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 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.
*******************************************************************************/

#ifndef CPU_X64_JIT_UNI_DW_CONV_KERNEL_UTILS_HPP
#define CPU_X64_JIT_UNI_DW_CONV_KERNEL_UTILS_HPP

#include <memory>
#include "common/c_types_map.hpp"
#include "common/memory_tracking.hpp"
#include "common/nstl.hpp"
#include "common/type_helpers.hpp"
#include "common/utils.hpp"

#include "cpu/x64/jit_generator.hpp"
#include "cpu/x64/jit_primitive_conf.hpp"

#include "cpu/x64/jit_avx512_core_bf16_dw_conv_kernel.hpp"
#include "cpu/x64/jit_avx512_core_f16_dw_conv_kernel.hpp"
#include "cpu/x64/jit_uni_dw_conv_kernel_f32.hpp"

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

template <cpu_isa_t isa, data_type_t kernel_dt>
struct jit_uni_dw_conv_fwd_kernel_t {

    jit_uni_dw_conv_fwd_kernel_t(
            const jit_conv_conf_t &ajcp, const memory_desc_t &dst_md)
        : ker_(utils::make_unique<jit_kernel_t>(ajcp, dst_md)) {}

    status_t create_kernel() {
        if (ker_) return ker_->create_kernel();
        return status::out_of_memory;
    }
    ~jit_uni_dw_conv_fwd_kernel_t() = default;

    static status_t init_conf(jit_conv_conf_t &jcp,
            const convolution_desc_t &cd, memory_desc_t &src_md,
            memory_desc_t &weights_md, memory_desc_t &bias_md,
            memory_desc_t &dst_md, primitive_attr_t &attr);

    static void init_scratchpad(memory_tracking::registrar_t &scratchpad,
            const jit_conv_conf_t &jcp);

    jit_generator_t *ker() const { return ker_.get(); }
    void operator()(const jit_conv_args_t *p) const { (*ker_)(p); }

private:
    constexpr static bool ker_condition_
            = (isa == avx512_core && kernel_dt == data_type::bf16)
            || (isa == avx512_core_fp16 && kernel_dt == data_type::f16);
    using jit_kernel_t = typename utils::conditional<ker_condition_,
            typename utils::conditional<kernel_dt == data_type::bf16,
                    jit_avx512_dw_conv_fwd_kernel_bf16_t,
                    jit_avx512_dw_conv_fwd_kernel_f16_t>::type,
            jit_uni_dw_conv_fwd_kernel_f32_t<isa>>::type;
    std::unique_ptr<jit_kernel_t> ker_;
    DNNL_DISALLOW_COPY_AND_ASSIGN(jit_uni_dw_conv_fwd_kernel_t)
};

template <cpu_isa_t isa, data_type_t kernel_dt>
struct jit_uni_dw_conv_bwd_data_kernel_t {

    jit_uni_dw_conv_bwd_data_kernel_t(const jit_conv_conf_t &ajcp)
        : ker_(utils::make_unique<jit_kernel_t>(ajcp)) {}

    status_t create_kernel() {
        if (ker_) return ker_->create_kernel();
        return status::out_of_memory;
    }
    ~jit_uni_dw_conv_bwd_data_kernel_t() = default;

    static status_t init_conf(jit_conv_conf_t &jcp,
            const convolution_desc_t &cd, memory_desc_t &diff_src_md,
            memory_desc_t &weights_md, memory_desc_t &diff_dst_md);

    static void init_scratchpad(memory_tracking::registrar_t &scratchpad,
            const jit_conv_conf_t &jcp);

    void operator()(const jit_conv_args_t *p) const { (*ker_)(p); }

private:
    using jit_kernel_t = typename utils::conditional<isa == avx512_core
                    && kernel_dt == data_type::bf16,
            jit_avx512_dw_conv_bwd_data_kernel_bf16_t,
            jit_uni_dw_conv_bwd_data_kernel_f32_t<isa>>::type;
    std::unique_ptr<jit_kernel_t> ker_;
    DNNL_DISALLOW_COPY_AND_ASSIGN(jit_uni_dw_conv_bwd_data_kernel_t)
};

template <cpu_isa_t isa, data_type_t kernel_dt>
struct jit_uni_dw_conv_bwd_weights_kernel_t {

    jit_uni_dw_conv_bwd_weights_kernel_t(const jit_conv_conf_t &ajcp)
        : ker_(utils::make_unique<jit_kernel_t>(ajcp)) {}

    status_t create_kernel() {
        if (ker_) return ker_->create_kernel();
        return status::out_of_memory;
    }
    ~jit_uni_dw_conv_bwd_weights_kernel_t() = default;

    static status_t init_conf(jit_conv_conf_t &jcp,
            const convolution_desc_t &cd, memory_desc_t &src_md,
            memory_desc_t &diff_weights_md, memory_desc_t &diff_bias_md,
            memory_desc_t &diff_dst_md, int nthreads);

    static void init_scratchpad(memory_tracking::registrar_t &scratchpad,
            const jit_conv_conf_t &jcp);

    static void partition_nthr_nxc(
            jit_conv_conf_t &jcp, int nthreads, bool prioritize_threading);
    static void balance(jit_conv_conf_t &jcp, int nthreads);

    void operator()(const jit_dw_conv_args_t *p) const { (*ker_)(p); }

private:
    using jit_kernel_t = typename utils::conditional<isa == avx512_core
                    && kernel_dt == data_type::bf16,
            jit_avx512_dw_conv_bwd_weights_kernel_bf16_t,
            jit_uni_dw_conv_bwd_weights_kernel_f32_t<isa>>::type;
    std::unique_ptr<jit_kernel_t> ker_;
    DNNL_DISALLOW_COPY_AND_ASSIGN(jit_uni_dw_conv_bwd_weights_kernel_t)
};

} // namespace x64
} // namespace cpu
} // namespace impl
} // namespace dnnl
#endif /* CPU_X64_JIT_UNI_DW_CONV_KERNEL_UTILS_HPP */