CUTLASS: cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK > Struct Template Reference

| | CUTLASS

CUDA Templates for Linear Algebra Subroutines and Solvers |

• Main Page

• Modules

• Namespaces

• Classes

• Files

• Class List

• Class Index

• Class Hierarchy

• Class Members

• cutlass

• epilogue

• threadblock

• DefaultInterleavedThreadMapTensorOp

Classes | Public Types | Static Public Attributes | List of all members

cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK > Struct Template Reference

Defines the optimal thread map for TensorOp accumulator layouts.

#include <default_thread_map_tensor_op.h>

|

Classes

| | struct | Detail | | |

|

Public Types

| | using | ThreadblockShape = ThreadblockShape_ | | | | using | WarpShape = WarpShape_ | | | | using | Element = Element_ | | | | using | Type = InterleavedOutputTileThreadMap< layout::PitchLinearShape< Detail::WarpCount::kM, Detail::WarpCount::kN >, layout::PitchLinearShape< WarpShape::kM/Detail::kTensorOpRows, WarpShape::kN/InterleavedK >, Detail::kThreads, kElementsPerAccess, sizeof_bits< Element >::value > | | |

|

Static Public Attributes

| | static int const | kPartitionsK = PartitionsK | | | | static int const | kElementsPerAccess = ElementsPerAccess | | | | static int const | kInterleavedK = InterleavedK | | |

Member Typedef Documentation

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

| using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::Element = Element_ |

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

| using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::ThreadblockShape = ThreadblockShape_ |

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

| using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::Type = InterleavedOutputTileThreadMap< layout::PitchLinearShape<Detail::WarpCount::kM, Detail::WarpCount::kN>, layout::PitchLinearShape<WarpShape::kM / Detail::kTensorOpRows, WarpShape::kN / InterleavedK>, Detail::kThreads, kElementsPerAccess, sizeof_bits<Element>::value> |

ThreadMap to be used by epilogue::PredicatedTileIterator satisfying concept InterleavedOutputTileThreadMap

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

| using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::WarpShape = WarpShape_ |

Member Data Documentation

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

|

| int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kElementsPerAccess = ElementsPerAccess |

| static |

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

|

| int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kInterleavedK = InterleavedK |

| static |

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>

|

| int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kPartitionsK = PartitionsK |

| static |

---

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

• [default_thread_map_tensor_op.h](default thread map tensor op_8h_source.html)

---

Generated by 1.8.11