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

/*******************************************************************************
* Copyright 2024 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 "common/primitive_attr_quant.hpp"
#include "common/primitive_hashing.hpp"
#include "common/verbose.hpp"

namespace dnnl {
namespace impl {

const quant_entry_t &default_quant_entry() {
    static const quant_entry_t default_quant_entry;
    return default_quant_entry;
}

size_t quant_entry_t::get_hash() const {
    size_t seed = 0;
    seed = hash_combine(seed, mask_);
    seed = hash_combine(seed, static_cast<size_t>(data_type_));
    seed = hash_combine(seed, group_ndims_);
    if (group_ndims_ > 0)
        seed = primitive_hashing::get_array_hash(
                seed, group_dims_, group_ndims_);
    seed = hash_combine(seed, is_host_scalar_);
    seed = hash_combine(seed, qmode_);
    return seed;
}

void quant_entry_t::serialize(serialization_stream_t &sstream) const {
    sstream.append(mask_);
    sstream.append(data_type_);
    sstream.append_array(group_ndims_, group_dims_);
    sstream.append(is_host_scalar_);
    sstream.append(qmode_);
}

quant_entry_t quant_entry_t::deserialize(deserializer_t &d) {
    quant_entry_t e;
    d.pop(e.mask_);
    d.pop(e.data_type_);
    size_t group_ndims;
    d.pop_array(group_ndims, e.group_dims_);
    e.group_ndims_ = static_cast<int>(group_ndims);
    d.pop(e.is_host_scalar_);
    d.pop(e.qmode_);
    return e;
}

std::string quant_entry_t::get_verbose() const {
    std::string s;
    s.append(std::to_string(mask_));
    s.append(":").append(dnnl_dt2str(data_type_));
    if (group_ndims_ > 0) {
        s.append(":")
                .append(std::to_string(group_dims_[0]))
                .append("x")
                .append(std::to_string(group_dims_[1]));
    }
    if (is_host_scalar_) { s.append(":host_scalar"); }
    if (qmode_ != quantization_mode::static_sazp) {
        s.append(":").append(dnnl_quantization_mode2str(qmode_));
    }
    return s;
}

std::ostream &operator<<(std::ostream &ss, const quant_entry_t &e) {
    ss << e.get_verbose();
    return ss;
}

size_t quant_entries_t::get_hash() const {
    size_t seed = 0;
    // Go through scales for all arguments.
    for (const auto &e : entries_) {
        seed = hash_combine(seed, e.first);
        seed = hash_combine(seed, e.second.get_hash());
    }
    return seed;
}

void quant_entries_t::serialize(serialization_stream_t &sstream) const {
    sstream.append(entries_.size());
    for (const auto &e : entries_) {
        sstream.append(e.first);
        sstream.append(e.second);
    }
}

template <typename T>
T deserialize_entries(deserializer_t &d) {
    T entries;
    size_t size = d.pop<size_t>();
    for (size_t i = 0; i < size; i++) {
        int arg = d.pop<int>();
        entries.set(arg, d.pop<quant_entry_t>());
    }
    return entries;
}

std::string quant_entries_t::get_verbose() const {
    std::string s;
    std::string empty_delim, attr_delim = "+";
    std::string delim = empty_delim;
    for (const auto &scale : entries_) {
        const auto &q = scale.second;
        if (q.has_default_values()) continue;

        int arg = scale.first;
        s.append(delim)
                .append(arg2str(arg))
                .append(":")
                .append(q.get_verbose());
        delim = attr_delim;
    }
    return s;
}

scales_t scales_t::deserialize(deserializer_t &d) {
    return deserialize_entries<scales_t>(d);
}

zero_points_t zero_points_t::deserialize(deserializer_t &d) {
    return deserialize_entries<zero_points_t>(d);
}

precomputed_reductions_t precomputed_reductions_t::deserialize(
        deserializer_t &d) {
    return deserialize_entries<precomputed_reductions_t>(d);
}

} // namespace impl
} // namespace dnnl