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

/**
 * \file dnn/src/naive/lrn/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/lrn/opr_impl.h"

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

namespace {

using namespace megdnn;
using Param = param::LRN;

template <typename T>
void forward(_megdnn_tensor_in src, _megdnn_tensor_out dst, const Param& param) {
    auto N = src.layout.shape[0], C = src.layout.shape[1], H = src.layout.shape[2],
         W = src.layout.shape[3];
    auto sptr = src.ptr<T>(), dptr = dst.ptr<T>();
    auto half_window = param.n / 2;
    rep(n, N) rep(hw, H * W) {
        rep(dc, C) {
            auto didx = n * C * H * W + dc * H * W + hw;
            size_t c_start = (dc >= half_window ? dc - half_window : 0u);
            size_t c_end = std::min(dc + half_window, C - 1);
            float suma2 = 0.0f;
            for (size_t sc = c_start; sc <= c_end; ++sc) {
                auto sidx = n * C * H * W + sc * H * W + hw;
                suma2 += sqr(sptr[sidx]);
            }
            float multiplicand = std::pow(param.k + param.alpha * suma2, -param.beta);
            float multiplier = sptr[didx];
            dptr[didx] = T(multiplier * multiplicand);
        }
    }
}

template <typename T>
void backward(
        _megdnn_tensor_in src, _megdnn_tensor_in /* dst */, _megdnn_tensor_in diff,
        _megdnn_tensor_out grad, const Param& param) {
    auto N = src.layout.shape[0], C = src.layout.shape[1], H = src.layout.shape[2],
         W = src.layout.shape[3];
    auto half_window = param.n / 2;
    auto k = param.k, alpha = param.alpha, beta = param.beta;
    auto sptr = src.ptr<T>(), hptr = diff.ptr<T>(), gptr = grad.ptr<T>();
    std::fill(gptr, gptr + N * C * H * W, 0);
    rep(n, N) rep(hw, H * W) {
        rep(dc, C) {
            auto didx = n * C * H * W + dc * H * W + hw;
            size_t sc_start = (dc >= half_window ? dc - half_window : 0u);
            size_t sc_end = std::min(dc + half_window, C - 1);
            float tmp = k;
            for (size_t sc = sc_start; sc <= sc_end; ++sc) {
                auto sidx = n * C * H * W + sc * H * W + hw;
                tmp += alpha * sqr(sptr[sidx]);
            }
            for (size_t sc = sc_start; sc <= sc_end; ++sc) {
                auto sidx = n * C * H * W + sc * H * W + hw;
                float res = sptr[didx] * -beta * std::pow(tmp, -beta - 1.0f) * 2.0f *
                            sptr[sidx] * alpha;
                if (sc == dc)
                    res += std::pow(tmp, -beta);
                gptr[sidx] += T(res * hptr[didx]);
            }
        }
    }
}

}  // anonymous namespace

namespace megdnn {
namespace naive {

void LRNForwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_out dst, _megdnn_workspace workspace) {
    check_exec(src.layout, dst.layout, workspace.size);
#define cb(DType)                                                               \
    if (src.layout.dtype == DType()) {                                          \
        MEGDNN_DISPATCH_CPU_KERN_OPR(                                           \
                forward<typename DTypeTrait<DType>::ctype>(src, dst, param())); \
        return;                                                                 \
    }
    MEGDNN_FOREACH_COMPUTING_DTYPE_FLOAT(cb)
#undef cb
    megdnn_assert_internal(0);
}

void LRNBackwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_in dst, _megdnn_tensor_in diff,
        _megdnn_tensor_out grad, _megdnn_workspace workspace) {
    check_exec(src.layout, dst.layout, diff.layout, grad.layout, workspace.size);
#define cb(DType)                                                                 \
    if (src.layout.dtype == DType()) {                                            \
        MEGDNN_DISPATCH_CPU_KERN_OPR(backward<typename DTypeTrait<DType>::ctype>( \
                src, dst, diff, grad, param()));                                  \
        return;                                                                   \
    }
    MEGDNN_FOREACH_COMPUTING_DTYPE_FLOAT(cb)
#undef cb
    megdnn_assert_internal(0);
}

}  // namespace naive
}  // namespace megdnn

// vim: syntax=cpp.doxygen