1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
|
/***************************************************************************************************
* Copyright (c) 2017-2018, 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.
*
**************************************************************************************************/
/*! \file
\brief Defies structural properties of single-precision GEMM.
*/
#pragma once
#include <cutlass/gemm/gemm.h>
#include <cutlass/gemm/gemm_epilogue.h>
#include <cutlass/gemm/gemm_epilogue_traits.h>
#include <cutlass/gemm/gemm_global_tile.h>
#include <cutlass/gemm/gemm_shared_tile.h>
#include <cutlass/gemm/gemm_traits.h>
#include <cutlass/gemm/thread_multiply_add.h>
namespace cutlass {
namespace gemm {
////////////////////////////////////////////////////////////////////////////////////////////////////
template <
/// The tile size for the GEMM KxNxM.
typename OutputTile_,
/// The number of accumulators per thread.
typename AccumulatorsPerThread_,
/// The number of scalars per LDG for A.
int kScalarsPerLdgA_ = 1,
/// The number of scalars per LDG for B.
int kScalarsPerLdgB_ = 1>
struct SgemmConfig
: public GemmConfig<
/// The scalar type for A.
float,
/// The scalar type for B.
float,
/// The scalar type for C.
float,
/// The scalar type for D.
float,
/// The tile size for the GEMM KxNxM.
OutputTile_,
/// The functor to do the math in the main loop.
ThreadMultiplyAdd<AccumulatorsPerThread_, Shape<1, 4, 8>, float, float, float>,
/// The number of scalars per LDG for A.
kScalarsPerLdgA_,
/// The number of scalars per STS for A.
kScalarsPerLdgA_,
/// The number of scalars per LDS for A.
4,
/// The number of scalars per LDG for B.
kScalarsPerLdgB_,
/// The number of scalars per STS for B.
kScalarsPerLdgB_,
/// The number of scalars per LDS for B.
4,
/// The number of scalars per LDG for C and STG for D.
1,
/// The number of scalars per STS for D.
4,
/// The number of scalars per LDS for D.
1,
/// The number of stages in shared memory.
2> {};
////////////////////////////////////////////////////////////////////////////////////////////////////
template <
/// The layout for A.
MatrixLayout::Kind kLayoutA_,
/// The layout for B.
MatrixLayout::Kind kLayoutB_,
/// The output tile.
typename OutputTile_ = Shape<8, 128, 128>,
/// The functor to use in the epilogue.
typename EpilogueFunctor_ = LinearScaling<float>,
/// The number of accumulators per thread.
typename AccumulatorsPerThread_ = Shape<8, 8, 8>,
/// The number of floats loaded in one LDG for A.
int kScalarsPerLdgA_ = 1,
/// The number of floats loaded in one LDG for B.
int kScalarsPerLdgB_ = 1,
/// The index.
typename Index_ = int,
/// The SGEMM config.
typename GemmConfig_ =
SgemmConfig<OutputTile_, AccumulatorsPerThread_, kScalarsPerLdgA_, kScalarsPerLdgB_>,
/// The traits class for the epilogue.
typename GemmEpilogueTraits_ =
SimplifiedGemmEpilogueTraits<GemmConfig_, EpilogueFunctor_, Index_> >
struct SgemmTraits : public SimplifiedGemmTraits<
// The layout for A.
kLayoutA_,
// The layout for B.
kLayoutB_,
// The config.
GemmConfig_,
// The epilogue.
GemmEpilogue<GemmEpilogueTraits_>,
// The index.
Index_> {};
////////////////////////////////////////////////////////////////////////////////////////////////////
} // namespace gemm
} // namespace cutlass
|
