Template Class BlockRadixRankMatch¶

Defined in File block_radix_rank.hpp

Nested Relationships¶

Nested Types¶

Struct BlockRadixRankMatch::TempStorage

Class Documentation¶

template<int BLOCK_DIM_X, int RADIX_BITS, bool IS_DESCENDING, BlockScanAlgorithm INNER_SCAN_ALGORITHM = BLOCK_SCAN_WARP_SCANS, int BLOCK_DIM_Y = 1, int BLOCK_DIM_Z = 1, int ARCH = 1> class hipcub::BlockRadixRankMatch¶

Radix-rank using match.any

Collective constructors

__device__ inline BlockRadixRankMatch(TempStorage &temp_storage)¶

Collective constructor using the specified memory allocation as temporary storage.

Parameters: temp_storage – Reference to memory allocation having layout type TempStorage

Raking

template<typename UnsignedBits, int KEYS_PER_THREAD, typename DigitExtractorT> __device__ __forceinline__ inline void RankKeys(UnsignedBits (&keys)[KEYS_PER_THREAD], int (&ranks)[KEYS_PER_THREAD], DigitExtractorT digit_extractor)¶

Rank keys.

Parameters

keys – Keys for this tile
ranks – For each key, the local rank within the tile
digit_extractor – The digit extractor

template<typename UnsignedBits, int KEYS_PER_THREAD, typename DigitExtractorT> __device__ __forceinline__ inline void RankKeys(UnsignedBits (&keys)[KEYS_PER_THREAD], int (&ranks)[KEYS_PER_THREAD], DigitExtractorT digit_extractor, int (&exclusive_digit_prefix)[BINS_TRACKED_PER_THREAD])¶

Rank keys. For the lower RADIX_DIGITS threads, digit counts for each digit are provided for the corresponding thread.

Parameters

keys – Keys for this tile
ranks – For each key, the local rank within the tile (out parameter)
digit_extractor – The digit extractor
exclusive_digit_prefix – The exclusive prefix sum for the digits [(threadIdx.x * BINS_TRACKED_PER_THREAD) … (threadIdx.x * BINS_TRACKED_PER_THREAD) + BINS_TRACKED_PER_THREAD - 1]

Public Types

enum [anonymous]¶

Values:

enumerator BINS_TRACKED_PER_THREAD¶: Number of bin-starting offsets tracked per thread.

struct TempStorage : public hipcub::Uninitialized<_TempStorage>¶

{BlockScan}

Public Types

enum [anonymous]¶: Values:

typedef UnitWord<_TempStorage>::DeviceWord DeviceWord¶: Biggest memory-access word that T is a whole multiple of and is not larger than the alignment of T.

Public Functions

__host__ __device__ __forceinline__ inline _TempStorage &Alias()¶: Alias.

Public Members

DeviceWord storage[WORDS]¶: Backing storage.