view release on metacpan or  search on metacpan

xgboost/cub/cub/warp/specializations/warp_scan_shfl.cuh  view on Meta::CPAN

 * Copyright (c) 2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2016, NVIDIA CORPORATION.  All rights reserved.
 * 
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in the
 *       documentation and/or other materials provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the
 *       names of its contributors may be used to endorse or promote products
 *       derived from this software without specific prior written permission.
 * 
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 ******************************************************************************/

/**
 * \file
 * cub::WarpScanShfl provides SHFL-based variants of parallel prefix scan of items partitioned across a CUDA thread warp.
 */

#pragma once

#include "../../thread/thread_operators.cuh"
#include "../../util_type.cuh"
#include "../../util_ptx.cuh"
#include "../../util_namespace.cuh"

/// Optional outer namespace(s)
CUB_NS_PREFIX

/// CUB namespace
namespace cub {

/**
 * \brief WarpScanShfl provides SHFL-based variants of parallel prefix scan of items partitioned across a CUDA thread warp.
 */
template <
    typename    T,                      ///< Data type being scanned
    int         LOGICAL_WARP_THREADS,   ///< Number of threads per logical warp
    int         PTX_ARCH>               ///< The PTX compute capability for which to to specialize this collective
struct WarpScanShfl
{
    //---------------------------------------------------------------------
    // Constants and type definitions
    //---------------------------------------------------------------------

    enum
    {
        /// Whether the logical warp size and the PTX warp size coincide
        IS_ARCH_WARP = (LOGICAL_WARP_THREADS == CUB_WARP_THREADS(PTX_ARCH)),

        /// The number of warp scan steps
        STEPS = Log2<LOGICAL_WARP_THREADS>::VALUE,

        /// The 5-bit SHFL mask for logically splitting warps into sub-segments starts 8-bits up
        SHFL_C = ((-1 << STEPS) & 31) << 8,
    };

    template <typename S>
    struct IntegerTraits
    {
        enum {
            ///Whether the data type is a small (32b or less) integer for which we can use a single SFHL instruction per exchange
            IS_SMALL_UNSIGNED = (Traits<S>::CATEGORY == UNSIGNED_INTEGER) && (sizeof(S) <= sizeof(unsigned int))
        };
    };

    /// Shared memory storage layout type
    struct TempStorage {};


    //---------------------------------------------------------------------
    // Thread fields
    //---------------------------------------------------------------------

    unsigned int lane_id;

    unsigned int member_mask;

    //---------------------------------------------------------------------
    // Construction
    //---------------------------------------------------------------------

    /// Constructor
    __device__ __forceinline__ WarpScanShfl(
        TempStorage &/*temp_storage*/)
    :
        lane_id(IS_ARCH_WARP ?
            LaneId() :
            LaneId() % LOGICAL_WARP_THREADS),
        member_mask(IS_ARCH_WARP ?
             0xffffffff :
             (0xffffffff >> (32 - LOGICAL_WARP_THREADS)) << (LaneId() / LOGICAL_WARP_THREADS))
    {}


    //---------------------------------------------------------------------
    // Inclusive scan steps
    //---------------------------------------------------------------------

    /// Inclusive prefix scan step (specialized for summation across int32 types)
    __device__ __forceinline__ int InclusiveScanStep(
        int             input,              ///< [in] Calling thread's input item.
        cub::Sum        /*scan_op*/,        ///< [in] Binary scan operator
        int             first_lane,         ///< [in] Index of first lane in segment
        int             offset)             ///< [in] Up-offset to pull from
    {
        int output;
        int shfl_c = first_lane | SHFL_C;   // Shuffle control (mask and first-lane)