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
163
164
165
166
167
168

/*******************************************************************************
* 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 GPU_INTEL_JIT_ELTWISE_INJECTOR_HPP
#define GPU_INTEL_JIT_ELTWISE_INJECTOR_HPP

#include <assert.h>

#include "common/c_types_map.hpp"
#include "common/utils.hpp"
#include "gpu/intel/jit/generator.hpp"

namespace dnnl {
namespace impl {
namespace gpu {
namespace intel {
namespace jit {

inline bool eltwise_injector_f32_is_supported(alg_kind_t alg) {
    using namespace alg_kind;
    return utils::one_of(alg, eltwise_elu, eltwise_elu_use_dst_for_bwd,
            eltwise_exp, eltwise_exp_use_dst_for_bwd, eltwise_gelu_tanh,
            eltwise_gelu_erf, eltwise_hardsigmoid, eltwise_hardswish,
            eltwise_log, eltwise_mish, eltwise_pow, eltwise_relu,
            eltwise_relu_use_dst_for_bwd, eltwise_soft_relu, eltwise_sqrt,
            eltwise_sqrt_use_dst_for_bwd, eltwise_square, eltwise_swish,
            eltwise_tanh, eltwise_tanh_use_dst_for_bwd, eltwise_abs,
            eltwise_round, eltwise_linear, eltwise_clip, eltwise_clip_v2,
            eltwise_clip_v2_use_dst_for_bwd, eltwise_logistic,
            eltwise_logistic_use_dst_for_bwd, eltwise_stochastic_round,
            eltwise_mx_scale);
}

template <typename ngen_generator_t>
struct eltwise_injector_f32_t {
    eltwise_injector_f32_t(ngen_generator_t *host, alg_kind_t alg, float alpha,
            float beta, float scale,
            const ngen::GRFRange &scratch = ngen::GRFRange(),
            bool is_fwd = true)
        : alg_(alg)
        , alpha_(alpha)
        , beta_(beta)
        , scale_(scale)
        , is_fwd_(is_fwd)
        , h(host)
        , scratch_(scratch) {

        assert(eltwise_injector_f32_is_supported(alg_));
        assert(scratch_.isEmpty() || (scratch_.getLen() >= min_scratch_regs()));
    }

    ngen::HW hw() const { return h->getHardware(); }

    int min_scratch_regs();
    int preferred_scratch_regs();
    void set_scratch(const ngen::GRFRange &scratch) { scratch_ = scratch; }

    void prepare();
    void compute(const ngen::GRF &reg) { compute(ngen::GRFRange(reg, 1)); }
    void compute(const ngen::GRFRange &regs, int seed = -1, int seed_off = 0,
            ngen::DataType = ngen::DataType::invalid);
    void compute(const int *grfs, int ngrf, int seed = -1, int seed_off = 0,
            ngen::DataType = ngen::DataType::invalid);

private:
    const alg_kind_t alg_;
    const float alpha_;
    const float beta_;
    const float scale_;
    const bool is_fwd_;

    ngen_generator_t *h;

    ngen::GRFRange scratch_;

    bool use_tanh_compat() const { return false; }

    int max_batch_size();
    int phase_count(alg_kind_t alg);

    void relu_zero_ns_prepare_fwd();
    void relu_prepare_bwd();
    void abs_prepare_bwd();
    void clip_prepare_bwd();
    void tanh_prepare_fwd();
    void tanh_prepare_fwd_compat();

    void relu_zero_ns_compute_fwd(int simd, const ngen::GRF &r);
    void relu_compute_fwd(int simd, const ngen::GRF &r, int phase, int off);
    void abs_compute_fwd(int simd, const ngen::GRF &r);
    void exp_compute_fwd(int simd, const ngen::GRF &r, int phase);
    void elu_compute_fwd(int simd, const ngen::GRF &r, int phase, int off);
    void gelu_erf_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off, int batch);
    void hardsigmoid_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off);
    void hardswish_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off);
    void log_compute_fwd(int simd, const ngen::GRF &r, int phase);
    void mish_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off, int batch);
    void pow_compute_fwd(int simd, const ngen::GRF &r, int phase, int off);
    void soft_relu_compute_fwd_inner(int simd, const ngen::GRF &input,
            const ngen::GRF &temp, const ngen::GRF &dest, int phase, int off,
            float alpha);
    void soft_relu_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off);
    void sqrt_compute_fwd(int simd, const ngen::GRF &r);
    void square_compute_fwd(int simd, const ngen::GRF &r);
    void round_compute_fwd(int simd, const ngen::GRF &r);
    void sround_compute_fwd(int simd, const ngen::GRF &r, int phase,
            const ngen::Subregister &seed, const ngen::DataType dst_dt,
            int off);
    void mx_scale_compute_fwd(int simd, const ngen::GRF &r, int phase,
            const ngen::Subregister &dst_base, int dst_off,
            const ngen::DataType dst_dt, int off);
    void philox_4x32(
            int simd, const ngen::Subregister &seed, const ngen::GRF &bias);
    void swish_compute_fwd(int simd, const ngen::GRF &r, int phase, int off);
    void tanh_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off, int batch);
    void tanh_compute_fwd_compat(
            int simd, const ngen::GRF &r, int phase, int off, int batch);
    void linear_compute_fwd(int simd, const ngen::GRF &r, int phase);
    void clip_compute_fwd(
            int simd, const ngen::GRF &r, int phase, float alpha, float beta);
    void gelu_tanh_compute_fwd(
            int simd, const ngen::GRF &r, int phase, int off);
    void logistic_compute_fwd(int simd, const ngen::GRF &r, int phase);

    void relu_compute_bwd(int simd, const ngen::GRF &r);
    void abs_compute_bwd(int simd, const ngen::GRF &r, int phase);
    void square_compute_bwd(int simd, const ngen::GRF &r);
    void linear_compute_bwd(int simd, const ngen::GRF &r);
    void clip_compute_bwd(
            int simd, const ngen::GRF &r, int phase, float alpha, float beta);
    void gelu_tanh_compute_bwd(
            int simd, const ngen::GRF &r, int phase, int off, int batch);

    const ngen::InstructionModifier le = ngen_generator_t::le;
    const ngen::InstructionModifier lt = ngen_generator_t::lt;
    const ngen::InstructionModifier ge = ngen_generator_t::ge;
    const ngen::InstructionModifier gt = ngen_generator_t::gt;
    const ngen::InstructionModifier eq = ngen_generator_t::eq;
    const ngen::InstructionModifier sat = ngen_generator_t::sat;
    const ngen::FlagRegister f0 = ngen_generator_t::f0;
};

} // namespace jit
} // namespace intel
} // namespace gpu
} // namespace impl
} // namespace dnnl

#endif // GPU_INTEL_JIT_ELTWISE_INJECTOR_HPP