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

/**
 * \file dnn/src/naive/local/opr_impl.h
 * 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 "megdnn/oprs.h"

namespace megdnn {
namespace naive {

class LocalForwardImpl : public LocalForward {
public:
    using LocalForward::LocalForward;
    void exec(
            _megdnn_tensor_in src, _megdnn_tensor_in filter, _megdnn_tensor_out dst,
            _megdnn_workspace workspace) override;
    size_t get_workspace_in_bytes(
            const TensorLayout&, const TensorLayout&, const TensorLayout&) override {
        return 0;
    }

    struct FloatNoncontigBatchKernParam {
        RefPtr src;
        RefPtr filter;
        RefPtr dst;
        size_t n, ic, ih, iw, oc, oh, ow, fh, fw;
        uint32_t ph, pw, sh, sw;
        ptrdiff_t inp_bs, out_bs;  //!< stride for batch of input, output
        void* workspace;
    };
    typedef void (*float_noncontig_batch_kern)(const FloatNoncontigBatchKernParam&);

    FloatNoncontigBatchKernParam make_float_kern_param(
            _megdnn_tensor_in src, _megdnn_tensor_in filter, _megdnn_tensor_out dst,
            _megdnn_workspace workspace) const;

    /*!
     * \brief get function address for computing kernel
     *
     * Used by GroupLocal impl, to compute on input tensors with
     * non-contiguous src/dst batch dimension
     *
     * No need to validate input \p src, \p filter and \p dst are used to
     * give actual input shapes for kerel dispatching.
     */
    virtual float_noncontig_batch_kern dispatch_float_noncontig_batch(
            const TensorLayout& src, const TensorLayout& filter,
            const TensorLayout& dst);

protected:
    //! implement exec() using kernel returned by
    //! dispatch_f32_noncontig_batch()
    void exec_use_float_noncontig_batch(
            _megdnn_tensor_in src, _megdnn_tensor_in filter, _megdnn_tensor_out dst,
            _megdnn_workspace workspace);

private:
    template <bool is_xcorr, typename dtype>
    static void naive_kern(const FloatNoncontigBatchKernParam& param);
};

class LocalBackwardDataImpl : public LocalBackwardData {
public:
    using LocalBackwardData::LocalBackwardData;
    void exec(
            _megdnn_tensor_in filter, _megdnn_tensor_in diff, _megdnn_tensor_out grad,
            _megdnn_workspace workspace) override;
    size_t get_workspace_in_bytes(
            const TensorLayout&, const TensorLayout&, const TensorLayout&) override {
        return 0;
    }
};

class LocalBackwardFilterImpl : public LocalBackwardFilter {
public:
    using LocalBackwardFilter::LocalBackwardFilter;
    void exec(
            _megdnn_tensor_in src, _megdnn_tensor_in diff, _megdnn_tensor_out grad,
            _megdnn_workspace workspace) override;
    size_t get_workspace_in_bytes(
            const TensorLayout&, const TensorLayout&, const TensorLayout&) override {
        return 0;
    }
};

}  // namespace naive
}  // namespace megdnn

//! define local variables for fields in LocalImpl::FloatNoncontigBatchKernParam
#define UNPACK_LOCAL_FLOAT_NONCONTIG_BATCH_KERN_PARAM(_p, _dtype)                   \
    const _dtype* src = static_cast<const _dtype*>(_p.src.get_ptr());               \
    const _dtype* filter = static_cast<const _dtype*>(_p.filter.get_ptr());         \
    _dtype* dst = static_cast<_dtype*>(_p.dst.get_ptr());                           \
    _dtype* workspace = static_cast<_dtype*>(_p.workspace);                         \
    const int N = _p.n, IC = _p.ic, IH = _p.ih, IW = _p.iw, OC = _p.oc, OH = _p.oh, \
              OW = _p.ow, FH = _p.fh, FW = _p.fw;                                   \
    const uint32_t PH = _p.ph, PW = _p.pw, SH = _p.sh, SW = _p.sw;                  \
    const ptrdiff_t INP_BS = _p.inp_bs, OUT_BS = _p.out_bs

// vim: syntax=cpp.doxygen