onednn-src 0.1.13

Source of oneAPI Deep Neural Network Library (oneDNN)
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

/*******************************************************************************
* Copyright 2020 Intel Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
*     http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*******************************************************************************/

#ifndef CPU_X64_LRN_LRN_AVX512_NHWC_EXECUTOR_HPP
#define CPU_X64_LRN_LRN_AVX512_NHWC_EXECUTOR_HPP

#include "cpu/x64/lrn/jit_avx512_common_lrn_bwd_nhwc.hpp"
#include "cpu/x64/lrn/jit_avx512_common_lrn_fwd_nhwc.hpp"
#include "cpu/x64/lrn/lrn_executor.hpp"

namespace dnnl {
namespace impl {
namespace cpu {
namespace x64 {
namespace lrn {

template <::dnnl::impl::data_type_t d_type, typename PD_T>
class lrn_avx512_nhwc_executor_fwd_t : public i_lrn_executor_t {
public:
    lrn_avx512_nhwc_executor_fwd_t(const PD_T *pd)
        : ker_(utils::make_unique<
                  lrn::jit_avx512_common_lrn_kernel_fwd_nhwc_t<d_type>>(pd->C(),
                  pd->desc()->prop_kind,
                  pd->desc()->lrn_alpha / pd->desc()->local_size,
                  pd->desc()->lrn_beta, pd->desc()->lrn_k,
                  pd->desc()->local_size))
        , N_(pd->MB())
        , C_(pd->C())
        , H_(pd->H())
        , W_(pd->W()) {}

    using data_t = typename prec_traits_t<d_type>::type;

    status_t create_kernel() override { return ker_->create_kernel(); }

    status_t execute(const exec_ctx_t &ctx) const override {
        status_t status = status::success;
        const auto src = CTX_IN_MEM(const data_t *, DNNL_ARG_SRC);
        const auto dst = CTX_OUT_CLEAN_MEM(data_t *, DNNL_ARG_DST, status);
        CHECK(status);
        const auto ws = CTX_OUT_CLEAN_MEM(data_t *, DNNL_ARG_WORKSPACE, status);
        CHECK(status);

        const auto ker = ker_.get();
        parallel_nd(
                N_, H_ * W_, [= COMPAT_THIS_CAPTURE](dim_t n, dim_t pixel_id) {
            typename lrn::jit_avx512_common_lrn_kernel_fwd_t<
                    d_type>::jit_args_fwd_t args;
            const auto offset = n * C_ * H_ * W_ + pixel_id * C_;
            const auto ws_offset0 = offset * 2;
            const auto ws_offset1 = ws_offset0 + C_;

            args.src = &src[offset];
            args.dst = &dst[offset];
            args.ws0 = ws ? &ws[ws_offset0] : nullptr;
            args.ws1 = ws ? &ws[ws_offset1] : nullptr;

            (*ker)(&args);
        });

        return status::success;
    }

    ~lrn_avx512_nhwc_executor_fwd_t() override = default;

private:
    std::unique_ptr<jit_avx512_common_lrn_kernel_fwd_nhwc_t<d_type>> ker_;
    const dim_t N_;
    const dim_t C_;
    const dim_t H_;
    const dim_t W_;
};
template <::dnnl::impl::data_type_t d_type, typename PD_T>
class lrn_avx512_nhwc_executor_bwd_t : public i_lrn_executor_t {
public:
    lrn_avx512_nhwc_executor_bwd_t(const PD_T *pd)
        : ker_ {utils::make_unique<
                  lrn::jit_avx512_common_lrn_kernel_bwd_nhwc_t<d_type>>(pd->C(),
                  pd->desc()->lrn_alpha / pd->desc()->local_size,
                  pd->desc()->lrn_beta, pd->desc()->local_size)}
        , N_(pd->MB())
        , C_(pd->C())
        , H_(pd->H())
        , W_(pd->W()) {}
    using data_t = typename prec_traits_t<d_type>::type;

    status_t create_kernel() override { return ker_->create_kernel(); }

    status_t execute(const exec_ctx_t &ctx) const override {
        status_t status = status::success;
        auto src = CTX_IN_MEM(data_t *, DNNL_ARG_SRC);
        auto diff_src = CTX_OUT_CLEAN_MEM(data_t *, DNNL_ARG_DIFF_SRC, status);
        CHECK(status);
        auto diff_dst = CTX_IN_MEM(data_t *, DNNL_ARG_DIFF_DST);
        auto ws = CTX_IN_MEM(data_t *, DNNL_ARG_WORKSPACE);

        const auto ker = ker_.get();
        parallel_nd(
                N_, H_ * W_, [= COMPAT_THIS_CAPTURE](dim_t n, dim_t pixel_id) {
            typename lrn::jit_avx512_common_lrn_kernel_bwd_nhwc_t<
                    d_type>::jit_args_bwd_t args;
            const auto offset = n * C_ * H_ * W_ + pixel_id * C_;
            const auto ws_offset0 = offset * 2;
            const auto ws_offset1 = ws_offset0 + C_;

            args.src = &src[offset];
            args.diff_dst = &diff_dst[offset];
            args.ws0 = &ws[ws_offset0];
            args.ws1 = &ws[ws_offset1];
            args.diff_src = &diff_src[offset];

            (*ker)(&args);
        });

        return status::success;
    }

    ~lrn_avx512_nhwc_executor_bwd_t() override = default;

private:
    std::unique_ptr<jit_avx512_common_lrn_kernel_bwd_nhwc_t<d_type>> ker_;
    const dim_t N_;
    const dim_t C_;
    const dim_t H_;
    const dim_t W_;
};

} // namespace lrn
} // namespace x64
} // namespace cpu
} // namespace impl
} // namespace dnnl

#endif