view release on metacpan or  search on metacpan

xgboost/cub/cub/warp/warp_scan.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
 * The cub::WarpScan class provides [<em>collective</em>](index.html#sec0) methods for computing a parallel prefix scan of items partitioned across a CUDA thread warp.
 */

#pragma once

#include "specializations/warp_scan_shfl.cuh"
#include "specializations/warp_scan_smem.cuh"
#include "../thread/thread_operators.cuh"
#include "../util_arch.cuh"
#include "../util_type.cuh"
#include "../util_namespace.cuh"

/// Optional outer namespace(s)
CUB_NS_PREFIX

/// CUB namespace
namespace cub {

/**
 * \addtogroup WarpModule
 * @{
 */

/**
 * \brief The WarpScan class provides [<em>collective</em>](index.html#sec0) methods for computing a parallel prefix scan of items partitioned across a CUDA thread warp.  ![](warp_scan_logo.png)
 *
 * \tparam T                        The scan input/output element type
 * \tparam LOGICAL_WARP_THREADS     <b>[optional]</b> The number of threads per "logical" warp (may be less than the number of hardware warp threads).  Default is the warp size associated with the CUDA Compute Capability targeted by the compiler (e.g...
 * \tparam PTX_ARCH                 <b>[optional]</b> \ptxversion
 *
 * \par Overview
 * - Given a list of input elements and a binary reduction operator, a [<em>prefix scan</em>](http://en.wikipedia.org/wiki/Prefix_sum)
 *   produces an output list where each element is computed to be the reduction
 *   of the elements occurring earlier in the input list.  <em>Prefix sum</em>
 *   connotes a prefix scan with the addition operator. The term \em inclusive indicates
 *   that the <em>i</em><sup>th</sup> output reduction incorporates the <em>i</em><sup>th</sup> input.
 *   The term \em exclusive indicates the <em>i</em><sup>th</sup> input is not incorporated into
 *   the <em>i</em><sup>th</sup> output reduction.
 * - Supports non-commutative scan operators
 * - Supports "logical" warps smaller than the physical warp size (e.g., a logical warp of 8 threads)
 * - The number of entrant threads must be an multiple of \p LOGICAL_WARP_THREADS
 *
 * \par Performance Considerations
 * - Uses special instructions when applicable (e.g., warp \p SHFL)
 * - 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 (<b><em>vs.</em></b> generic scan)
 *     - The architecture's warp size is a whole multiple of \p LOGICAL_WARP_THREADS
 *
 * \par Simple Examples
 * \warpcollective{WarpScan}
 * \par
 * The code snippet below illustrates four concurrent warp prefix sums within a block of
 * 128 threads (one per each of the 32-thread warps).
 * \par
 * \code
 * #include <cub/cub.cuh>
 *
 * __global__ void ExampleKernel(...)
 * {
 *     // Specialize WarpScan for type int
 *     typedef cub::WarpScan<int> WarpScan;
 *
 *     // Allocate WarpScan shared memory for 4 warps
 *     __shared__ typename WarpScan::TempStorage temp_storage[4];
 *
 *     // Obtain one input item per thread
 *     int thread_data = ...
 *
 *     // Compute warp-wide prefix sums
 *     int warp_id = threadIdx.x / 32;
 *     WarpScan(temp_storage[warp_id]).ExclusiveSum(thread_data, thread_data);
 *
 * \endcode
 * \par
 * Suppose the set of input \p thread_data across the block of threads is <tt>{1, 1, 1, 1, ...}</tt>.
 * The corresponding output \p thread_data in each of the four warps of threads will be
 * <tt>0, 1, 2, 3, ..., 31}</tt>.
 *
 * \par
 * The code snippet below illustrates a single warp prefix sum within a block of
 * 128 threads.
 * \par
 * \code