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

/*******************************************************************************
* Copyright 2021 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.
*******************************************************************************/

#include "graph/backend/dnnl/kernels/reorder.hpp"
#include "graph/backend/dnnl/patterns/fusions.hpp"
#include "graph/backend/dnnl/patterns/pattern_matcher_pass.hpp"
#include "graph/backend/dnnl/patterns/utils.hpp"

namespace dnnl {
namespace impl {
namespace graph {
namespace dnnl_impl {
namespace pattern {

namespace pm = graph::utils::pm;
using in_edges_t = pm::in_edges_t;
using pb_graph = pm::pb_graph_t;
using FCreatePattern = graph::pass::FCreatePattern;

DNNL_BACKEND_REGISTER_PATTERN_DEF_BEGIN(reorder_fusion)

DNNL_BACKEND_REGISTER_PATTERN_MATCHER_PASS(dnnl, fp_reorder_sum)
        .set_priority(10.1f)
        .set_kind(partition_kind_t::misc_post_ops)
        .set_attr<FCreatePattern>("FCreatePattern",
                [](const std::shared_ptr<pb_graph> &pgraph) -> void {
                    pm::pb_op_t *reorder
                            = pgraph->append_op(graph::op_kind::Reorder);
                    pm::pb_op_t *add = pgraph->append_op(
                            graph::op_kind::Add, {in_edge(0, reorder, 0)});
                    add->append_decision_function([](op_t *graph_op) -> bool {
                        return !graph_op->has_attr(op_attr::auto_broadcast)
                                || graph_op->get_attr<std::string>(
                                           op_attr::auto_broadcast)
                                == "none";
                    });
                })
        .set_attr<FCreateKernel>("FCreateKernel", []() -> kernel_ptr {
            return std::make_shared<float_reorder>();
        });

DNNL_BACKEND_REGISTER_PATTERN_MATCHER_PASS(dnnl, x8_reorder)
        .set_priority(10.1f)
        .set_kind(partition_kind_t::misc_quantized_post_ops)
        .set_attr<FCreatePattern>("FCreatePattern",
                [](const std::shared_ptr<pb_graph> &pgraph) -> void {
                    pm::pb_op_t *dequant
                            = pgraph->append_op(graph::op_kind::Dequantize);
                    dequant->append_decision_function(is_int8_quantization);
                    pm::pb_op_t *reorder = pgraph->append_op(
                            graph::op_kind::Reorder, {in_edge(0, dequant, 0)});
                    auto quant = pgraph->append_op(
                            graph::op_kind::Quantize, {in_edge(0, reorder, 0)});
                    quant->append_decision_function(is_int8_quantization);
                })
        .set_attr<FCreateKernel>("FCreateKernel", []() -> kernel_ptr {
            return std::make_shared<quantized_reorder>();
        });

/*
Currently DNNL Backend doesn't support Post-sum/binary with zero points
on GPU, while CPU supports.
*/
#if DNNL_CPU_RUNTIME != DNNL_RUNTIME_NONE
DNNL_BACKEND_REGISTER_PATTERN_MATCHER_PASS(dnnl, x8_reorder_sum_cpu)
        .set_priority(10.2f)
        .set_engine_kind(engine_kind::cpu)
        .set_kind(partition_kind_t::misc_quantized_post_ops)
        .set_attr<FCreatePattern>("FCreatePattern",
                [](const std::shared_ptr<pb_graph> &pgraph) -> void {
                    pm::pb_op_t *dequant
                            = pgraph->append_op(graph::op_kind::Dequantize);
                    dequant->append_decision_function(is_int8_quantization);
                    pm::pb_op_t *dequant_other
                            = pgraph->append_op(graph::op_kind::Dequantize);
                    dequant_other->append_decision_function(
                            is_int8_quantization);

                    pm::pb_op_t *reorder = pgraph->append_op(
                            graph::op_kind::Reorder, {in_edge(0, dequant, 0)});
                    pm::pb_op_t *add = pgraph->append_op(graph::op_kind::Add,
                            {in_edge(0, reorder, 0),
                                    in_edge(1, dequant_other, 0)});
                    add->append_decision_function([](op_t *graph_op) -> bool {
                        return !graph_op->has_attr(op_attr::auto_broadcast)
                                || graph_op->get_attr<std::string>(
                                           op_attr::auto_broadcast)
                                == "none";
                    });

                    auto quant = pgraph->append_op(
                            graph::op_kind::Quantize, {in_edge(0, add, 0)});
                    quant->append_decision_function(is_int8_quantization);
                })
        .set_attr<FCreateKernel>("FCreateKernel", []() -> kernel_ptr {
            return std::make_shared<quantized_reorder>();
        });
#endif
/*
Currently DNNL Backend doesn't support Post-sum/binary with zero points
on GPU, while CPU supports.
*/
#if DNNL_GPU_RUNTIME != DNNL_RUNTIME_NONE
DNNL_BACKEND_REGISTER_PATTERN_MATCHER_PASS(dnnl, x8_reorder_sum_gpu)
        .set_priority(10.2f)
        .set_engine_kind(engine_kind::gpu)
        .set_kind(partition_kind_t::misc_quantized_post_ops)
        .set_attr<FCreatePattern>("FCreatePattern",
                [](const std::shared_ptr<pb_graph> &pgraph) -> void {
                    pm::pb_op_t *dequant
                            = pgraph->append_op(graph::op_kind::Dequantize);
                    dequant->append_decision_function(is_int8_quantization);
                    pm::pb_op_t *dequant_other
                            = pgraph->append_op(graph::op_kind::Dequantize);
                    dequant_other->append_decision_function(
                            is_int8_quantization);
                    dequant_other->append_decision_function(
                            check_zps_values<0>);

                    pm::pb_op_t *reorder = pgraph->append_op(
                            graph::op_kind::Reorder, {in_edge(0, dequant, 0)});
                    pm::pb_op_t *add = pgraph->append_op(graph::op_kind::Add,
                            {in_edge(0, reorder, 0),
                                    in_edge(1, dequant_other, 0)});
                    add->append_decision_function([](op_t *graph_op) -> bool {
                        return !graph_op->has_attr(op_attr::auto_broadcast)
                                || graph_op->get_attr<std::string>(
                                           op_attr::auto_broadcast)
                                == "none";
                    });

                    auto quant = pgraph->append_op(
                            graph::op_kind::Quantize, {in_edge(0, add, 0)});
                    quant->append_decision_function(is_int8_quantization);
                })
        .set_attr<FCreateKernel>("FCreateKernel", []() -> kernel_ptr {
            return std::make_shared<quantized_reorder>();
        });
#endif
DNNL_BACKEND_REGISTER_PATTERN_DEF_END

} // namespace pattern
} // namespace dnnl_impl
} // namespace graph
} // namespace impl
} // namespace dnnl