MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias > Struct Template Reference

MoeFlatmmHostArgs&lt; ScaleM, ScaleN, ExpertBias &gt; Struct Template Reference#

Composable Kernel: ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias > Struct Template Reference
Public Member Functions | Public Attributes | List of all members
ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias > Struct Template Reference

#include <moe_flatmm_kernel.hpp>

Inheritance diagram for ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >:
ck_tile::ScaleFlatmmHostArgs< ScaleM, ScaleN, NumDTensor > ck_tile::BaseFlatmmHostArgs< NumDTensor >

Public Member Functions

CK_TILE_HOST MoeFlatmmHostArgs () noexcept=default
CK_TILE_HOST MoeFlatmmHostArgs (const ck_tile::index_t *p_sorted_token_ids_, const void *p_sorted_expert_weights_, const ck_tile::index_t *p_sorted_expert_ids_, const ck_tile::index_t *p_max_token_id_, const void *a_ptr_, const void *b_ptr_, void *c_ptr_, ck_tile::index_t NumTokens_, ck_tile::index_t NumExperts_, ck_tile::index_t TopK_, ck_tile::index_t k_batch_, ck_tile::index_t M_, ck_tile::index_t N_, ck_tile::index_t K_, ck_tile::index_t stride_A_, ck_tile::index_t stride_B_, ck_tile::index_t stride_C_, ScaleM scale_m_={}, ScaleN scale_n_={}, ExpertBias exp_bias_={})
CK_TILE_HOST MoeFlatmmHostArgs (const ck_tile::index_t *p_sorted_token_ids_, const void *p_sorted_expert_weights_, const ck_tile::index_t *p_sorted_expert_ids_, const ck_tile::index_t *p_max_token_id_, const void *a_ptr_, const void *b_ptr_, void *c_ptr_, ck_tile::index_t NumTokens_, ck_tile::index_t NumExperts_, ck_tile::index_t TopK_, ck_tile::index_t k_batch_, ck_tile::index_t M_, ck_tile::index_t N_, ck_tile::index_t K_, ck_tile::index_t stride_A_, ck_tile::index_t stride_B_, ck_tile::index_t stride_C_, ck_tile::index_t n_padded_zeros_=0, ck_tile::index_t k_padded_zeros_=0, ScaleM scale_m_={}, ScaleN scale_n_={}, ExpertBias exp_bias_={})
Public Member Functions inherited from ck_tile::ScaleFlatmmHostArgs< ScaleM, ScaleN, NumDTensor >
CK_TILE_HOST ScaleFlatmmHostArgs ()=default
CK_TILE_HOST ScaleFlatmmHostArgs (const void *a_ptr_, const void *b_shuffle_ptr_, const std::array< const void *, NumDTensor > &ds_ptr_, void *c_ptr_, index_t k_batch_, index_t M_, index_t N_, index_t K_, index_t stride_A_, index_t stride_B_, const std::array< index_t, NumDTensor > &stride_Ds_, index_t stride_C_, ScaleM scale_m_=nullptr, ScaleN scale_n_=nullptr)
Public Member Functions inherited from ck_tile::BaseFlatmmHostArgs< NumDTensor >
CK_TILE_HOST BaseFlatmmHostArgs ()=default
CK_TILE_HOST BaseFlatmmHostArgs (const void *a_ptr_, const void *b_ptr_, const std::array< const void *, NumDTensor > &ds_ptr_, void *e_ptr_, index_t k_batch_, index_t M_, index_t N_, index_t K_, index_t stride_A_, index_t stride_B_, const std::array< index_t, NumDTensor > &stride_Ds_, index_t stride_E_)

Public Attributes

ck_tile::index_t NumTokens
ck_tile::index_t NumExperts
ck_tile::index_t TopK
const ck_tile::index_t * p_sorted_token_ids
const ck_tile::index_t * p_sorted_expert_ids
const ck_tile::index_t * p_max_token_id
const void * p_sorted_expert_weights
const ck_tile::index_t n_padded_zeros
const ck_tile::index_t k_padded_zeros
ExpertBias exp_bias
Public Attributes inherited from ck_tile::ScaleFlatmmHostArgs< ScaleM, ScaleN, NumDTensor >
ScaleM scale_m = nullptr
ScaleN scale_n = nullptr
Public Attributes inherited from ck_tile::BaseFlatmmHostArgs< NumDTensor >
const void * a_ptr
const void * b_ptr
const std::array< const void *, NumDTensor > ds_ptr
union { 
   void *   e_ptr 
   void *   c_ptr 
}; 
index_t M
index_t N
index_t K
index_t stride_A
index_t stride_B
const std::array< index_t, NumDTensor > stride_Ds
union { 
   index_t   stride_E 
   index_t   stride_C 
}; 
index_t k_batch

Constructor & Destructor Documentation

◆ MoeFlatmmHostArgs() [1/3]

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
CK_TILE_HOST ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::MoeFlatmmHostArgs ( )
defaultnoexcept

◆ MoeFlatmmHostArgs() [2/3]

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
CK_TILE_HOST ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::MoeFlatmmHostArgs ( const ck_tile::index_t * p_sorted_token_ids_,
const void * p_sorted_expert_weights_,
const ck_tile::index_t * p_sorted_expert_ids_,
const ck_tile::index_t * p_max_token_id_,
const void * a_ptr_,
const void * b_ptr_,
void * c_ptr_,
ck_tile::index_t NumTokens_,
ck_tile::index_t NumExperts_,
ck_tile::index_t TopK_,
ck_tile::index_t k_batch_,
ck_tile::index_t M_,
ck_tile::index_t N_,
ck_tile::index_t K_,
ck_tile::index_t stride_A_,
ck_tile::index_t stride_B_,
ck_tile::index_t stride_C_,
ScaleM scale_m_ = {},
ScaleN scale_n_ = {},
ExpertBias exp_bias_ = {} )
inline

◆ MoeFlatmmHostArgs() [3/3]

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
CK_TILE_HOST ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::MoeFlatmmHostArgs ( const ck_tile::index_t * p_sorted_token_ids_,
const void * p_sorted_expert_weights_,
const ck_tile::index_t * p_sorted_expert_ids_,
const ck_tile::index_t * p_max_token_id_,
const void * a_ptr_,
const void * b_ptr_,
void * c_ptr_,
ck_tile::index_t NumTokens_,
ck_tile::index_t NumExperts_,
ck_tile::index_t TopK_,
ck_tile::index_t k_batch_,
ck_tile::index_t M_,
ck_tile::index_t N_,
ck_tile::index_t K_,
ck_tile::index_t stride_A_,
ck_tile::index_t stride_B_,
ck_tile::index_t stride_C_,
ck_tile::index_t n_padded_zeros_ = 0,
ck_tile::index_t k_padded_zeros_ = 0,
ScaleM scale_m_ = {},
ScaleN scale_n_ = {},
ExpertBias exp_bias_ = {} )
inline

Member Data Documentation

◆ exp_bias

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
ExpertBias ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::exp_bias

◆ k_padded_zeros

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const ck_tile::index_t ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::k_padded_zeros

◆ n_padded_zeros

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const ck_tile::index_t ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::n_padded_zeros

◆ NumExperts

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
ck_tile::index_t ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::NumExperts

◆ NumTokens

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
ck_tile::index_t ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::NumTokens

◆ p_max_token_id

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const ck_tile::index_t* ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::p_max_token_id

◆ p_sorted_expert_ids

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const ck_tile::index_t* ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::p_sorted_expert_ids

◆ p_sorted_expert_weights

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const void* ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::p_sorted_expert_weights

◆ p_sorted_token_ids

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
const ck_tile::index_t* ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::p_sorted_token_ids

◆ TopK

template<class ScaleM = FlatmmScalePointer<-1>, class ScaleN = FlatmmScalePointer<-1>, class ExpertBias = FlatmmScalePointer<-1>>
ck_tile::index_t ck_tile::MoeFlatmmHostArgs< ScaleM, ScaleN, ExpertBias >::TopK
The documentation for this struct was generated from the following file: