default_thread_map_tensor_op.h

Go to the documentation of this file.

/***************************************************************************************************
 * Copyright (c) 2017-2019, 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 TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 **************************************************************************************************/
#pragma once

#include "predicated_tile_iterator.h"
#include "cutlass/gemm/gemm.h"
#include "cutlass/layout/pitch_linear.h"


namespace cutlass {
namespace epilogue {
namespace threadblock {


template <
  typename ThreadblockShape_,
  typename WarpShape_,
  int PartitionsK,
  typename Element_,
  int ElementsPerAccess
>
struct DefaultThreadMapTensorOp {

  using ThreadblockShape = ThreadblockShape_;
  using WarpShape = WarpShape_;
  static int const kPartitionsK = PartitionsK;
  using Element = Element_;
  static int const kElementsPerAccess = ElementsPerAccess;

  //
  // Definitions
  //

  struct Detail {

    static int const kTensorOpRows = 8;
    static int const kWarpSize = 32;

    static_assert(
      !(ThreadblockShape::kM % WarpShape::kM) &&
      !(ThreadblockShape::kM % WarpShape::kM), "Divisibility");

    using WarpCount = gemm::GemmShape<
      ThreadblockShape::kM / WarpShape::kM,
      ThreadblockShape::kN / WarpShape::kN,
      kPartitionsK
    >;

    static int const kThreads = WarpCount::kCount * kWarpSize;
  };

  //
  // ThreadMap
  //
  
  using Type = OutputTileOptimalThreadMap <
    OutputTileShape<ThreadblockShape::kN, Detail::kTensorOpRows, Detail::WarpCount::kM, 1, 1>,
    OutputTileShape<1, WarpShape::kM / Detail::kTensorOpRows, 1, 1, WarpShape::kM / Detail::kTensorOpRows>,
    Detail::kThreads,
    kElementsPerAccess,
    sizeof_bits<Element>::value
  >;
};


template <typename ThreadblockShape_, typename WarpShape_, int PartitionsK,
          typename Element_, int ElementsPerAccess, int InterleavedK>
struct DefaultInterleavedThreadMapTensorOp {
  using ThreadblockShape = ThreadblockShape_;
  using WarpShape = WarpShape_;
  static int const kPartitionsK = PartitionsK;
  using Element = Element_;
  static int const kElementsPerAccess = ElementsPerAccess;
  static int const kInterleavedK = InterleavedK;

  //
  // Definitions
  //

  struct Detail {
    static int const kTensorOpRows = 8;
    static int const kWarpSize = 32;

    static_assert(!(ThreadblockShape::kM % WarpShape::kM) &&
                      !(ThreadblockShape::kM % WarpShape::kM),
                  "Divisibility");

    using WarpCount =
        gemm::GemmShape<ThreadblockShape::kM / WarpShape::kM,
                        ThreadblockShape::kN / WarpShape::kN, kPartitionsK>;

    static int const kThreads = WarpCount::kCount * kWarpSize;
  };

  //
  // ThreadMap
  //

  using Type = InterleavedOutputTileThreadMap<
      layout::PitchLinearShape<Detail::WarpCount::kM, Detail::WarpCount::kN>,
      layout::PitchLinearShape<WarpShape::kM / Detail::kTensorOpRows,
                               WarpShape::kN / InterleavedK>,
      Detail::kThreads, kElementsPerAccess, sizeof_bits<Element>::value>;
};


} // namespace threadblock
} // namespace epilogue
} // namespace cutlass