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
|
/*
// Copyright (c) 2016 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 "batch_norm_inst.h"
#include "primitive_gpu_base.h"
#include "implementation_map.h"
#include "error_handler.h"
#include "kernel_selector_helper.h"
#include "batch_norm/batch_norm_kernel_base.h"
#include "batch_norm/batch_norm_kernel_selector.h"
#include "eltwise/eltwise_kernel_selector.h"
#include "eltwise/eltwise_kernel_base.h"
namespace cldnn { namespace gpu {
struct batch_norm_gpu : typed_primitive_gpu_impl<batch_norm>
{
using parent = typed_primitive_gpu_impl<batch_norm>;
using parent::parent;
protected:
virtual kernel::kernel_arguments_data get_arguments(typed_primitive_inst<batch_norm>& instance, int32_t) const override
{
kernel::kernel_arguments_data args;
if (!instance.use_global_stats())
{
args.inputs = { &instance.input_memory() };
if (instance.forwad_pass())
args.inputs.push_back(&instance.inv_variance_memory());
}
else
{
args.inputs = { &instance.input_memory(), &instance.mean_memory(), &instance.variance_memory() };
}
args.output = &instance.output_memory();
return args;
}
public:
static primitive_impl* create(const batch_norm_node &arg)
{
if (!arg.use_global_stats())
{
auto norm_params = get_default_params<kernel_selector::batch_norm_params>(arg);
auto norm_optional_params = get_default_optional_params<kernel_selector::batch_norm_optional_params>(arg.get_program());
norm_params.batchNormParams.epsilon = arg.get_primitive()->epsilon;
norm_params.batchNormParams.with_inv_var = arg.forwad_pass();
auto& kernel_selector = kernel_selector::batch_norm_kernel_selector::Instance();
auto best_kernels = kernel_selector.GetBestKernels(norm_params, norm_optional_params);
CLDNN_ERROR_BOOL(arg.id(), "Best_kernel.empty()", best_kernels.empty(), "Cannot find a proper kernel with this arguments");
auto norm = new batch_norm_gpu(arg, best_kernels[0]);
return norm;
}
else
{
auto ew_params = get_default_params<kernel_selector::eltwise_params>(arg);
auto ew_optional_params = get_default_optional_params<kernel_selector::eltwise_optional_params>(arg.get_program());
const float epsilon =
(arg.input().get_output_layout().data_type == data_types::f16) ?
std::max(0.00007f, arg.get_primitive()->epsilon) : // prevent underflow if the epsilon is too small for fp16
arg.get_primitive()->epsilon;
ew_params.inputs.push_back(convert_data_tensor(arg.mean().get_output_layout()));
ew_params.inputs.push_back(convert_data_tensor(arg.variance().get_output_layout()));
ew_params.eltwiseParams.operations.push_back({
{ kernel_selector::eltwise_params::InputType::Buffer(0), kernel_selector::eltwise_params::InputType::Buffer(1) },
kernel_selector::eltwise_mode::SUB });
ew_params.eltwiseParams.operations.push_back({
{ kernel_selector::eltwise_params::InputType::Buffer(2), kernel_selector::eltwise_params::InputType::Scalar(epsilon) },
kernel_selector::eltwise_mode::ADD });
ew_params.eltwiseParams.operations.push_back({
{ kernel_selector::eltwise_params::InputType::Intermediate(1) },
kernel_selector::eltwise_mode::RSQRT });
ew_params.eltwiseParams.operations.push_back({
{ kernel_selector::eltwise_params::InputType::Intermediate(0), kernel_selector::eltwise_params::InputType::Intermediate(2) },
kernel_selector::eltwise_mode::MUL });
ew_params.eltwiseParams.layoutBased = true;
auto& kernel_selector = kernel_selector::eltwise_kernel_selector::Instance();
auto best_kernels = kernel_selector.GetBestKernels(ew_params, ew_optional_params);
CLDNN_ERROR_BOOL(arg.id(), "Best_kernel.empty()", best_kernels.empty(), "Cannot find a proper kernel with this arguments");
auto norm = new batch_norm_gpu(arg, best_kernels[0]);
return norm;
}
};
};
namespace {
struct attach {
attach() {
auto val_fw = batch_norm_gpu::create;
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::yxfb), val_fw);
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::yxfb), val_fw);
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::bfyx), val_fw);
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::bfyx), val_fw);
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::byxf), val_fw);
implementation_map<batch_norm>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::byxf), val_fw);
}
~attach() {}
};
attach attach_impl;
}
} }
|
