OpenFPM  5.2.0
Project that contain the implementation of distributed structures
Data Structures | Public Types | Public Member Functions | Static Public Member Functions | Data Fields | Static Public Attributes
cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp > Struct Template Reference

AgentReduce implements a stateful abstraction of CUDA thread blocks for participating in device-wide reduction . More...

Data Structures

struct  _TempStorage
 Shared memory type required by this thread block. More...
 
struct  TempStorage
 Alias wrapper allowing storage to be unioned. More...
 

Public Types

enum  {
  BLOCK_THREADS = AgentReducePolicy::BLOCK_THREADS , ITEMS_PER_THREAD = AgentReducePolicy::ITEMS_PER_THREAD , VECTOR_LOAD_LENGTH = CUB_MIN(ITEMS_PER_THREAD, AgentReducePolicy::VECTOR_LOAD_LENGTH) , TILE_ITEMS = BLOCK_THREADS * ITEMS_PER_THREAD ,
  ATTEMPT_VECTORIZATION
}
 Constants.
 
typedef std::iterator_traits< InputIteratorT >::value_type InputT
 The input value type.
 
typedef If<(Equals< typename std::iterator_traits< OutputIteratorT >::value_type, void >::VALUE), typename std::iterator_traits< InputIteratorT >::value_type, typename std::iterator_traits< OutputIteratorT >::value_type >::Type OutputT
 The output value type.
 
typedef CubVector< InputT, AgentReducePolicy::VECTOR_LOAD_LENGTH >::Type VectorT
 Vector type of InputT for data movement.
 
typedef If< IsPointer< InputIteratorT >::VALUE, CacheModifiedInputIterator< AgentReducePolicy::LOAD_MODIFIER, InputT, OffsetT >, InputIteratorT >::Type WrappedInputIteratorT
 Input iterator wrapper type (for applying cache modifier)
 
typedef BlockReduce< OutputT, BLOCK_THREADS, AgentReducePolicy::BLOCK_ALGORITHM > BlockReduceT
 Parameterized BlockReduce primitive.
 

Public Member Functions

__device__ __forceinline__ AgentReduce (TempStorage &temp_storage, InputIteratorT d_in, ReductionOp reduction_op)
 
template<int IS_FIRST_TILE>
__device__ __forceinline__ void ConsumeTile (OutputT &thread_aggregate, OffsetT block_offset, int, Int2Type< true >, Int2Type< false >)
 
template<int IS_FIRST_TILE>
__device__ __forceinline__ void ConsumeTile (OutputT &thread_aggregate, OffsetT block_offset, int, Int2Type< true >, Int2Type< true >)
 
template<int IS_FIRST_TILE, int CAN_VECTORIZE>
__device__ __forceinline__ void ConsumeTile (OutputT &thread_aggregate, OffsetT block_offset, int valid_items, Int2Type< false >, Int2Type< CAN_VECTORIZE >)
 
template<int CAN_VECTORIZE>
__device__ __forceinline__ OutputT ConsumeRange (GridEvenShare< OffsetT > &even_share, Int2Type< CAN_VECTORIZE > can_vectorize)
 Reduce a contiguous segment of input tiles. More...
 
__device__ __forceinline__ OutputT ConsumeRange (OffsetT block_offset, OffsetT block_end)
 Reduce a contiguous segment of input tiles. More...
 
__device__ __forceinline__ OutputT ConsumeTiles (GridEvenShare< OffsetT > &even_share)
 

Static Public Member Functions

template<typename Iterator >
static __device__ __forceinline__ bool IsAligned (Iterator d_in, Int2Type< true >)
 
template<typename Iterator >
static __device__ __forceinline__ bool IsAligned (Iterator, Int2Type< false >)
 

Data Fields

_TempStorage & temp_storage
 Reference to temp_storage.
 
InputIteratorT d_in
 Input data to reduce.
 
WrappedInputIteratorT d_wrapped_in
 Wrapped input data to reduce.
 
ReductionOp reduction_op
 Binary reduction operator.
 

Static Public Attributes

static const CacheLoadModifier LOAD_MODIFIER = AgentReducePolicy::LOAD_MODIFIER
 
static const BlockReduceAlgorithm BLOCK_ALGORITHM = AgentReducePolicy::BLOCK_ALGORITHM
 

Detailed Description

template<typename AgentReducePolicy, typename InputIteratorT, typename OutputIteratorT, typename OffsetT, typename ReductionOp>
struct cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >

AgentReduce implements a stateful abstraction of CUDA thread blocks for participating in device-wide reduction .

Each thread reduces only the values it loads. If FIRST_TILE, this partial reduction is stored into thread_aggregate. Otherwise it is accumulated into thread_aggregate. < Binary reduction operator type having member T operator()(const T &a, const T &b)

Definition at line 99 of file agent_reduce.cuh.

Constructor & Destructor Documentation

◆ AgentReduce()

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
__device__ __forceinline__ cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::AgentReduce ( TempStorage &  temp_storage,
InputIteratorT  d_in,
ReductionOp  reduction_op 
)
inline

Constructor

Parameters
temp_storageReference to temp_storage
d_inInput data to reduce
reduction_opBinary reduction operator

Definition at line 195 of file agent_reduce.cuh.

Member Function Documentation

◆ ConsumeRange() [1/2]

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
template<int CAN_VECTORIZE>
__device__ __forceinline__ OutputT cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeRange ( GridEvenShare< OffsetT > &  even_share,
Int2Type< CAN_VECTORIZE >  can_vectorize 
)
inline

Reduce a contiguous segment of input tiles.

Parameters
even_shareGridEvenShare descriptor
can_vectorizeWhether or not we can vectorize loads

Definition at line 312 of file agent_reduce.cuh.

◆ ConsumeRange() [2/2]

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
__device__ __forceinline__ OutputT cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeRange ( OffsetT  block_offset,
OffsetT  block_end 
)
inline

Reduce a contiguous segment of input tiles.

Parameters
[in]block_offsetThreadblock begin offset (inclusive)
[in]block_endThreadblock end offset (exclusive)

Definition at line 352 of file agent_reduce.cuh.

◆ ConsumeTile() [1/3]

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
template<int IS_FIRST_TILE, int CAN_VECTORIZE>
__device__ __forceinline__ void cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeTile ( OutputT &  thread_aggregate,
OffsetT  block_offset,
int  valid_items,
Int2Type< false >  ,
Int2Type< CAN_VECTORIZE >   
)
inline

Consume a partial tile of input

Parameters
block_offsetThe offset the tile to consume
valid_itemsThe number of valid items in the tile

Definition at line 277 of file agent_reduce.cuh.

◆ ConsumeTile() [2/3]

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
template<int IS_FIRST_TILE>
__device__ __forceinline__ void cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeTile ( OutputT &  thread_aggregate,
OffsetT  block_offset,
int  ,
Int2Type< true >  ,
Int2Type< false >   
)
inline

Consume a full tile of input (non-vectorized)

Parameters
block_offsetThe offset the tile to consume

Definition at line 215 of file agent_reduce.cuh.

◆ ConsumeTile() [3/3]

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
template<int IS_FIRST_TILE>
__device__ __forceinline__ void cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeTile ( OutputT &  thread_aggregate,
OffsetT  block_offset,
int  ,
Int2Type< true >  ,
Int2Type< true >   
)
inline

Consume a full tile of input (vectorized)

Parameters
block_offsetThe offset the tile to consume

Definition at line 238 of file agent_reduce.cuh.

◆ ConsumeTiles()

template<typename AgentReducePolicy , typename InputIteratorT , typename OutputIteratorT , typename OffsetT , typename ReductionOp >
__device__ __forceinline__ OutputT cub::AgentReduce< AgentReducePolicy, InputIteratorT, OutputIteratorT, OffsetT, ReductionOp >::ConsumeTiles ( GridEvenShare< OffsetT > &  even_share)
inline

Reduce a contiguous segment of input tiles

Parameters
[in]even_shareGridEvenShare descriptor

Definition at line 368 of file agent_reduce.cuh.

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