cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, 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< Shape::kContiguous/Layout::kElementsPerAccess/LdsmShapeContiguous, 1 >
	Number and arrangement of LDSM instructions. More...

Static Public Attributes
static int const	kLdsmOpOuter = Layout::kElementsPerAccess
static int const	kLdsmOpInner = 8
static int const	LdsmShapeStrided
	Shape of one individual LDSM instruction. More...
static int const	LdsmShapeContiguous = 4 / LdsmShapeStrided
static int const	kGroupsPerTile
	Number of groups for each tile. More...

Member Typedef Documentation

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

using cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmIterations = layout::PitchLinearShape< Shape::kContiguous / Layout::kElementsPerAccess / LdsmShapeContiguous, 1>

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

using cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, 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 PartitionsK_>

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kGroupsPerTile

static

Initial value:

=
        Shape::kStrided / InstructionShape::kStrided

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

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kLdsmOpInner = 8

static

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

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::kLdsmOpOuter = Layout::kElementsPerAccess

static

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

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmShapeContiguous = 4 / LdsmShapeStrided

static

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

int const cutlass::gemm::warp::MmaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::TensorOpMultiplicandCongruous< sizeof_bits< Element_ >::value, 64 >, InstructionShape_, OpDelta_, 32, PartitionsK_ >::Policy::LdsmShapeStrided

static

Initial value:

=
        InstructionShape::kStrided / kLdsmOpInner

---

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

• mma_tensor_op_tile_iterator.h