cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy Struct Reference

Internal structure of iterator - made public to enable introspection.

#include <mma_tensor_op_tile_iterator.h>

Public Types
using	LdsmShape = layout::PitchLinearShape< LdsmShapeContiguous, LdsmShapeStrided >
using	LdsmIterations = layout::PitchLinearShape< 1, Shape::kStrided/kLdsmOpInner/LdsmShape::kStrided >
	Number and arrangement of LDSM instructions. More...

Static Public Attributes
static int const	kLdsmOpOuter = Layout::kElementsPerAccess
static int const	kLdsmOpInner = 8
static int const	LdsmShapeContiguous
	Shape of one individual LDSM instruction. More...
static int const	LdsmShapeStrided
static int const	kGroupsPerTile

Member Typedef Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

using cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmIterations = layout::PitchLinearShape<1, Shape::kStrided / kLdsmOpInner / LdsmShape::kStrided>

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

using cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmShape = layout::PitchLinearShape<LdsmShapeContiguous, LdsmShapeStrided>

Member Data Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kGroupsPerTile

static

Initial value:

= Layout::TileShape::kContiguous /
                                      Layout::kFactor / LdsmShape::kContiguous

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kLdsmOpInner = 8

static

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kLdsmOpOuter = Layout::kElementsPerAccess

static

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmShapeContiguous

static

Initial value:

=
        InstructionShape::kContiguous / kLdsmOpOuter

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int Crosswise, int PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, Crosswise >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmShapeStrided

static

Initial value:

=
        ((4 / LdsmShapeContiguous * kLdsmOpInner) > Shape::kStrided)
            ? (Shape::kStrided / kLdsmOpInner)
            : (4 / LdsmShapeContiguous)

---

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

• mma_tensor_op_tile_iterator.h