gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t Struct Reference

BATCH_GEMM arguments. More...

#include <batch_gemm.hpp>

Public Member Functions
	arguments_t ()=default
	Constructs arguments with default method.
	arguments_t (uint32_t batch_size_, uint32_t matrix_m_, uint32_t matrix_k_, uint32_t matrix_n_, matA_base_t matA_base_, uint32_t matA_ld_, matB_base_t matB_base_, uint32_t matB_ld_, matC_base_t matC_base_, uint32_t matC_ld_, epilogue_args_t epilogue_args_={})
	Constructs arguments with initialization list.
	arguments_t (const arguments_t &args)
arguments_t &	operator= (const arguments_t &args)

Public Attributes
uint32_t	batch_size
	Is the number of total batches.
uint32_t	matrix_m
	Is the size of the m dimension of the matrix multiplication (m x k x n).
uint32_t	matrix_k
	Is the size of the k dimension of the matrix multiplication (m x k x n).
uint32_t	matrix_n
	Is the size of the n dimension of the matrix multiplication (m x k x n).
uint32_t	matA_ld
	Is the leading dimension (pitch) size of the matrix A in memory.
uint32_t	matB_ld
	Is the leading dimension (pitch) size of the matrix B in memory.
uint32_t	matC_ld
	Is the leading dimension (pitch) size of the matrix C in memory.
matA_base_t	matA_base
	Is the base address of matrix A.
matB_base_t	matB_base
	Is the base address of matrix B.
matC_base_t	matC_base
	Is the base address of matrix C.
epilogue_args_t	epilogue_args
	Is the epilogue arguments.

Static Public Attributes
static constexpr bool	host_callable = true
	Set for device copyable.

Detailed Description

template<typename gemm_t_, typename epilogue_t_, gpu_arch arch_tag_>
struct gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t

BATCH_GEMM arguments.

This is the interface for users to pass the application-related runtime variables.

Constructor & Destructor Documentation

arguments_t() [1/3]

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::arguments_t ( )

inlinedefault

Constructs arguments with default method.

arguments_t() [2/3]

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::arguments_t ( uint32_t  batch_size_, uint32_t  matrix_m_, uint32_t  matrix_k_, uint32_t  matrix_n_, matA_base_t  matA_base_, uint32_t  matA_ld_, matB_base_t  matB_base_, uint32_t  matB_ld_, matC_base_t  matC_base_, uint32_t  matC_ld_, epilogue_args_t  epilogue_args_ = {}  )

inline

Constructs arguments with initialization list.

Parameters

batch_size_	Is the number of total batches.
matrix_m_	Is the size of the m dimension of the matrix multiplication (m x k x n).
matrix_k_	Is the size of the k dimension of the matrix multiplication (m x k x n).
matrix_n_	Is the size of the n dimension of the matrix multiplication (m x k x n).
matA_base_	Is the base address of matrix A.
matA_ld_	Is the leading dimension (pitch) size of the matrix A in memory.
matB_base_	Is the base address of matrix B.
matB_ld_	Is the leading dimension (pitch) size of the matrix B in memory.
matC_base_	Is the base address of matrix C.
matC_ld_	Is the leading dimension (pitch) size of the matrix C in memory.
epilogue_args_	Is the epilogue arguments.

arguments_t() [3/3]

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::arguments_t ( const arguments_t &  args )

inline

Member Function Documentation

operator=()

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

arguments_t & gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::operator= ( const arguments_t &  args )

inline

Member Data Documentation

batch_size

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::batch_size

Is the number of total batches.

epilogue_args

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

epilogue_args_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::epilogue_args

Is the epilogue arguments.

host_callable

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

constexpr bool gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::host_callable = true

staticconstexpr

Set for device copyable.

matA_base

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

matA_base_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matA_base

Is the base address of matrix A.

matA_ld

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matA_ld

Is the leading dimension (pitch) size of the matrix A in memory.

matB_base

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

matB_base_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matB_base

Is the base address of matrix B.

matB_ld

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matB_ld

Is the leading dimension (pitch) size of the matrix B in memory.

matC_base

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

matC_base_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matC_base

Is the base address of matrix C.

matC_ld

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matC_ld

Is the leading dimension (pitch) size of the matrix C in memory.

matrix_k

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matrix_k

Is the size of the k dimension of the matrix multiplication (m x k x n).

matrix_m

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matrix_m

Is the size of the m dimension of the matrix multiplication (m x k x n).

matrix_n

template<typename gemm_t_ , typename epilogue_t_ , gpu_arch arch_tag_>

uint32_t gpu::xetla::kernel::batch_gemm_t< gemm_t_, epilogue_t_, arch_tag_ >::arguments_t::matrix_n

Is the size of the n dimension of the matrix multiplication (m x k x n).