device_normalization_bwd_gamma_beta_impl.hpp File Reference

• include
• ck
• tensor_operation
• gpu
• device
• impl

Classes | Namespaces | Functions

device_normalization_bwd_gamma_beta_impl.hpp File Reference

#include <iostream>
#include <vector>
#include "ck/tensor_operation/gpu/device/device_normalization_bwd_gamma_beta.hpp"
#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_bwd_gamma_beta.hpp"
#include "ck/tensor_description/tensor_descriptor.hpp"
#include "ck/tensor_description/tensor_descriptor_helper.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
#include "ck/host_utility/device_prop.hpp"
#include "ck/host_utility/kernel_launch.hpp"

Go to the source code of this file.

Classes
struct	ck::tensor_operation::device::DeviceNormalizationBwdGammaBetaImpl< DYDataType, XDataType, MeanInvStdDataType, ComputeDataType, DGammaDataType, DBetaDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, IsDYFastestDimReduced, DYSrcVectorSize, IsXFastestDimReduced, XSrcVectorSize, IsMeanInvStdFastestDimReduced, MeanInvStdSrcVectorSize, DGammaDstVectorSize, DBetaDstVectorSize >
struct	ck::tensor_operation::device::DeviceNormalizationBwdGammaBetaImpl< DYDataType, XDataType, MeanInvStdDataType, ComputeDataType, DGammaDataType, DBetaDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, IsDYFastestDimReduced, DYSrcVectorSize, IsXFastestDimReduced, XSrcVectorSize, IsMeanInvStdFastestDimReduced, MeanInvStdSrcVectorSize, DGammaDstVectorSize, DBetaDstVectorSize >::Argument
struct	ck::tensor_operation::device::DeviceNormalizationBwdGammaBetaImpl< DYDataType, XDataType, MeanInvStdDataType, ComputeDataType, DGammaDataType, DBetaDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, IsDYFastestDimReduced, DYSrcVectorSize, IsXFastestDimReduced, XSrcVectorSize, IsMeanInvStdFastestDimReduced, MeanInvStdSrcVectorSize, DGammaDstVectorSize, DBetaDstVectorSize >::Invoker

Namespaces
namespace	ck
namespace	ck::tensor_operation
namespace	ck::tensor_operation::device

Functions
template<typename GridwiseReduction, typename DYDataType, typename XDataType, typename MeanInvStdDataType, typename DGammaDataType, typename DBetaDataType, typename GridDesc_M_K, typename GridDesc_M>
__global__ void	ck::tensor_operation::device::kernel_normalization_bwd_gamma_beta (const GridDesc_M_K dy_grid_desc_m_k, const GridDesc_M_K x_grid_desc_m_k, const GridDesc_M_K mean_grid_desc_m_k, const GridDesc_M_K inv_std_grid_desc_m_k, const GridDesc_M dgamma_grid_desc_m, const GridDesc_M dbeta_grid_desc_m, index_t num_k_block_tile_iteration, const DYDataType *const __restrict__ p_dy_global, const XDataType *const __restrict__ p_x_global, const MeanInvStdDataType *const __restrict__ p_mean_global, const MeanInvStdDataType *const __restrict__ p_inv_std_global, DGammaDataType *const __restrict__ p_dgamma_global, DBetaDataType *const __restrict__ p_dbeta_global)