Cutlass
CUDA Templates for Linear Algebra Subroutines and Solvers
dgemm_traits.h
Go to the documentation of this file.
1 /***************************************************************************************************
2  * Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without modification, are permitted
5  * provided that the following conditions are met:
6  * * Redistributions of source code must retain the above copyright notice, this list of
7  * conditions and the following disclaimer.
8  * * Redistributions in binary form must reproduce the above copyright notice, this list of
9  * conditions and the following disclaimer in the documentation and/or other materials
10  * provided with the distribution.
11  * * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used
12  * to endorse or promote products derived from this software without specific prior written
13  * permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
17  * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
19  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
20  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
21  * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
22  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
23  *
24  **************************************************************************************************/
28 #pragma once
29 
30 #include <cutlass/gemm/gemm.h>
37 
38 namespace cutlass {
39 namespace gemm {
40 
42 
43 template <
45  typename OutputTile_,
47  typename AccumulatorsPerThread_,
49  int kScalarsPerLdgA_ = 1,
51  int kScalarsPerLdgB_ = 1>
53  : public GemmConfig<
55  double,
57  double,
59  double,
61  double,
63  OutputTile_,
65  ThreadMultiplyAdd<AccumulatorsPerThread_, Shape<1, 4, 8>, double, double, double>,
67  kScalarsPerLdgA_,
69  kScalarsPerLdgA_,
71  2,
73  kScalarsPerLdgB_,
75  kScalarsPerLdgB_,
77  2,
79  1,
81  2,
83  1,
85  2> {};
86 
88 
89 template <
91  MatrixLayout::Kind kLayoutA_,
93  MatrixLayout::Kind kLayoutB_,
95  typename OutputTile_ = Shape<8, 64, 128>,
97  typename EpilogueFunctor_ = LinearScaling<double>,
99  typename AccumulatorsPerThread_ = Shape<8, 8, 8>,
101  int kScalarsPerLdgA_ = 1,
103  int kScalarsPerLdgB_ = 1,
105  typename Index_ = int,
107  typename GemmConfig_ =
110  typename GemmEpilogueTraits_ =
113  // The layout for A.
114  kLayoutA_,
115  // The layout for B.
116  kLayoutB_,
117  // The config.
118  GemmConfig_,
119  // The epilogue.
120  GemmEpilogue<GemmEpilogueTraits_>,
121  // The index.
122  Index_> {};
123 
125 
126 } // namespace gemm
127 } // namespace cutlass
Definition: convert.h:33
Defines iterators for efficiently loading and storing to global memory.
Defines structural properties of complete GEMM computation.
Template implementing matrix multiply-add operations on fragments.
Implements the epilogue phase of the GEMM kernel that efficiently updates global memory with the comp...
Defines iterators for efficiently loading and storing tiles to and from shared memory.
Definition: gemm_traits.h:79
Definition: dgemm_traits.h:112
Definition: dgemm_traits.h:52
A Shape implementing Layout Concept describing the dimensions of a cube.
Definition: shape.h:64
Definition: gemm_epilogue_traits.h:300
Kind
Definition: matrix_traits.h:36
Functor to compute linear combination of fragments.
Definition: linear_scaling.h:40
Implements a software-pipelined efficient GEMM.
Defines structural properties of the GEMM epilogue.
Definition: gemm_traits.h:723