#include <get_cutlass_moe_mm_data.h>

Inheritance diagram for infini::ops::GetCutlassMoeMmData:

Public Member Functions
	GetCutlassMoeMmData (const Tensor topk_ids, const int64_t num_experts, const int64_t n, const int64_t k, Tensor expert_offsets, Tensor problem_sizes1, Tensor problem_sizes2, Tensor input_permutation, Tensor output_permutation, std::optional< Tensor > blockscale_offsets=std::nullopt)

	GetCutlassMoeMmData (const Tensor topk_ids, const int64_t num_experts, const int64_t n, const int64_t k, const bool is_gated, Tensor expert_offsets, Tensor problem_sizes1, Tensor problem_sizes2, Tensor input_permutation, Tensor output_permutation, std::optional< Tensor > blockscale_offsets=std::nullopt)

void	operator() (const Tensor topk_ids, const int64_t num_experts, const int64_t n, const int64_t k, Tensor expert_offsets, Tensor problem_sizes1, Tensor problem_sizes2, Tensor input_permutation, Tensor output_permutation, std::optional< Tensor > blockscale_offsets=std::nullopt) const

virtual void	operator() (const Tensor topk_ids, const int64_t num_experts, const int64_t n, const int64_t k, const bool is_gated, Tensor expert_offsets, Tensor problem_sizes1, Tensor problem_sizes2, Tensor input_permutation, Tensor output_permutation, std::optional< Tensor > blockscale_offsets=std::nullopt) const =0

Public Member Functions inherited from infini::ops::Operator< GetCutlassMoeMmData >
void	operator() (const Handle &handle, const Args &... args)

void	operator() (const Args &... args) const

void	operator() (const Handle &handle, const Args &... args)

void	operator() (const Args &... args) const

Public Member Functions inherited from infini::ops::OperatorBase
virtual	~OperatorBase ()=default

virtual std::size_t	workspace_size_in_bytes () const

void	set_handle (const Handle &handle)

void	set_config (const Config &config)

void	set_stream (void *stream)

void	set_workspace (void *workspace)

void	set_workspace_size_in_bytes (std::size_t workspace_size_in_bytes)

virtual	~OperatorBase ()=default

virtual std::size_t	workspace_size_in_bytes () const

void	set_handle (const Handle &handle)

void	set_config (const Config &config)

void	set_stream (void *stream)

void	set_workspace (void *workspace)

void	set_workspace_size_in_bytes (std::size_t workspace_size_in_bytes)

Protected Member Functions
void	ValidateCallMetadata (const Tensor topk_ids, const int64_t num_experts, const int64_t n, const int64_t k, const bool is_gated, const Tensor expert_offsets, const Tensor problem_sizes1, const Tensor problem_sizes2, const Tensor input_permutation, const Tensor output_permutation, const std::optional< Tensor > blockscale_offsets) const

Protected Attributes
Tensor	topk_ids_metadata_

Tensor	expert_offsets_metadata_

Tensor	problem_sizes1_metadata_

Tensor	problem_sizes2_metadata_

Tensor	input_permutation_metadata_

Tensor	output_permutation_metadata_

std::optional< Tensor >	blockscale_offsets_metadata_

int64_t	num_experts_ {0}

int64_t	n_ {0}

int64_t	k_ {0}

bool	is_gated_ {true}

Tensor::Size	numel_ {0}

Tensor::Size	topk_ {0}

int	device_index_ {0}

Protected Attributes inherited from infini::ops::OperatorBase
std::unique_ptr< Handle >	handle_ptr_

std::unique_ptr< Config >	config_ptr_

void *	stream_ {nullptr}

void *	workspace_ {nullptr}

std::size_t	workspace_size_in_bytes_ {0}

Additional Inherited Members
Static Public Member Functions inherited from infini::ops::Operator< GetCutlassMoeMmData >
static void	clear_cache ()

static void	clear_cache ()

static std::unique_ptr< Operator >	Make (const Config &config, const Tensor tensor, Args &&... args)

static std::unique_ptr< Operator >	Make (const Tensor tensor, Args &&... args)

static std::unique_ptr< Operator >	Make (const Config &config, const std::vector< Tensor > tensors, Args &&... args)

static std::unique_ptr< Operator >	Make (const std::vector< Tensor > tensors, Args &&... args)

static std::unique_ptr< Operator >	Make (const Config &config, const Tensor tensor, Args &&... args)

static std::unique_ptr< Operator >	Make (const Tensor tensor, Args &&... args)

static std::unique_ptr< Operator >	Make (const Config &config, const std::vector< Tensor > tensors, Args &&... args)

static std::unique_ptr< Operator >	Make (const std::vector< Tensor > tensors, Args &&... args)

static void	Call (const Handle &handle, const Config &config, const Args &... args)

static void	Call (const Tensor tensor, const Args &... args)

static auto	Call (const TensorLike &tensor, const Args &... args)

static void	Call (const Handle &handle, const Config &config, const Args &... args)

static void	Call (const Tensor tensor, const Args &... args)

static auto	Call (const TensorLike &tensor, const Args &... args)

static std::vector< std::size_t >	active_implementation_indices (Device::Type dev_type)

static std::vector< std::size_t >	active_implementation_indices (Device::Type dev_type)

Static Protected Attributes inherited from infini::ops::Operator< GetCutlassMoeMmData >
static constexpr Device::Type	device_type_

static constexpr std::size_t	implementation_index_

Constructor & Destructor Documentation

◆ GetCutlassMoeMmData() [1/2]

infini::ops::GetCutlassMoeMmData::GetCutlassMoeMmData	(	const Tensor	topk_ids,
		const int64_t	num_experts,
		const int64_t	n,
		const int64_t	k,
		Tensor	expert_offsets,
		Tensor	problem_sizes1,
		Tensor	problem_sizes2,
		Tensor	input_permutation,
		Tensor	output_permutation,
		std::optional< Tensor >	blockscale_offsets = `std::nullopt`
	)

inline

◆ GetCutlassMoeMmData() [2/2]

infini::ops::GetCutlassMoeMmData::GetCutlassMoeMmData	(	const Tensor	topk_ids,
		const int64_t	num_experts,
		const int64_t	n,
		const int64_t	k,
		const bool	is_gated,
		Tensor	expert_offsets,
		Tensor	problem_sizes1,
		Tensor	problem_sizes2,
		Tensor	input_permutation,
		Tensor	output_permutation,
		std::optional< Tensor >	blockscale_offsets = `std::nullopt`
	)

inline

Member Function Documentation

◆ operator()() [1/2]

virtual void infini::ops::GetCutlassMoeMmData::operator()	(	const Tensor	topk_ids,
		const int64_t	num_experts,
		const int64_t	n,
		const int64_t	k,
		const bool	is_gated,
		Tensor	expert_offsets,
		Tensor	problem_sizes1,
		Tensor	problem_sizes2,
		Tensor	input_permutation,
		Tensor	output_permutation,
		std::optional< Tensor >	blockscale_offsets = `std::nullopt`
	)		const

pure virtual

◆ operator()() [2/2]

void infini::ops::GetCutlassMoeMmData::operator()	(	const Tensor	topk_ids,
		const int64_t	num_experts,
		const int64_t	n,
		const int64_t	k,
		Tensor	expert_offsets,
		Tensor	problem_sizes1,
		Tensor	problem_sizes2,
		Tensor	input_permutation,
		Tensor	output_permutation,
		std::optional< Tensor >	blockscale_offsets = `std::nullopt`
	)		const

inline

◆ ValidateCallMetadata()

void infini::ops::GetCutlassMoeMmData::ValidateCallMetadata	(	const Tensor	topk_ids,
		const int64_t	num_experts,
		const int64_t	n,
		const int64_t	k,
		const bool	is_gated,
		const Tensor	expert_offsets,
		const Tensor	problem_sizes1,
		const Tensor	problem_sizes2,
		const Tensor	input_permutation,
		const Tensor	output_permutation,
		const std::optional< Tensor >	blockscale_offsets
	)		const

inlineprotected

Member Data Documentation

◆ blockscale_offsets_metadata_

std::optional<Tensor> infini::ops::GetCutlassMoeMmData::blockscale_offsets_metadata_

protected

◆ device_index_

int infini::ops::GetCutlassMoeMmData::device_index_ {0}

protected

◆ expert_offsets_metadata_

Tensor infini::ops::GetCutlassMoeMmData::expert_offsets_metadata_

protected

◆ input_permutation_metadata_

Tensor infini::ops::GetCutlassMoeMmData::input_permutation_metadata_

protected

◆ is_gated_

bool infini::ops::GetCutlassMoeMmData::is_gated_ {true}

protected

◆ k_

int64_t infini::ops::GetCutlassMoeMmData::k_ {0}

protected

◆ n_

int64_t infini::ops::GetCutlassMoeMmData::n_ {0}

protected

◆ num_experts_

int64_t infini::ops::GetCutlassMoeMmData::num_experts_ {0}

protected

◆ numel_

Tensor::Size infini::ops::GetCutlassMoeMmData::numel_ {0}

protected

◆ output_permutation_metadata_

Tensor infini::ops::GetCutlassMoeMmData::output_permutation_metadata_

protected

◆ problem_sizes1_metadata_

Tensor infini::ops::GetCutlassMoeMmData::problem_sizes1_metadata_

protected

◆ problem_sizes2_metadata_

Tensor infini::ops::GetCutlassMoeMmData::problem_sizes2_metadata_

protected

◆ topk_

Tensor::Size infini::ops::GetCutlassMoeMmData::topk_ {0}

protected

◆ topk_ids_metadata_

Tensor infini::ops::GetCutlassMoeMmData::topk_ids_metadata_

protected

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

get_cutlass_moe_mm_data.h

Public Member Functions

Protected Member Functions

Protected Attributes

Additional Inherited Members

Constructor & Destructor Documentation

◆ GetCutlassMoeMmData() [1/2]

◆ GetCutlassMoeMmData() [2/2]

Member Function Documentation

◆ operator()() [1/2]

◆ operator()() [2/2]

◆ ValidateCallMetadata()

Member Data Documentation

◆ blockscale_offsets_metadata_

◆ device_index_

◆ expert_offsets_metadata_

◆ input_permutation_metadata_

◆ is_gated_

◆ k_

◆ n_

◆ num_experts_

◆ numel_

◆ output_permutation_metadata_

◆ problem_sizes1_metadata_

◆ problem_sizes2_metadata_

◆ topk_

◆ topk_ids_metadata_