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
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212

/**
 * \file dnn/src/naive/group_local/opr_impl.cpp
 * 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.
 */
#include "src/naive/group_local/opr_impl.h"

#include <cstring>
#include "src/naive/handle.h"

namespace {

template <typename dtype>
void forward(
        const dtype* src, const dtype* filter, dtype* dst, size_t N, size_t IC,
        size_t IH, size_t IW, size_t FH, size_t FW, size_t OC, size_t OH, size_t OW,
        size_t group, size_t pad_h, size_t pad_w, size_t stride_h, size_t stride_w) {
    size_t ICg = IC / group;
    size_t OCg = OC / group;
    for (size_t n = 0; n < N; ++n)
        for (size_t gid = 0; gid < group; ++gid)
            for (size_t ocg = 0; ocg < OCg; ++ocg)
                for (size_t oh = 0; oh < OH; ++oh)
                    for (size_t ow = 0; ow < OW; ++ow) {
                        float res = 0;
                        size_t oc = gid * OCg + ocg;
                        for (size_t fh = 0; fh < FH; ++fh)
                            for (size_t fw = 0; fw < FW; ++fw)
                                for (size_t icg = 0; icg < ICg; ++icg) {
                                    size_t ih = oh * stride_h - pad_h + fh;
                                    size_t iw = ow * stride_w - pad_w + fw;
                                    size_t ic = gid * ICg + icg;
                                    if (ih < IH && iw < IW) {
                                        auto fval = filter[(
                                                (((((gid * OH + oh) * OW + ow) * ICg +
                                                   icg) * FH +
                                                  fh) * FW +
                                                 fw) * OCg +
                                                ocg)];
                                        auto sval =
                                                src[n * IC * IH * IW + ic * IH * IW +
                                                    ih * IW + iw];
                                        res += fval * sval;
                                    }
                                }
                        dst[n * OC * OH * OW + oc * OH * OW + oh * OW + ow] = res;
                    }
}

void backward_data(
        const float* filter, const float* diff, float* grad, size_t N, size_t IC,
        size_t IH, size_t IW, size_t FH, size_t FW, size_t OC, size_t OH, size_t OW,
        size_t group, size_t pad_h, size_t pad_w, size_t stride_h, size_t stride_w) {
    auto ICg = IC / group;
    auto OCg = OC / group;
    memset(grad, 0, sizeof(float) * N * IC * IH * IW);
    for (size_t n = 0; n < N; ++n)
        for (size_t gid = 0; gid < group; ++gid)
            for (size_t ocg = 0; ocg < OCg; ++ocg)
                for (size_t oh = 0; oh < OH; ++oh)
                    for (size_t ow = 0; ow < OW; ++ow) {
                        size_t oc = gid * OCg + ocg;
                        for (size_t fh = 0; fh < FH; ++fh)
                            for (size_t fw = 0; fw < FW; ++fw)
                                for (size_t icg = 0; icg < ICg; ++icg) {
                                    size_t ih = oh * stride_h - pad_h + fh;
                                    size_t iw = ow * stride_w - pad_w + fw;
                                    size_t ic = gid * ICg + icg;
                                    if (ih < IH && iw < IW) {
                                        auto fval = filter[(
                                                (((((gid * OH + oh) * OW + ow) * ICg +
                                                   icg) * FH +
                                                  fh) * FW +
                                                 fw) * OCg +
                                                ocg)];
                                        auto dval =
                                                diff[n * OC * OH * OW + oc * OH * OW +
                                                     oh * OW + ow];
                                        auto& sval =
                                                grad[n * IC * IH * IW + ic * IH * IW +
                                                     ih * IW + iw];
                                        sval += fval * dval;
                                    }
                                }
                    }
}

void backward_filter(
        const float* src, const float* diff, float* grad, size_t N, size_t IC,
        size_t IH, size_t IW, size_t FH, size_t FW, size_t OC, size_t OH, size_t OW,
        size_t group, size_t pad_h, size_t pad_w, size_t stride_h, size_t stride_w) {
    auto ICg = IC / group;
    auto OCg = OC / group;
    memset(grad, 0, sizeof(float) * group * OH * OW * ICg * FH * FW * OCg);
    for (size_t n = 0; n < N; ++n)
        for (size_t gid = 0; gid < group; ++gid)
            for (size_t ocg = 0; ocg < OCg; ++ocg)
                for (size_t oh = 0; oh < OH; ++oh)
                    for (size_t ow = 0; ow < OW; ++ow) {
                        size_t oc = gid * OCg + ocg;
                        for (size_t fh = 0; fh < FH; ++fh)
                            for (size_t fw = 0; fw < FW; ++fw)
                                for (size_t icg = 0; icg < ICg; ++icg) {
                                    size_t ih = oh * stride_h - pad_h + fh;
                                    size_t iw = ow * stride_w - pad_w + fw;
                                    size_t ic = gid * ICg + icg;
                                    if (ih < IH && iw < IW) {
                                        auto sval =
                                                src[n * IC * IH * IW + ic * IH * IW +
                                                    ih * IW + iw];
                                        auto& fval = grad[(
                                                (((((gid * OH + oh) * OW + ow) * ICg +
                                                   icg) * FH +
                                                  fh) * FW +
                                                 fw) * OCg +
                                                ocg)];
                                        auto dval =
                                                diff[n * OC * OH * OW + oc * OH * OW +
                                                     oh * OW + ow];
                                        fval += sval * dval;
                                    }
                                }
                    }
}

}  // anonymous namespace

namespace megdnn {
namespace naive {

void GroupLocalForwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_in filter, _megdnn_tensor_out dst,
        _megdnn_workspace workspace) {
    check_exec(src.layout, filter.layout, dst.layout, workspace.size);
    auto N = src.layout.shape[0], IC = src.layout.shape[1], IH = src.layout.shape[2],
         IW = src.layout.shape[3];
    auto group = filter.layout.shape[0];
    auto FH = filter.layout.shape[4], FW = filter.layout.shape[5];
    auto OC = dst.layout.shape[1], OH = dst.layout.shape[2], OW = dst.layout.shape[3];
    auto pad_h = param().pad_h;
    auto pad_w = param().pad_w;
    auto stride_h = param().stride_h;
    auto stride_w = param().stride_w;
    if (src.layout.dtype == dtype::Float32() &&
        filter.layout.dtype == dtype::Float32() &&
        dst.layout.dtype == dtype::Float32()) {
        MEGDNN_DISPATCH_CPU_KERN_OPR(
                forward(src.ptr<dt_float32>(), filter.ptr<dt_float32>(),
                        dst.ptr<dt_float32>(), N, IC, IH, IW, FH, FW, OC, OH, OW, group,
                        pad_h, pad_w, stride_h, stride_w));
    } else if (DNN_FLOAT16_SELECT(
                       src.layout.dtype == dtype::Float16() &&
                               filter.layout.dtype == dtype::Float16() &&
                               dst.layout.dtype == dtype::Float16(),
                       false)) {
        DNN_INC_FLOAT16(MEGDNN_DISPATCH_CPU_KERN_OPR(forward(
                src.ptr<dt_float16>(), filter.ptr<dt_float16>(), dst.ptr<dt_float16>(),
                N, IC, IH, IW, FH, FW, OC, OH, OW, group, pad_h, pad_w, stride_h,
                stride_w)););

    } else {
        megdnn_assert_internal(false);
    }
}

void GroupLocalBackwardDataImpl::exec(
        _megdnn_tensor_in filter, _megdnn_tensor_in diff, _megdnn_tensor_out grad,
        _megdnn_workspace workspace) {
    check_exec(filter.layout, diff.layout, grad.layout, workspace.size);
    auto N = grad.layout.shape[0], IC = grad.layout.shape[1], IH = grad.layout.shape[2],
         IW = grad.layout.shape[3];
    auto group = filter.layout.shape[0];
    auto FH = filter.layout.shape[4], FW = filter.layout.shape[5];
    auto OC = diff.layout.shape[1], OH = diff.layout.shape[2],
         OW = diff.layout.shape[3];
    auto pad_h = param().pad_h;
    auto pad_w = param().pad_w;
    auto stride_h = param().stride_h;
    auto stride_w = param().stride_w;
    MEGDNN_DISPATCH_CPU_KERN_OPR(backward_data(
            filter.ptr<dt_float32>(), diff.ptr<dt_float32>(), grad.ptr<dt_float32>(), N,
            IC, IH, IW, FH, FW, OC, OH, OW, group, pad_h, pad_w, stride_h, stride_w));
}

void GroupLocalBackwardFilterImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_in diff, _megdnn_tensor_out grad,
        _megdnn_workspace workspace) {
    check_exec(src.layout, diff.layout, grad.layout, workspace.size);
    auto N = src.layout.shape[0], IC = src.layout.shape[1], IH = src.layout.shape[2],
         IW = src.layout.shape[3];
    auto group = grad.layout.shape[0];
    auto FH = grad.layout.shape[4], FW = grad.layout.shape[5];
    auto OC = diff.layout.shape[1], OH = diff.layout.shape[2],
         OW = diff.layout.shape[3];
    auto pad_h = param().pad_h;
    auto pad_w = param().pad_w;
    auto stride_h = param().stride_h;
    auto stride_w = param().stride_w;
    MEGDNN_DISPATCH_CPU_KERN_OPR(backward_filter(
            src.ptr<dt_float32>(), diff.ptr<dt_float32>(), grad.ptr<dt_float32>(), N,
            IC, IH, IW, FH, FW, OC, OH, OW, group, pad_h, pad_w, stride_h, stride_w));
}

}  // namespace naive
}  // namespace megdnn

// vim: syntax=cpp.doxygen