F16xMXF4FlatmmPipelineAgBgCrPolicy Struct Reference

F16xMXF4FlatmmPipelineAgBgCrPolicy Struct Reference#

Composable Kernel: ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy Struct Reference
ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy Struct Reference

#include <mixed_prec_flatmm_pipeline_agmem_bgmem_creg_v1_policy.hpp>

Inheritance diagram for ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy:
ck_tile::UniversalFlatmmPipelineAgBgCrPolicy

Static Public Member Functions

template<typename Problem, typename NativeADramTensorView>
static CK_TILE_HOST_DEVICE constexpr auto TransformF16xF4_ATensorView (const NativeADramTensorView &a_dram_view)
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeF16xF4_ReadALdsBlockDescriptor ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeF16xF4_WriteALdsBlockDescriptor ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeF16xF4_ALDS_TileDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeFp4BFlatDramTileDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeFp4ScaleBFlatDramTileDistribution ()
Static Public Member Functions inherited from ck_tile::UniversalFlatmmPipelineAgBgCrPolicy
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeALdsBlockDescriptor ()
template<typename Problem, typename DataType, index_t MNPerBlock, index_t XPerTile>
static CK_TILE_HOST_DEVICE constexpr auto GetGlobalVectorLoadSize ()
 Get the maximum global memory vector load size.
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto GetVectorSizeA ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto GetVectorSizeB ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr index_t GetSmemSizeA ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr index_t GetSmemSize ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto GetSmemPackA ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto GetKBPerLoad ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeALDS_WarpTileDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeADramTileDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeADramDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeBFlatDramTileDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto MakeShuffledARegBlockDistribution ()
template<typename Problem>
static CK_TILE_HOST_DEVICE constexpr auto GetBlockFlatmm ()

Static Public Attributes

static constexpr auto I0 = number<0>{}
static constexpr auto I1 = number<1>{}
static constexpr auto I2 = number<2>{}
static constexpr index_t KBPerLoad = 32
static constexpr index_t N_Pack = 2
static constexpr index_t K_Pack = 2
Static Public Attributes inherited from ck_tile::UniversalFlatmmPipelineAgBgCrPolicy
static constexpr auto I0 = number<0>{}
static constexpr auto I1 = number<1>{}
static constexpr auto I2 = number<2>{}

Member Function Documentation

◆ MakeF16xF4_ALDS_TileDistribution()

template<typename Problem>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::MakeF16xF4_ALDS_TileDistribution ( )
inlinestaticconstexpr

◆ MakeF16xF4_ReadALdsBlockDescriptor()

template<typename Problem>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::MakeF16xF4_ReadALdsBlockDescriptor ( )
inlinestaticconstexpr

◆ MakeF16xF4_WriteALdsBlockDescriptor()

template<typename Problem>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::MakeF16xF4_WriteALdsBlockDescriptor ( )
inlinestaticconstexpr

◆ MakeFp4BFlatDramTileDistribution()

template<typename Problem>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::MakeFp4BFlatDramTileDistribution ( )
inlinestaticconstexpr

◆ MakeFp4ScaleBFlatDramTileDistribution()

template<typename Problem>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::MakeFp4ScaleBFlatDramTileDistribution ( )
inlinestaticconstexpr

◆ TransformF16xF4_ATensorView()

template<typename Problem, typename NativeADramTensorView>
CK_TILE_HOST_DEVICE constexpr auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::TransformF16xF4_ATensorView ( const NativeADramTensorView & a_dram_view)
inlinestaticconstexpr

Member Data Documentation

◆ I0

auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::I0 = number<0>{}
staticconstexpr

◆ I1

auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::I1 = number<1>{}
staticconstexpr

◆ I2

auto ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::I2 = number<2>{}
staticconstexpr

◆ K_Pack

index_t ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::K_Pack = 2
staticconstexpr

◆ KBPerLoad

index_t ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::KBPerLoad = 32
staticconstexpr

◆ N_Pack

index_t ck_tile::F16xMXF4FlatmmPipelineAgBgCrPolicy::N_Pack = 2
staticconstexpr

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