cub::WarpReduce

template<typename T, int LOGICAL_WARP_THREADS = CUB_PTX_WARP_THREADS, int LEGACY_PTX_ARCH = 0> class WarpReduce

The WarpReduce class provides collective methods for computing a parallel reduction of items partitioned across a CUDA thread warp.

Overview

A reduction (or fold) uses a binary combining operator to compute a single aggregate from a list of input elements.
Supports “logical” warps smaller than the physical warp size (e.g., logical warps of 8 threads)
The number of entrant threads must be an multiple of LOGICAL_WARP_THREADS

Performance Considerations

Uses special instructions when applicable (e.g., warp SHFL instructions)
Uses synchronization-free communication between warp lanes when applicable
Incurs zero bank conflicts for most types
Computation is slightly more efficient (i.e., having lower instruction overhead) for:
- Summation (vs. generic reduction)
- The architecture’s warp size is a whole multiple of LOGICAL_WARP_THREADS

Simple Examples

Every thread in the warp uses the WarpReduce class by first specializing the WarpReduce type, then instantiating an instance with parameters for communication, and finally invoking or more collective member functions.

The code snippet below illustrates four concurrent warp sum reductions within a block of 128 threads (one per each of the 32-thread warps).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for 4 warps
    __shared__ typename WarpReduce::TempStorage temp_storage[4];

    // Obtain one input item per thread
    int thread_data = ...

    // Return the warp-wide sums to each lane0 (threads 0, 32, 64, and 96)
    int warp_id = threadIdx.x / 32;
    int aggregate = WarpReduce(temp_storage[warp_id]).Sum(thread_data);

Suppose the set of input thread_data across the block of threads is {0, 1, 2, 3, ..., 127}. The corresponding output aggregate in threads 0, 32, 64, and 96 will be 496, 1520, 2544, and 3568, respectively (and is undefined in other threads).

The code snippet below illustrates a single warp sum reduction within a block of 128 threads.

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;
    ...

    // Only the first warp performs a reduction
    if (threadIdx.x < 32)
    {
        // Obtain one input item per thread
        int thread_data = ...

        // Return the warp-wide sum to lane0
        int aggregate = WarpReduce(temp_storage).Sum(thread_data);

Suppose the set of input thread_data across the warp of threads is {0, 1, 2, 3, ..., 31}. The corresponding output aggregate in thread0 will be 496 (and is undefined in other threads).

Template Parameters

T – The reduction input/output element type
LOGICAL_WARP_THREADS – [optional] The number of threads per “logical” warp (may be less than the number of hardware warp threads). Default is the warp size of the targeted CUDA compute-capability (e.g., 32 threads for SM20).
LEGACY_PTX_ARCH – [optional] Unused.

Collective constructors

inline WarpReduce(TempStorage &temp_storage)

Collective constructor using the specified memory allocation as temporary storage. Logical warp and lane identifiers are constructed from threadIdx.x.

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

Summation reductions

inline T Sum(T input)

Computes a warp-wide sum in the calling warp. The output is valid in warp lane₀.

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates four concurrent warp sum reductions within a block of 128 threads (one per each of the 32-thread warps).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for 4 warps
    __shared__ typename WarpReduce::TempStorage temp_storage[4];

    // Obtain one input item per thread
    int thread_data = ...

    // Return the warp-wide sums to each lane0
    int warp_id = threadIdx.x / 32;
    int aggregate = WarpReduce(temp_storage[warp_id]).Sum(thread_data);

Suppose the set of input thread_data across the block of threads is {0, 1, 2, 3, ..., 127}. The corresponding output aggregate in threads 0, 32, 64, and 96 will 496, 1520, 2544, and 3568, respectively (and is undefined in other threads).

Parameters: input – [in] Calling thread’s input

inline T Sum(T input, int valid_items)

Computes a partially-full warp-wide sum in the calling warp. The output is valid in warp lane₀.

All threads across the calling warp must agree on the same value for valid_items. Otherwise the result is undefined.

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a sum reduction within a single, partially-full block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(int *d_data, int valid_items)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item per thread if in range
    int thread_data;
    if (threadIdx.x < valid_items)
        thread_data = d_data[threadIdx.x];

    // Return the warp-wide sums to each lane0
    int aggregate = WarpReduce(temp_storage).Sum(
        thread_data, valid_items);

Suppose the input d_data is {0, 1, 2, 3, 4, ... and valid_items is 4. The corresponding output aggregate in lane₀ is 6 (and is undefined in other threads).

Parameters

input – [in] Calling thread’s input
valid_items – [in] Total number of valid items in the calling thread’s logical warp (may be less than LOGICAL_WARP_THREADS)

template<typename FlagT> inline T HeadSegmentedSum(T input, FlagT head_flag)

Computes a segmented sum in the calling warp where segments are defined by head-flags. The sum of each segment is returned to the first lane in that segment (which always includes lane₀).

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a head-segmented warp sum reduction within a block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item and flag per thread
    int thread_data = ...
    int head_flag = ...

    // Return the warp-wide sums to each lane0
    int aggregate = WarpReduce(temp_storage).HeadSegmentedSum(
        thread_data, head_flag);

Suppose the set of input thread_data and head_flag across the block of threads is {0, 1, 2, 3, ..., 31 and is {1, 0, 0, 0, 1, 0, 0, 0, ..., 1, 0, 0, 0, respectively. The corresponding output aggregate in threads 0, 4, 8, etc. will be 6, 22, 38, etc. (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
head_flag – [in] Head flag denoting whether or not input is the start of a new segment

template<typename FlagT> inline T TailSegmentedSum(T input, FlagT tail_flag)

Computes a segmented sum in the calling warp where segments are defined by tail-flags. The sum of each segment is returned to the first lane in that segment (which always includes lane₀).

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a tail-segmented warp sum reduction within a block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item and flag per thread
    int thread_data = ...
    int tail_flag = ...

    // Return the warp-wide sums to each lane0
    int aggregate = WarpReduce(temp_storage).TailSegmentedSum(
        thread_data, tail_flag);

Suppose the set of input thread_data and tail_flag across the block of threads is {0, 1, 2, 3, ..., 31} and is {0, 0, 0, 1, 0, 0, 0, 1, ..., 0, 0, 0, 1}, respectively. The corresponding output aggregate in threads 0, 4, 8, etc. will be 6, 22, 38, etc. (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
tail_flag – [in] Head flag denoting whether or not input is the start of a new segment

Generic reductions

template<typename ReductionOp> inline T Reduce(T input, ReductionOp reduction_op)

Computes a warp-wide reduction in the calling warp using the specified binary reduction functor. The output is valid in warp lane₀.

Supports non-commutative reduction operators

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates four concurrent warp max reductions within a block of 128 threads (one per each of the 32-thread warps).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for 4 warps
    __shared__ typename WarpReduce::TempStorage temp_storage[4];

    // Obtain one input item per thread
    int thread_data = ...

    // Return the warp-wide reductions to each lane0
    int warp_id = threadIdx.x / 32;
    int aggregate = WarpReduce(temp_storage[warp_id]).Reduce(
        thread_data, cuda::maximum<>{});

Suppose the set of input thread_data across the block of threads is {0, 1, 2, 3, ..., 127}. The corresponding output aggregate in threads 0, 32, 64, and 96 will be 31, 63, 95, and 127, respectively (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
reduction_op – [in] Binary reduction operator

template<typename ReductionOp> inline T Reduce(T input, ReductionOp reduction_op, int valid_items)

Computes a partially-full warp-wide reduction in the calling warp using the specified binary reduction functor. The output is valid in warp lane₀.

All threads across the calling warp must agree on the same value for valid_items. Otherwise the result is undefined.

Supports non-commutative reduction operators

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a max reduction within a single, partially-full block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(int *d_data, int valid_items)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item per thread if in range
    int thread_data;
    if (threadIdx.x < valid_items)
        thread_data = d_data[threadIdx.x];

    // Return the warp-wide reductions to each lane0
    int aggregate = WarpReduce(temp_storage).Reduce(
        thread_data, cuda::maximum<>{}, valid_items);

Suppose the input d_data is {0, 1, 2, 3, 4, ... } and valid_items is 4. The corresponding output aggregate in thread0 is 3 (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
reduction_op – [in] Binary reduction operator
valid_items – [in] Total number of valid items in the calling thread’s logical warp (may be less than LOGICAL_WARP_THREADS)

template<typename ReductionOp, typename FlagT> inline T HeadSegmentedReduce(T input, FlagT head_flag, ReductionOp reduction_op)

Computes a segmented reduction in the calling warp where segments are defined by head-flags. The reduction of each segment is returned to the first lane in that segment (which always includes lane₀).

Supports non-commutative reduction operators

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a head-segmented warp max reduction within a block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item and flag per thread
    int thread_data = ...
    int head_flag = ...

    // Return the warp-wide reductions to each lane0
    int aggregate = WarpReduce(temp_storage).HeadSegmentedReduce(
        thread_data, head_flag, cuda::maximum<>{});

Suppose the set of input thread_data and head_flag across the block of threads is {0, 1, 2, 3, ..., 31} and is {1, 0, 0, 0, 1, 0, 0, 0, ..., 1, 0, 0, 0}, respectively. The corresponding output aggregate in threads 0, 4, 8, etc. will be 3, 7, 11, etc. (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
head_flag – [in] Head flag denoting whether or not input is the start of a new segment
reduction_op – [in] Reduction operator

template<typename ReductionOp, typename FlagT> inline T TailSegmentedReduce(T input, FlagT tail_flag, ReductionOp reduction_op)

Computes a segmented reduction in the calling warp where segments are defined by tail-flags. The reduction of each segment is returned to the first lane in that segment (which always includes lane₀).

Supports non-commutative reduction operators

A subsequent __syncwarp() warp-wide barrier should be invoked after calling this method if the collective’s temporary storage (e.g., temp_storage) is to be reused or repurposed.

Snippet

The code snippet below illustrates a tail-segmented warp max reduction within a block of 32 threads (one warp).

#include <cub/cub.cuh>

__global__ void ExampleKernel(...)
{
    // Specialize WarpReduce for type int
    using WarpReduce = cub::WarpReduce<int>;

    // Allocate WarpReduce shared memory for one warp
    __shared__ typename WarpReduce::TempStorage temp_storage;

    // Obtain one input item and flag per thread
    int thread_data = ...
    int tail_flag = ...

    // Return the warp-wide reductions to each lane0
    int aggregate = WarpReduce(temp_storage).TailSegmentedReduce(
        thread_data, tail_flag, cuda::maximum<>{});

Suppose the set of input thread_data and tail_flag across the block of threads is {0, 1, 2, 3, ..., 31} and is {0, 0, 0, 1, 0, 0, 0, 1, ..., 0, 0, 0, 1}, respectively. The corresponding output aggregate in threads 0, 4, 8, etc. will be 3, 7, 11, etc. (and is undefined in other threads).

Template Parameters

ReductionOp – [inferred] Binary reduction operator type having member T operator()(const T &a, const T &b)

Parameters

input – [in] Calling thread’s input
tail_flag – [in] Tail flag denoting whether or not input is the end of the current segment
reduction_op – [in] Reduction operator

struct TempStorage : public Uninitialized<_TempStorage>: The operations exposed by WarpReduce require a temporary memory allocation of this nested type for thread communication. This opaque storage can be allocated directly using the __shared__ keyword. Alternatively, it can be aliased to externally allocated memory (shared or global) or union’d with other storage allocation types to facilitate memory reuse.