tensor_op_policy.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 "cutlass/matrix_shape.h"
#include "cutlass/layout/matrix.h"


namespace cutlass {
namespace epilogue {
namespace warp {


template <
  typename WarpShape,     
  typename OperatorShape, 
  typename Layout         
>
struct TensorOpPolicy; 


template <
  typename WarpShape,           
  typename OperatorShape        
>
struct TensorOpPolicy<WarpShape, OperatorShape, layout::RowMajor> {

  using OperatorCount = MatrixShape<
    WarpShape::kM / OperatorShape::kM,
    WarpShape::kN / OperatorShape::kN
  >;

  //
  // Hard-coded constants regarding Tensor Operations
  //

  static int const kElementsPerAccess = 2;
  static int const kRowsPerIteration = 8;

  //
  // Derived quantities
  //

  // Number of 'externally visible' iterations per actual instruction
  static int const kIterationsPerInstruction = OperatorShape::kM / kRowsPerIteration;

  // Number of externally visible iterations
  static int const kIterations = OperatorCount::kRow * kIterationsPerInstruction;

  static int const kAccumulatorRowStride = kElementsPerAccess;
  static int const kAccumulatorColumnStride = kElementsPerAccess * OperatorCount::kRow * kIterationsPerInstruction;
};


template <
    typename WarpShape,  
    typename OperatorShape,   
    int InterleavedK     
    >
struct TensorOpPolicy<WarpShape, OperatorShape,
                      layout::ColumnMajorInterleaved<InterleavedK> > {
  using OperatorCount = MatrixShape<WarpShape::kM / OperatorShape::kM,
                                    WarpShape::kN / OperatorShape::kN>;

  //
  // Hard-coded constants regarding Tensor Operations
  //

  static int const kElementsPerAccess = 2;
  static int const kRowsPerIteration = 8;

  //
  // Derived quantities
  //

  // Number of 'externally visible' iterations per actual instruction
  static int const kIterationsPerInstruction =
      OperatorShape::kM / kRowsPerIteration;

  // Number of externally visible iterations
  static int const kIterations = WarpShape::kN / InterleavedK *
                                 OperatorCount::kRow *
                                 kIterationsPerInstruction;
};


} // namespace warp
} // namespace epilogue
} // namespace cutlass