GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce > Struct Template Reference

GridwiseNormalizationNaiveVariance_mk_to_mk&lt; XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce &gt; Struct Template Reference#

Composable Kernel: ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce > Struct Template Reference
Public Types | Static Public Member Functions | Static Public Attributes | List of all members
ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce > Struct Template Reference

#include <gridwise_normalization_naive_variance.hpp>

Public Types

using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>
using ThreadBufferDimAccessOrder
using ThreadClusterArrangeOrder
using ThreadBufferLengths_M_K = Sequence<MThreadSliceSize, XSrcVectorSize>
using ThreadBufferLengths_M = Sequence<MThreadSliceSize>
using ThreadReduceSrcDesc_M_K
using ThreadReduceDstDesc_M
using BlockwiseSumReduce
using ThreadwiseSumReduce
using PassThroughOp = tensor_operation::element_wise::PassThrough

Static Public Member Functions

static __device__ void Run (const GridDesc_M_K &x_grid_desc_m_k, const GridDesc_M_K &gamma_grid_desc_m_k, const GridDesc_M_K &beta_grid_desc_m_k, const GridDesc_M_K &y_grid_desc_m_k, const GridDesc_M &save_mean_grid_desc_m, const GridDesc_M &save_inv_std_grid_desc_m, index_t num_k_block_tile_iteration, ComputeDataType epsilon, const XDataType *const __restrict__ p_x_global, const GammaDataType *const __restrict__ p_gamma_global, const BetaDataType *const __restrict__ p_beta_global, YDataType *const __restrict__ p_y_global, SaveMeanInvStdDataType *const __restrict__ p_save_mean_global, SaveMeanInvStdDataType *const __restrict__ p_save_inv_std_global, const YElementwiseOperation y_elementwise_op)

Static Public Attributes

static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0)
static constexpr auto thread_cluster_desc
static constexpr auto thread_buffer_desc_m_k
static constexpr auto thread_buffer_desc_m
static constexpr auto I0 = Number<0>{}
static constexpr auto I1 = Number<1>{}
static constexpr auto I2 = Number<2>{}
static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize
static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize
static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize
static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{}

Member Typedef Documentation

◆ BlockwiseSumReduce

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::BlockwiseSumReduce
Initial value:
PartitionedBlockwiseReduction<ComputeDataType,
BlockSize,
ThreadClusterLengths_M_K,
ThreadClusterArrangeOrder,
reduce::Add,
true>
ck::BlockwiseSoftmax::ThreadClusterLengths_M_K
decltype(ThreadClusterDesc_M_K{}.GetLengths()) ThreadClusterLengths_M_K
Definition blockwise_softmax.hpp:69
ck::GridwiseNormalizationBwdData_mk_to_mk::ThreadClusterArrangeOrder
DYThreadBufferDimAccessOrder ThreadClusterArrangeOrder
Definition gridwise_normalization_bwd_data.hpp:86
ck::PartitionedBlockwiseReduction
Definition reduction_functions_blockwise.hpp:28
ck::reduce::Add
Definition reduction_operator.hpp:37

◆ PassThroughOp

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::PassThroughOp = tensor_operation::element_wise::PassThrough

◆ ThreadBufferDimAccessOrder

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadBufferDimAccessOrder
Initial value:
typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type
ck::Sequence
Definition utility/sequence.hpp:43
ck::conditional
Definition utility/functional.hpp:100

◆ ThreadBufferLengths_M

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadBufferLengths_M = Sequence<MThreadSliceSize>

◆ ThreadBufferLengths_M_K

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadBufferLengths_M_K = Sequence<MThreadSliceSize, XSrcVectorSize>

◆ ThreadClusterArrangeOrder

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadClusterArrangeOrder
Initial value:
typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type

◆ ThreadClusterLengths_M_K

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>

◆ ThreadReduceDstDesc_M

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadReduceDstDesc_M
Initial value:
decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})))
ck::Number
integral_constant< index_t, N > Number
Definition number.hpp:12
ck::make_naive_tensor_descriptor_packed
__host__ __device__ constexpr auto make_naive_tensor_descriptor_packed(const Tuple< Lengths... > &lengths)
Definition tensor_descriptor_helper.hpp:101
ck::make_tuple
__host__ __device__ constexpr auto make_tuple(Xs &&... xs)
Definition utility/tuple.hpp:211

◆ ThreadReduceSrcDesc_M_K

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadReduceSrcDesc_M_K
Initial value:
decltype(make_naive_tensor_descriptor_packed(
make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})))

◆ ThreadwiseSumReduce

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
using ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadwiseSumReduce
Initial value:
ThreadwiseReduction<ComputeDataType,
ThreadReduceSrcDesc_M_K,
ThreadReduceDstDesc_M,
reduce::Add,
true>
ck::GridwiseNormalizationNaiveVariance_mk_to_mk::ThreadReduceDstDesc_M
decltype(make_naive_tensor_descriptor_packed(make_tuple(Number< MThreadSliceSize >{}))) ThreadReduceDstDesc_M
Definition gridwise_normalization_naive_variance.hpp:82
ck::GridwiseNormalizationNaiveVariance_mk_to_mk::ThreadReduceSrcDesc_M_K
decltype(make_naive_tensor_descriptor_packed( make_tuple(Number< MThreadSliceSize >{}, Number< XSrcVectorSize >{}))) ThreadReduceSrcDesc_M_K
Definition gridwise_normalization_naive_variance.hpp:80
ck::ThreadwiseReduction
Definition reduction_functions_threadwise.hpp:23

Member Function Documentation

◆ Run()

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
__device__ void ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::Run ( const GridDesc_M_K & x_grid_desc_m_k,
const GridDesc_M_K & gamma_grid_desc_m_k,
const GridDesc_M_K & beta_grid_desc_m_k,
const GridDesc_M_K & y_grid_desc_m_k,
const GridDesc_M & save_mean_grid_desc_m,
const GridDesc_M & save_inv_std_grid_desc_m,
index_t num_k_block_tile_iteration,
ComputeDataType epsilon,
const XDataType *const __restrict__ p_x_global,
const GammaDataType *const __restrict__ p_gamma_global,
const BetaDataType *const __restrict__ p_beta_global,
YDataType *const __restrict__ p_y_global,
SaveMeanInvStdDataType *const __restrict__ p_save_mean_global,
SaveMeanInvStdDataType *const __restrict__ p_save_inv_std_global,
const YElementwiseOperation y_elementwise_op )
inlinestatic

Member Data Documentation

◆ I0

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::I0 = Number<0>{}
staticconstexpr

◆ I1

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::I1 = Number<1>{}
staticconstexpr

◆ I2

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::I2 = Number<2>{}
staticconstexpr

◆ K_BlockTileSize

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
index_t ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::K_BlockTileSize = KThreadClusterSize * KThreadSliceSize
staticconstexpr

◆ K_BlockTileStepSize

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
index_t ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize
staticconstexpr

◆ M_BlockTileSize

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
index_t ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::M_BlockTileSize = MThreadClusterSize * MThreadSliceSize
staticconstexpr

◆ reorder_thread_cluster

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
bool ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::reorder_thread_cluster = (XSrcVectorDim == 0)
staticconstexpr

◆ thread_buffer_desc_m

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::thread_buffer_desc_m
staticconstexpr
Initial value:
=
make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}))

◆ thread_buffer_desc_m_k

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::thread_buffer_desc_m_k
staticconstexpr
Initial value:
= make_naive_tensor_descriptor_packed(
make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}))

◆ thread_cluster_desc

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::thread_cluster_desc
staticconstexpr
Initial value:
=
make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{})
ck::make_cluster_descriptor
__host__ __device__ constexpr auto make_cluster_descriptor(const Lengths &lengths, ArrangeOrder order=typename arithmetic_sequence_gen< 0, Lengths::Size(), 1 >::type{})
Definition tensor_description/cluster_descriptor.hpp:13
ck::GridwiseNormalizationNaiveVariance_mk_to_mk::ThreadClusterArrangeOrder
typename conditional< reorder_thread_cluster, Sequence< 1, 0 >, Sequence< 0, 1 > >::type ThreadClusterArrangeOrder
Definition gridwise_normalization_naive_variance.hpp:66
ck::GridwiseNormalizationNaiveVariance_mk_to_mk::ThreadClusterLengths_M_K
Sequence< MThreadClusterSize, KThreadClusterSize > ThreadClusterLengths_M_K
Definition gridwise_normalization_naive_variance.hpp:61

◆ ThreadBufferNumber

template<typename XDataType, typename GammaDataType, typename BetaDataType, typename YDataType, typename SaveMeanInvStdDataType, typename ComputeDataType, typename YElementwiseOperation, typename GridDesc_M_K, typename GridDesc_M, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorDim, index_t YDstVectorSize, index_t SaveMeanInvStdDstVectorSize, bool SweepOnce>
auto ck::GridwiseNormalizationNaiveVariance_mk_to_mk< XDataType, GammaDataType, BetaDataType, YDataType, SaveMeanInvStdDataType, ComputeDataType, YElementwiseOperation, GridDesc_M_K, GridDesc_M, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorDim, YDstVectorSize, SaveMeanInvStdDstVectorSize, SweepOnce >::ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{}
staticconstexpr
The documentation for this struct was generated from the following file: