CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
simd.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  **************************************************************************************************/
29 #pragma once
30 
31 #include "../array.h"
32 #include "../numeric_types.h"
33 
34 namespace cutlass {
35 namespace arch {
36 
38 
39 //
40 // Element-wise operators
41 //
42 
44 template <typename T, int N>
45 Array<T, N> operator*(Array<T, N> const &a, Array<T, N> const &b) {
46  Array<T, N> d;
48  for (int i = 0; i < N; ++i) {
49  d[i] = a[i] * b[i];
50  }
51  return d;
52 }
53 
55 template <typename T, int N>
56 Array<T, N> operator+(Array<T, N> const &a, Array<T, N> const &b) {
57  Array<T, N> d;
59  for (int i = 0; i < N; ++i) {
60  d[i] = a[i] + b[i];
61  }
62  return d;
63 }
64 
66 template <typename T, int N>
67 Array<T, N> operator-(Array<T, N> const &a, Array<T, N> const &b) {
68  Array<T, N> d;
70  for (int i = 0; i < N; ++i) {
71  d[i] = a[i] - b[i];
72  }
73  return d;
74 }
75 
77 
78 //
79 // Multiply-accumulate operators
80 //
81 
83 template <typename T, int N>
84 Array<T, N> mac(Array<T, N> const &a, Array<T, N> const &b, Array<T, N> const &c) {
85  Array<T, N> d;
87  for (int i = 0; i < N; ++i) {
88  d[i] = a[i] * b[i] + c;
89  }
90  return d;
91 }
92 
94 
95 //
96 // Dot product operator
97 //
98 
100 template <typename Element, typename Accumulator, int N>
101 Accumulator dot(Array<T, N> const &a, Array<T, N> const &b, Accumulator accum) {
103  for (int i = 0; i < N; ++i) {
104  accum += a[i] * b[i];
105  }
106  return accum;
107 }
108 
110 
111 } // namespace arch
112 } // namespace cutlass
113 
115 
116 #include "simd_sm60.h"
117 #include "simd_sm61.h"
118 
Definition: aligned_buffer.h:35
Templates exposing SIMD operators for SM60.
#define CUTLASS_PRAGMA_UNROLL
Definition: cutlass.h:110
CUTLASS_HOST_DEVICE Array< T, N > operator-(Array< T, N > const &a, Array< T, N > const &b)
Definition: simd.h:67
CUTLASS_HOST_DEVICE Array< T, N > operator*(Array< T, N > const &a, Array< T, N > const &b)
Definition: simd.h:45
CUTLASS_HOST_DEVICE Accumulator dot(Array< T, N > const &a, Array< T, N > const &b, Accumulator accum)
Definition: simd.h:101
#define CUTLASS_HOST_DEVICE
Definition: cutlass.h:89
CUTLASS_HOST_DEVICE Array< T, N > operator+(Array< T, N > const &a, Array< T, N > const &b)
Definition: simd.h:56
Templates exposing SIMD operators for SM60.
CUTLASS_HOST_DEVICE Array< T, N > mac(Array< T, N > const &a, Array< T, N > const &b, Array< T, N > const &c)
Definition: simd.h:84