megenginelite-sys 1.8.2

A safe megenginelite wrapper in Rust
Documentation 
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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150

/**
 * \file dnn/src/cuda/elemwise/kern_wrapper.cuh
 * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
 *
 * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 */

#pragma once

#include "src/common/elemwise/kern_defs.cuh"
#include "src/cuda/elemwise_helper.cuh"

namespace megdnn {
namespace cuda {

template <int arity, class KernImpl, typename enable = void>
struct ElemArithKernWrapper;

template <class KernImpl>
struct ElemArithKernWrapper<
        1, KernImpl,
        typename std::enable_if<
                !std::is_same<typename KernImpl::ctype, dt_int8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_uint8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    ctype* dst;

#if MEGDNN_CC_CUDA
    __device__ void operator()(uint32_t idx, ctype x) { dst[idx] = KernImpl::apply(x); }
#endif
};
template <class KernImpl>
struct ElemArithKernWrapper<
        2, KernImpl,
        typename std::enable_if<
                !std::is_same<typename KernImpl::ctype, dt_int8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_uint8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    ctype* dst;

#if MEGDNN_CC_CUDA
    __device__ void operator()(uint32_t idx, ctype x, ctype y) {
        dst[idx] = KernImpl::apply(x, y);
    }
#endif
};
template <class KernImpl>
struct ElemArithKernWrapper<
        3, KernImpl,
        typename std::enable_if<
                !std::is_same<typename KernImpl::ctype, dt_int8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_uint8>::value &&
                !std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    ctype* dst;

#if MEGDNN_CC_CUDA
    __device__ void operator()(uint32_t idx, ctype x, ctype y, ctype z) {
        dst[idx] = KernImpl::apply(x, y, z);
    }
#endif
};

template <class KernImpl>
struct ElemArithKernWrapper<
        1, KernImpl,
        typename std::enable_if<
                std::is_same<typename KernImpl::ctype, dt_int8>::value ||
                std::is_same<typename KernImpl::ctype, dt_uint8>::value ||
                std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    using VectTypeTrait = elemwise_intl::VectTypeTrait<ctype>;
    typedef typename VectTypeTrait::vect_type vect_type;
    ctype* dst;
#if MEGDNN_CC_CUDA
    __device__ __forceinline__ void operator()(uint32_t idx, ctype x) {
        dst[idx] = KernImpl::apply(x);
    }
    __device__ __forceinline__ void operator()(uint32_t idx, vect_type x) {
        ctype a = KernImpl::apply(x.x);
        ctype b = KernImpl::apply(x.y);
        ctype g = KernImpl::apply(x.z);
        ctype r = KernImpl::apply(x.w);
        *(vect_type*)(&dst[idx]) = VectTypeTrait::make_vector(a, b, g, r);
    }
#endif
};

template <class KernImpl>
struct ElemArithKernWrapper<
        2, KernImpl,
        typename std::enable_if<
                std::is_same<typename KernImpl::ctype, dt_int8>::value ||
                std::is_same<typename KernImpl::ctype, dt_uint8>::value ||
                std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    using VectTypeTrait = elemwise_intl::VectTypeTrait<ctype>;
    typedef typename VectTypeTrait::vect_type vect_type;
    ctype* dst;
#if MEGDNN_CC_CUDA
    __device__ __forceinline__ void operator()(uint32_t idx, ctype x, ctype y) {
        dst[idx] = KernImpl::apply(x, y);
    }
    __device__ __forceinline__ void operator()(uint32_t idx, vect_type x, vect_type y) {
        ctype a = KernImpl::apply(x.x, y.x);
        ctype b = KernImpl::apply(x.y, y.y);
        ctype g = KernImpl::apply(x.z, y.z);
        ctype r = KernImpl::apply(x.w, y.w);
        *(vect_type*)(&dst[idx]) = VectTypeTrait::make_vector(a, b, g, r);
    }
#endif
};

template <class KernImpl>
struct ElemArithKernWrapper<
        3, KernImpl,
        typename std::enable_if<
                std::is_same<typename KernImpl::ctype, dt_int8>::value ||
                std::is_same<typename KernImpl::ctype, dt_uint8>::value ||
                std::is_same<typename KernImpl::ctype, dt_bool>::value>::type> {
    typedef typename KernImpl::ctype ctype;
    using VectTypeTrait = elemwise_intl::VectTypeTrait<ctype>;
    typedef typename VectTypeTrait::vect_type vect_type;
    ctype* dst;
#if MEGDNN_CC_CUDA
    __device__ __forceinline__ void operator()(
            uint32_t idx, ctype x, ctype y, ctype z) {
        dst[idx] = KernImpl::apply(x, y, z);
    }
    __device__ __forceinline__ void operator()(
            uint32_t idx, vect_type x, vect_type y, vect_type z) {
        ctype a = KernImpl::apply(x.x, y.x, z.x);
        ctype b = KernImpl::apply(x.y, y.y, z.y);
        ctype g = KernImpl::apply(x.z, y.z, z.z);
        ctype r = KernImpl::apply(x.w, y.w, z.w);
        *(vect_type*)(&dst[idx]) = VectTypeTrait::make_vector(a, b, g, r);
    }
#endif
};

}  // namespace cuda
}  // namespace megdnn

// vim: ft=cpp syntax=cpp.doxygen