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 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.
*/
#include "src/cl/CLTensorArgument.h"
#include "ckw/Error.h"
#include "src/cl/CLHelpers.h"
#include "src/cl/CLTensorComponent.h"
#include "src/ITensorArgument.h"
#include "src/ITensorComponent.h"
#include "src/types/TensorComponentType.h"
#include <algorithm>
#include <vector>
namespace ckw
{
CLTensorArgument::CLTensorArgument(const std::string &name, const TensorInfo &info, bool return_dims_by_value)
{
_return_dims_by_value = return_dims_by_value;
_basename = name;
_info = info;
}
CLTensorArgument::~CLTensorArgument() = default;
CLTensorComponent &CLTensorArgument::cl_component(TensorComponentType x)
{
// Return the component if it has already been created.
{
const auto it =
std::find_if(_components_used.begin(), _components_used.end(),
[=](const std::unique_ptr<CLTensorComponent> &item) { return item->component_type() == x; });
if (it != _components_used.end())
{
return **it;
}
}
if (_return_dims_by_value)
{
uint32_t component_type = static_cast<uint32_t>(x);
const bool is_dimension = (component_type & static_cast<uint32_t>(TensorComponentBitmask::Dimension)) != 0;
const bool is_folded_dimensions =
(component_type & static_cast<uint32_t>(TensorComponentBitmask::FoldedDimensions)) != 0;
constexpr auto bitmask_all = static_cast<uint32_t>(TensorComponentIndexBitmask::All);
constexpr auto bitmask_index_0 = static_cast<uint32_t>(TensorComponentIndexBitmask::Index0);
#ifdef COMPUTE_KERNEL_WRITER_ASSERTS_ENABLED
constexpr auto bitmask_index_1 = static_cast<uint32_t>(TensorComponentIndexBitmask::Index1);
constexpr auto bitmask_index_2 = static_cast<uint32_t>(TensorComponentIndexBitmask::Index2);
constexpr auto bitmask_index_3 = static_cast<uint32_t>(TensorComponentIndexBitmask::Index3);
#endif // COMPUTE_KERNEL_WRITER_ASSERTS_ENABLED
// Make sure that the encoding of component type hasn't changed and each nibble is 4 bits apart.
CKW_ASSERT(bitmask_all == (bitmask_index_0 | bitmask_index_1 | bitmask_index_2 | bitmask_index_3));
CKW_ASSERT(bitmask_index_0 == bitmask_index_1 >> 4);
CKW_ASSERT(bitmask_index_1 == bitmask_index_2 >> 4);
CKW_ASSERT(bitmask_index_2 == bitmask_index_3 >> 4);
// If we have a dimension or folded dimensions, we can return the corresponding value if it is not dynamic (not equal to -1)
if (is_dimension == true || is_folded_dimensions == true)
{
component_type = component_type & bitmask_all;
int32_t idx = 1;
for (int32_t i = 0; i < tensor_component_index_max_count; ++i)
{
uint32_t dim_idx = component_type & bitmask_index_0;
if (dim_idx == 0)
{
// Stop at the first nibble containing 0
break;
}
// Subtract - 1. Please refer to the TensorComponentIndexBitmask documentation
dim_idx -= 1;
// Get the dimension value
const int32_t dim_val = _info.shape()[dim_idx];
if (dim_val == kDynamicTensorDimensionValue)
{
// We cannot return the dimension by value if it is dynamic.
// Therefore, force the idx variable to kDynamicTensorDimensionValue and break the loop.
idx = kDynamicTensorDimensionValue;
break;
}
idx *= dim_val;
// Go to the next nibble
component_type >>= 4;
}
if (idx != kDynamicTensorDimensionValue)
{
_components_used.emplace_back(std::make_unique<CLTensorComponent>(*this, x, idx));
return *_components_used.back();
}
}
}
_components_used.emplace_back(std::make_unique<CLTensorComponent>(*this, x));
return *_components_used.back();
}
ITile &CLTensorArgument::component(TensorComponentType x)
{
return cl_component(x);
}
TensorStorageVariable &CLTensorArgument::storage(TensorStorageType x)
{
// Return the storage if it has already been created.
{
const auto it = std::find_if(_storages_used.begin(), _storages_used.end(),
[=](const TensorStorageVariable &item) { return item.type == x; });
if (it != _storages_used.end())
{
return *it;
}
}
TensorStorageVariable t;
t.val = create_storage_name(x);
t.type = x;
_storages_used.emplace_back(t);
return _storages_used.back();
}
std::string CLTensorArgument::create_storage_name(TensorStorageType x) const
{
std::string var_name = _basename;
switch (x)
{
case TensorStorageType::BufferUint8Ptr:
var_name += "_ptr";
break;
case TensorStorageType::Texture2dReadOnly:
case TensorStorageType::Texture2dWriteOnly:
var_name += "_img2d";
break;
default:
CKW_ASSERT_FAILED_MSG("Unsupported tensor storage");
return "";
}
return var_name;
}
std::vector<TensorStorageVariable> CLTensorArgument::storages() const
{
std::vector<TensorStorageVariable> storages;
storages.reserve(_storages_used.size());
std::copy(_storages_used.begin(), _storages_used.end(), std::back_inserter(storages));
return storages;
}
std::vector<const ITensorComponent *> CLTensorArgument::components() const
{
std::vector<const ITensorComponent *> components;
for (const auto &component : _components_used)
{
if (component->is_assignable())
{
components.push_back(component.get());
}
}
return components;
}
} // namespace ckw
|
