xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/kernels/CLArgMinMaxLayerKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2020 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_CLARGMINMAXLAYERKERNEL_H
25 #define ARM_COMPUTE_CLARGMINMAXLAYERKERNEL_H
26 
27 #include "arm_compute/core/Types.h"
28 #include "src/core/CL/ICLKernel.h"
29 
30 namespace arm_compute
31 {
32 class ICLTensor;
33 
34 /** Interface for the reduction operation kernel
35  *
36  * @note The default data type for an uninitialized output tensor is
37  *       signed 32-bit integer (S32). It is the user's responsibility to check
38  *       that the results do not overflow because the indices are computed
39  *       in unsigned 32-bit (U32).
40  */
41 class CLArgMinMaxLayerKernel : public ICLKernel
42 {
43 public:
44     /** Default constructor */
45     CLArgMinMaxLayerKernel();
46     /** Prevent instances of this class from being copied (As this class contains pointers) */
47     CLArgMinMaxLayerKernel(const CLArgMinMaxLayerKernel &) = delete;
48     /** Prevent instances of this class from being copied (As this class contains pointers) */
49     CLArgMinMaxLayerKernel &operator=(const CLArgMinMaxLayerKernel &) = delete;
50     /** Allow instances of this class to be moved */
51     CLArgMinMaxLayerKernel(CLArgMinMaxLayerKernel &&) = default;
52     /** Allow instances of this class to be moved */
53     CLArgMinMaxLayerKernel &operator=(CLArgMinMaxLayerKernel &&) = default;
54     /** Default destructor */
55     ~CLArgMinMaxLayerKernel() = default;
56 
57     /** Set the input and output tensors.
58      *
59      * @param[in]  input       Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/S32/F16/F32.
60      * @param[in]  prev_output Destination tensor of the previous iterations of @ref CLArgMinMaxLayerKernel. Data types supported: U32/S32
61      *                         Has to be nullptr for the first iteration
62      * @param[out] output      Destination tensor. Data types supported: U32/S32
63      *                         Output will have the same number of dimensions as input.
64      * @param[in]  axis        Axis along which to reduce. Supported reduction axis : 0,1,2,3
65      * @param[in]  op          Reduction operation to perform. Only ArgMin and ArgMax are supported.
66      */
67     void configure(const ICLTensor *input, const ICLTensor *prev_output, ICLTensor *output, unsigned int axis, ReductionOperation op);
68     /** Set the input and output tensors.
69      *
70      * @param[in]  compile_context The compile context to be used.
71      * @param[in]  input           Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/S32/F16/F32.
72      * @param[in]  prev_output     Destination tensor of the previous iterations of @ref CLArgMinMaxLayerKernel. Data types supported: U32/S32
73      *                             Has to be nullptr for the first iteration
74      * @param[out] output          Destination tensor. Data types supported: U32/S32
75      *                             Output will have the same number of dimensions as input.
76      * @param[in]  axis            Axis along which to reduce. Supported reduction axis : 0,1,2,3
77      * @param[in]  op              Reduction operation to perform. Only ArgMin and ArgMax are supported.
78      */
79     void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *prev_output, ICLTensor *output, unsigned int axis, ReductionOperation op);
80 
81     /** Static function to check if given info will lead to a valid configuration of @ref CLArgMinMaxLayerKernel.
82      *
83      * @param[in] input       Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/S32/F16/F32.
84      * @param[in] prev_output Destination tensor info of the previous iterations. Data types supported: U32/S32
85      *                        Has to be nullptr for the first iteration
86      * @param[in] output      Destination tensor info. Data types supported: U32/S32
87      *                        Output will have the same number of dimensions as input.
88      * @param[in] axis        Axis along which to reduce. Supported reduction axis : 0,1,2,3
89      * @param[in] op          Reduction operation to perform.  Only ArgMin and ArgMax are supported.
90      *
91      * @return a status
92      */
93     static Status validate(const ITensorInfo *input, const ITensorInfo *prev_output, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
94 
95     // Inherited methods overridden:
96     void run(const Window &window, cl::CommandQueue &queue) override;
97 
98 private:
99     const ICLTensor   *_input;
100     const ICLTensor   *_prev_output;
101     ICLTensor         *_output;
102     unsigned int       _reduction_axis;
103     ReductionOperation _op;
104 };
105 } // namespace arm_compute
106 #endif /* ARM_COMPUTE_CLARGMINMAXLAYERKERNEL_H */
107