gemv.h

Go to the documentation of this file.

/***************************************************************************************************
 * Copyright (c) 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/cutlass.h"
#include "cutlass/array.h"
#include "cutlass/numeric_types.h"
#include "cutlass/matrix_shape.h"

#include "cutlass/gemm/gemm.h"


namespace cutlass {
namespace gemm {
namespace threadblock {


template <
  class Core_ //< GemvCore
>
class Gemv {
public:
  using Shape = typename Core_::Shape;

  using Operator = typename Core_::Operator;

  using IteratorA = typename Core_::IteratorA;

  using IteratorB = typename Core_::IteratorB;

  using IteratorC = typename Core_::IteratorC;

  using FragmentA = typename IteratorA::Fragment;

  using FragmentB = typename IteratorB::Fragment;

  using FragmentC = typename Operator::FragmentC;

  using ThreadShape = typename Core_::ThreadShape;

public:
  CUTLASS_DEVICE
  Gemv() { }

  CUTLASS_DEVICE
  void operator()(
    GemmCoord const &problem_size,    
    FragmentC &accum,                 
    IteratorA iterator_A,             
    IteratorB iterator_B,             
    FragmentC const &src_accum) {     

    //
    // Prologue
    //

    FragmentA frag_A;
    FragmentB frag_B;
    frag_A.clear();
    frag_B.clear();

    iterator_A.load(frag_A);
    iterator_B.load(frag_B);
    ++iterator_A;
    ++iterator_B;

    //
    // Mainloop
    //
    Operator thread_mma;
    int gemm_k = problem_size.k();

    if (gemm_k < Shape::kK)
    {
      iterator_A.clear_mask();
      iterator_B.clear_mask();
    }

    // iterate over K to accumulate result
    CUTLASS_GEMM_LOOP
    for (; gemm_k > 0; gemm_k -= Shape::kK) {
      thread_mma(accum, frag_A, frag_B, accum);

      iterator_A.load(frag_A);
      iterator_B.load(frag_B);
      ++iterator_A;
      ++iterator_B;

      if (gemm_k < Shape::kK)
      {
        iterator_A.clear_mask();
        iterator_B.clear_mask();
      }
    }

  }
};


} // namespace threadblock
} // namespace gemm
} // namespace cutlass