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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
|
/*
* Copyright (c) 2023-2024 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#ifndef ACL_TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE_H
#define ACL_TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE_H
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h"
#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h"
#include "tests/framework/Fixture.h"
#include "tests/validation/reference/ActivationLayer.h"
using namespace arm_compute::experimental::dynamic_fusion;
namespace arm_compute
{
namespace test
{
namespace validation
{
template <typename TensorType, typename AccessorType, typename FunctionType, typename T, typename... TArgs>
class DynamicFusionActivationValidationFixture : public framework::Fixture
{
public:
void setup(TensorShape shape, bool fuse, DataType data_type, ActivationLayerInfo act_info, TArgs... args)
{
_fuse = fuse;
_data_type = data_type;
_function = act_info.activation();
_target = compute_target(shape, args...);
_reference = compute_reference(shape, act_info);
}
protected:
std::vector<T> get_boundary_values(T min, T max)
{
// This function will return a vector filled with the following values that can
// represent two partitions derived from equivalent partitioning.
// * Lower partition: min, min + delta, lower quarter (nominal), center - delta
// * Upper partition: center, center + delta, upper quarter (nominal), max - delta, max
const auto delta = is_data_type_float(_data_type) ? T(0.1f) : T(1);
const auto center_value = (min + max) / 2;
const auto lower_quarter = (min + center_value) / 2;
const auto upper_quarter = (center_value + max) / 2;
std::vector<T> boundary_values{};
// To ensure all the inserted values are within the given range after subtracing/adding delta
auto insert_values = [&boundary_values, &min, &max](const std::initializer_list<T> &new_values)
{
for (auto &v : new_values)
{
if (v >= min && v <= max)
{
boundary_values.emplace_back(v);
}
}
};
insert_values({min, static_cast<T>(min + delta), static_cast<T>(lower_quarter),
static_cast<T>(center_value - delta)}); // lower partition
insert_values({static_cast<T>(center_value), static_cast<T>(center_value + delta),
static_cast<T>(upper_quarter), static_cast<T>(max - delta), max}); // upper partition
return boundary_values;
}
template <typename U>
void fill(U &&tensor)
{
float min_bound = 0;
float max_bound = 0;
std::tie(min_bound, max_bound) = get_activation_layer_test_bounds<T>(_function, _data_type);
library->fill_static_values(tensor, get_boundary_values(static_cast<T>(min_bound), static_cast<T>(max_bound)));
}
TensorType compute_target(const TensorShape &shape, TArgs... args)
{
// Create a new workload sketch
CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
GpuWorkloadContext context{&cl_compile_ctx};
GpuWorkloadSketch sketch{&context};
// Create sketch tensors
ITensorInfo *src_info = context.create_tensor_info(TensorInfo(shape, 1, _data_type));
ITensorInfo *dst_info = context.create_tensor_info(TensorInfo(shape, 1, _data_type));
ITensorInfo *ans_0_info = FunctionType::create_op(sketch, src_info, args...);
if (_fuse)
{
ITensorInfo *ans_1_info = FunctionType::create_op(sketch, ans_0_info, args...);
GpuOutput::create_op(sketch, ans_1_info, dst_info);
}
else
{
GpuOutput::create_op(sketch, ans_0_info, dst_info);
}
// Configure runtime
ClWorkloadRuntime runtime;
runtime.configure(sketch);
// Construct user tensors
TensorType t_src{};
TensorType t_dst{};
// Initialize user tensors
t_src.allocator()->init(*src_info);
t_dst.allocator()->init(*dst_info);
// Allocate and fill user tensors
t_src.allocator()->allocate();
t_dst.allocator()->allocate();
fill(AccessorType(t_src));
// Run runtime
runtime.run({&t_src, &t_dst});
return t_dst;
}
SimpleTensor<T> compute_reference(const TensorShape &shape, ActivationLayerInfo act_info)
{
// Create reference
SimpleTensor<T> src{shape, _data_type, 1};
// Fill reference
fill(src);
auto tmp = reference::activation_layer<T>(src, act_info);
if (_fuse)
{
auto dst = reference::activation_layer<T>(tmp, act_info);
return dst;
}
else
{
return tmp;
}
}
protected:
ActivationLayerInfo::ActivationFunction _function{};
bool _fuse{false};
DataType _data_type{};
TensorType _target{};
SimpleTensor<T> _reference{};
};
template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
class DynamicFusionSigmoidValidationFixture
: public DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>
{
public:
void setup(TensorShape shape, bool fuse, DataType data_type)
{
ActivationLayerInfo act_info{ActivationLayerInfo::ActivationFunction::LOGISTIC};
DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, fuse,
data_type, act_info);
}
};
template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
class DynamicFusionTanhValidationFixture
: public DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>
{
public:
void setup(TensorShape shape, bool fuse, DataType data_type)
{
ActivationLayerInfo act_info{ActivationLayerInfo::ActivationFunction::TANH, 1.0f, 1.0f};
DynamicFusionActivationValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, fuse,
data_type, act_info);
}
};
} // namespace validation
} // namespace test
} // namespace arm_compute
#endif // ACL_TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_ACTIVATIONFIXTURE_H
|
