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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162

/*******************************************************************************
* Copyright 2019 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.
*******************************************************************************/

// Common for RNN and LSTM cell execution

#include "gpu/generic/sycl/rnn/ref_rnn.hpp"

namespace dnnl {
namespace impl {
namespace gpu {
namespace generic {
namespace sycl {

using namespace dnnl::impl::utils;
using namespace rnn_utils;

status_t ref_rnn_fwd_t::cell_execution(const cell_ctx_t &cell_struct) {

    auto cell_layer = cell_struct.workspace.states_range(cell_struct.lay,
            cell_struct.lay, cell_struct.dir, cell_struct.dir, cell_struct.iter,
            cell_struct.iter);

    auto iter_off = cell_struct.iter == 0
            ? (-1 * (cell_struct.rnn.n_dir - 1) * (cell_struct.rnn.n_iter + 1))
                    - 1
            : cell_struct.iter - 1;
    auto cell_iter = cell_struct.workspace.states_range(cell_struct.lay + 1,
            cell_struct.lay + 1, cell_struct.dir, cell_struct.dir, iter_off,
            iter_off);

    auto scratch_gates = cell_struct.scratch.gates(0);

    auto wei_layer
            = cell_struct.user_data.wei_layer(cell_struct.lay, cell_struct.dir);
    auto wei_iter
            = cell_struct.user_data.wei_iter(cell_struct.lay, cell_struct.dir);

    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, wei_layer,
            cell_layer, scratch_gates, matmul_layer_fwd));

    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, wei_iter,
            cell_iter, scratch_gates, matmul_iter_fwd));

    CHECK(rnn_bias(cell_struct.ctx, cell_struct.rnn.mb, cell_struct.rnn.dhc,
            cell_struct.iter, cell_struct.lay, cell_struct.dir,
            cell_struct.workspace, cell_struct.scratch, cell_struct.user_data));

    return status::success;
}

status_t ref_rnn_bwd_t::cell_execution(const cell_ctx_t &cell_struct) {

    auto wei_layer = cell_struct.user_data.wei_layer(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir);
    auto wei_iter = cell_struct.user_data.wei_iter(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir);

    auto ws_gates = cell_struct.workspace.gates(
            cell_struct.rnn.n_layer - cell_struct.lay, cell_struct.dir,
            cell_struct.rnn.n_iter - cell_struct.iter - 1);

    // take into account reading first layer from end of state
    // subsequent layers at written at forward pass'
    // last layer step location in state
    auto dsl_lay_off = cell_struct.iter == 0
            ? cell_struct.rnn.n_layer * 2 - cell_struct.lay - 1
            : cell_struct.rnn.n_layer - cell_struct.lay - 1;
    // take into account for bidirectional case
    // any timesteps after first one
    // write out from bottom up of timestep state location
    auto dsl_iter_offset = cell_struct.iter == 0 ? 0
            : cell_struct.dir == 1
            ? cell_struct.iter
            : cell_struct.rnn.n_iter - cell_struct.iter + 1;
    auto diff_cell_layer = cell_struct.scratch.diff_states(
            dsl_lay_off, cell_struct.dir, dsl_iter_offset);

    // account for bidirectional cases needing to jumping over
    //  n_iter state blocks
    auto dir_off = cell_struct.dir == 1
            ? cell_struct.iter + 1
            : cell_struct.rnn.n_iter - cell_struct.iter;
    auto diff_cell_iter = cell_struct.scratch.diff_states(
            cell_struct.rnn.n_layer - cell_struct.lay, cell_struct.dir,
            dir_off);

    auto wei_cell_layer = cell_struct.workspace.states_range(
            cell_struct.rnn.n_layer - 1 - cell_struct.lay,
            cell_struct.rnn.n_layer - 1 - cell_struct.lay, cell_struct.dir,
            cell_struct.dir, cell_struct.rnn.n_iter - cell_struct.iter - 1,
            cell_struct.rnn.n_iter - cell_struct.iter - 1);

    auto wci_offset = cell_struct.rnn.n_iter - cell_struct.iter - 2;

    if (cell_struct.rnn.n_dir == 2) {
        wci_offset = cell_struct.rnn.n_iter - cell_struct.iter - 1 == 0
                ? cell_struct.rnn.n_iter - cell_struct.iter
                        - cell_struct.rnn.n_iter - 3
                : cell_struct.rnn.n_iter - cell_struct.iter - 2;
    }

    auto wei_cell_iter = cell_struct.workspace.states_range(
            cell_struct.rnn.n_layer - cell_struct.lay,
            cell_struct.rnn.n_layer - cell_struct.lay, cell_struct.dir,
            cell_struct.dir, wci_offset, wci_offset);

    auto diff_gates = cell_struct.scratch.diff_gates(0);

    CHECK(rnn_bias(cell_struct.ctx, cell_struct.rnn.mb, cell_struct.rnn.dhc,
            cell_struct.iter, cell_struct.lay, cell_struct.dir,
            cell_struct.rnn.n_layer, diff_cell_layer, diff_cell_iter,
            cell_struct.user_data, ws_gates, diff_gates));

    auto dsi_offset = cell_struct.dir == 1
            ? cell_struct.iter + 1
            : cell_struct.rnn.n_iter - cell_struct.iter;

    auto diff_states_layer = cell_struct.scratch.diff_states(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir,
            dsi_offset, 0);

    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, wei_iter,
            diff_gates, diff_states_layer, matmul_iter_bwd));

    auto diff_states_iter = cell_struct.scratch.diff_states(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir,
            dsi_offset, 1);

    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, wei_layer,
            diff_gates, diff_states_iter, matmul_layer_bwd));

    auto diff_wei_layer = cell_struct.user_data.diff_wei_layer(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir);
    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, diff_gates,
            wei_cell_layer, diff_wei_layer, matmul_diff_wei_layer));

    auto diff_wei_iter = cell_struct.user_data.diff_wei_iter(
            cell_struct.rnn.n_layer - cell_struct.lay - 1, cell_struct.dir);
    CHECK(matmul_primitive(cell_struct.engine, cell_struct.ctx, diff_gates,
            wei_cell_iter, diff_wei_iter, matmul_diff_wei_iter));

    return status::success;
}

} // namespace sycl
} // namespace generic
} // namespace gpu
} // namespace impl
} // namespace dnnl