CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
wmma_tensor_op_policy.h
Go to the documentation of this file.
1 /***************************************************************************************************
2  * Copyright (c) 2017-2019, 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  **************************************************************************************************/
31 #pragma once
32 
33 #include "cutlass/arch/wmma.h"
34 #include "cutlass/matrix_shape.h"
35 #include "cutlass/layout/matrix.h"
36 
37 #if defined(CUTLASS_ARCH_WMMA_ENABLED)
38 
40 
41 namespace cutlass {
42 namespace epilogue {
43 namespace warp {
44 
46 
48 template <
49  typename WarpShape,
50  typename OperatorShape,
51  typename Layout
52 >
53 struct WmmaTensorOpPolicy;
54 
56 
58 template <
59  typename WarpShape,
60  typename OperatorShape
61 >
62 struct WmmaTensorOpPolicy<WarpShape, OperatorShape, layout::RowMajor> {
63 
65  using OperatorCount = MatrixShape<
66  WarpShape::kM / OperatorShape::kM,
67  WarpShape::kN / OperatorShape::kN
68  >;
69 
70  //
71  // Hard-coded constants regarding Tensor Operations
72  //
73  static int const kElementsPerAccess = 2;
74  static int const kRowsPerIteration = OperatorShape::kM;
75  static int const kWmmaFragmentsPerAccess = 1;
76 
77  //
78  // Derived quantities
79  //
80 
81  // Number of externally visible iterations
82  static int const kIterations = OperatorCount::kRow;
83 
84 };
85 
87 
88 } // namespace warp
89 } // namespace epilogue
90 } // namespace cutlass
91 
93 
94 #endif
95 
Definition: aligned_buffer.h:35
Defines a Shape template for matrix tiles.
Defines layout functions used by TensorRef and derived classes.
Templates exposing architecture support for warp matrix multiply-add (WMMA) operations.