ...]

Returns

Flattened tensor descriptor: [M_total, K_total] for GEMM computation

Removes batch dimensions and flattens M and K dimensions for efficient GEMM execution

◆ Make_B_GridDescriptor_N_K()

template<ck_tile::index_t NumDimG, ck_tile::index_t NumDimM, ck_tile::index_t NumDimN, ck_tile::index_t NumDimK>

CK_TILE_HOST constexpr auto ck_tile::TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK >::Make_B_GridDescriptor_N_K ( const std::vector< ck_tile::index_t > & B_dims = {}, const std::vector< ck_tile::index_t > & B_strides = {} )

inlinestaticconstexpr

Creates a tensor descriptor for input tensor B with batch dimensions removed.

Parameters

B_dims	Dimension vector for tensor B: [G0, G1, ..., N0, N1, N2, ..., K0, K1, K2, ...]
B_strides	Stride vector for tensor B: [G0, G1, ..., N0, N1, N2, ..., K0, K1, K2, ...]

Returns

Flattened tensor descriptor: [N_total, K_total] for GEMM computation

Removes batch dimensions and flattens N and K dimensions for efficient GEMM execution

◆ Make_E_GridDescriptor_M_N()

template<ck_tile::index_t NumDimG, ck_tile::index_t NumDimM, ck_tile::index_t NumDimN, ck_tile::index_t NumDimK>

CK_TILE_HOST constexpr auto ck_tile::TensorDescriptorUtils< NumDimG, NumDimM, NumDimN, NumDimK >::Make_E_GridDescriptor_M_N ( const std::vector< ck_tile::index_t > & E_dims = {}, const std::vector< ck_tile::index_t > & E_strides = {} )

inlinestaticconstexpr

Creates a tensor descriptor for output tensor E with batch dimensions removed.

Parameters

E_dims	Dimension vector for tensor E: [G0, G1, ..., M0, M1, M2, ..., N0, N1, N2, ...]
E_strides	Stride vector for tensor E: [G0, G1, ..., M0, M1, M2, ..., N0, N1, N2, ...]

Returns

Flattened tensor descriptor: [M_total, N_total] for GEMM computation

Removes batch dimensions and flattens M and N dimensions for efficient GEMM execution

---

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

• tensor_descriptor_utils.hpp