Compute Library
 22.05
ClElementwiseKernel.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 #ifndef ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H
25 #define ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H
26 
27 #include "src/core/KernelTypes.h"
28 #include "src/core/common/Macros.h"
30 #include "src/gpu/cl/IClKernel.h"
31 
32 namespace arm_compute
33 {
34 namespace opencl
35 {
36 namespace kernels
37 {
38 /** Interface for an element-wise operation kernel
39  *
40  * Element-wise operation is computed by:
41  * @f[ dst(x,y) = OP(src1(x,y), src2(x,y))@f]
42  *
43  * For binary elementwise ops in-place cannot be enabled by passing nullptr to dst, it can only be enabled by passing either src1 or src2 to dst instead.
44  *
45  */
47 {
48 public:
51 
52  // Inherited methods overridden:
53  void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override;
54 
55 protected:
56  /** The name of the operation */
57  virtual std::string name() = 0;
58 
59  /** Configure kernel for a given list of arguments
60  *
61  * @param[in] src1 First source tensor info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/F16/U32/S32/F32.
62  * @param[in] src2 Second source tensor info. Data types supported: same as @p src1.
63  * @param[in] dst Destination tensor info. Data types supported: same as @p src1.
64  *
65  * @return a pair of Status and Window
66  */
67  virtual std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) = 0;
68 
69  /** Generate the build options for the specific kernel
70  *
71  * @reutrn a CLBuildOptions struct
72  */
73  virtual CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) = 0;
74 
75  /** Generate the identifier for tuning
76  *
77  * @reutrn a string
78  */
79  virtual std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) = 0;
80 
81  /** Commmon configure function for element-wise operators with no additional options (e.g., Div, Min, Max, SquaredDiff)
82  *
83  */
84  void configure_common(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst);
85 
86  ActivationLayerInfo _act_info{};
87 };
88 
90 {
91 public:
92  ClLogicalBinaryKernel() = default;
94  /** Function to configure kernel
95  *
96  * @param[in] compile_context The compile context to be used.
97  * @param[in] op Logical binary operation to be executed.
98  * @param[in] src1 First source tensor info. Data types supported: U8.
99  * @param[in] src2 Second source tensor info. Data types supported: same as @p src1.
100  * @param[in] dst Destination tensor info. Data types supported: same as @p src1.
101  */
102  void configure(const ClCompileContext &compile_context, LogicalOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst);
103  /** Static function to check if given info will lead to a valid configuration
104  *
105  * Similar to @ref ClLogicalBinaryKernel::configure()
106  *
107  * @return a status
108  */
109  static Status validate(LogicalOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst);
110 
111 private:
112  // Inherited methods overridden:
113  std::string name() override;
114  std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) override;
115  CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) override;
116  std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) override;
117 
119 };
120 
121 /** Addition operation */
123 {
124 public:
125  ClSaturatedArithmeticKernel() = default;
127  /** Static function to check if given info will lead to a valid configuration of @ref ClSaturatedArithmeticKernel
128  *
129  * @param[in] compile_context The compile context to be used.
130  * @param[in] op Arithmetic operation to be executed.
131  * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
132  * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
133  * @param[in] output Output tensor info. Data types supported: Same as @p input1.
134  * @param[in] policy Policy to use to handle overflow.
135  * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
136  */
137  void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy,
138  const ActivationLayerInfo &act_info = ActivationLayerInfo());
139 
140  /** Static function to check if given info will lead to a valid configuration
141  *
142  * Similar to @ref ClSaturatedArithmeticKernel::configure()
143  *
144  * @return a status
145  */
146  static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy,
147  const ActivationLayerInfo &act_info = ActivationLayerInfo());
148 
149 protected:
150  // Inherited methods overridden:
151  std::string name() override;
152  std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) override;
153  CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
154  std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) override;
155 
156 private:
157  ConvertPolicy _policy{};
158  ArithmeticOperation _op{};
159 };
160 
162 {
163 public:
164  ClArithmeticKernel() = default;
166 
167  /** Static function to check if given info will lead to a valid configuration of @ref ClArithmeticKernel
168  *
169  * @param[in] compile_context The compile context to be used.
170  * @param[in] op Arithmetic operation to be executed.
171  * @param[in] src1 First source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
172  * @param[in] src2 Second source tensor info. Data types supported: same as @p src1.
173  * @param[in] dst Destination tensor info. Data types supported: same as @p src1.
174  * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
175  */
176  void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst,
177  const ActivationLayerInfo &act_info = ActivationLayerInfo());
178 
179  /** Static function to check if given info will lead to a valid configuration
180  *
181  * Similar to @ref ClArithmeticKernel::configure()
182  *
183  * @return a status
184  */
185  static Status validate(ArithmeticOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info = ActivationLayerInfo());
186 
187 protected:
188  // Inherited methods overridden:
189  std::string name() override;
190  std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) override;
191  CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) override;
192  std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) override;
193 
194 private:
195  ArithmeticOperation _op{};
196 };
197 } // namespace kernels
198 } // namespace opencl
199 } // namespace arm_compute
200 #endif /* ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H */
ArithmeticOperation
Available element-wise operations.
Definition: Types.h:489
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
Status class.
Definition: Error.h:52
Common interface for all the OpenCL kernels.
Definition: ICLKernel.h:81
Activation Layer Information class.
Definition: Types.h:1625
Copyright (c) 2017-2022 Arm Limited.
void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override
Interface for an element-wise operation kernel.
const char * name
CLCompileContext class.
Tensor packing service.
Definition: ITensorPack.h:39
std::string kernel_name
LogicalOperation
List of supported logical operations.
Definition: KernelTypes.h:30
Describe a multidimensional execution window.
Definition: Window.h:39
ConvertPolicy
Policy to handle integer overflow.
Definition: Types.h:404
Status validate(const ITensorInfo *scores_in, const ITensorInfo *boxes_in, const ITensorInfo *batch_splits_in, const ITensorInfo *scores_out, const ITensorInfo *boxes_out, const ITensorInfo *classes, const ITensorInfo *batch_splits_out, const ITensorInfo *keeps, const ITensorInfo *keeps_size, const BoxNMSLimitInfo info)