• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016-2020 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_NEPIXELWISEMULTIPLICATIONKERNEL_H
25 #define ARM_COMPUTE_NEPIXELWISEMULTIPLICATIONKERNEL_H
26 
27 #include "arm_compute/core/Types.h"
28 #include "src/core/NEON/INEKernel.h"
29 
30 namespace arm_compute
31 {
32 class ITensor;
33 
34 /** Interface for the kernel to perform addition between two tensors */
35 class NEPixelWiseMultiplicationKernel : public INEKernel
36 {
37 public:
name()38     const char *name() const override
39     {
40         return "NEPixelWiseMultiplicationKernel";
41     }
42     /** Default constructor */
43     NEPixelWiseMultiplicationKernel();
44     /** Prevent instances of this class from being copied (As this class contains pointers) */
45     NEPixelWiseMultiplicationKernel(const NEPixelWiseMultiplicationKernel &) = delete;
46     /** Prevent instances of this class from being copied (As this class contains pointers) */
47     NEPixelWiseMultiplicationKernel &operator=(const NEPixelWiseMultiplicationKernel &) = delete;
48     /** Allow instances of this class to be moved */
49     NEPixelWiseMultiplicationKernel(NEPixelWiseMultiplicationKernel &&) = default;
50     /** Allow instances of this class to be moved */
51     NEPixelWiseMultiplicationKernel &operator=(NEPixelWiseMultiplicationKernel &&) = default;
52     /** Default destructor */
53     ~NEPixelWiseMultiplicationKernel() = default;
54     /** Initialise the kernel's input, output and border mode.
55      *
56      * Valid configurations (Input1,Input2) -> Output :
57      *
58      *                                                       Support: Broadcast? Scale=1/255?
59      *   - (U8,U8)                         -> U8, S16                 N          Y
60      *   - (U8,S16)                        -> S16                     N          Y
61      *   - (S16,U8)                        -> S16                     N          Y
62      *   - (S16,S16)                       -> S16                     N          Y
63      *   - (S32,S32)                       -> S32                     Y          N
64      *   - (F16,F16)                       -> F16                     N          Y
65      *   - (F32,F32)                       -> F32                     Y          Y
66      *   - (QASYMM8,QASYMM8)               -> QASYMM8                 Y          Y
67      *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED          Y          Y
68      *   - (QSYMM16,QSYMM16)               -> QSYMM16, S32            N          Y
69      *
70      * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported.
71      *       For all other scale values only round to zero (implemented as round towards minus infinity) is supported.
72      *
73      * @param[in]  input1          First input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
74      * @param[in]  input2          Second input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
75      * @param[out] output          Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
76      * @param[in]  scale           Scale to apply after multiplication.
77      *                             Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15.
78      *                             If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255
79      * @param[in]  overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype
80      * @param[in]  rounding_policy Rounding policy.
81      */
82     void configure(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy);
83     /** Static function to check if given info will lead to a valid configuration of @ref NEPixelWiseMultiplicationKernel
84      *
85      * Valid configurations (Input1,Input2) -> Output :
86      *                                                       Support: Broadcast? Scale=1/255?
87      *   - (U8,U8)                         -> U8, S16                 N          Y
88      *   - (U8,S16)                        -> S16                     N          Y
89      *   - (S16,U8)                        -> S16                     N          Y
90      *   - (S16,S16)                       -> S16                     N          Y
91      *   - (S32,S32)                       -> S32                     Y          N
92      *   - (F16,F16)                       -> F16                     N          Y
93      *   - (F32,F32)                       -> F32                     Y          Y
94      *   - (QASYMM8,QASYMM8)               -> QASYMM8                 Y          Y
95      *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED          Y          Y
96      *   - (QSYMM16,QSYMM16)               -> QSYMM16, S32            N          Y
97      *
98      * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported.
99      *       For all other scale values only round to zero (implemented as round towards minus infinity) is supported.
100      *
101      * @param[in] input1          First input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
102      * @param[in] input2          Second input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
103      * @param[in] output          Output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32
104      * @param[in] scale           Scale to apply after multiplication.
105      *                            Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15.
106      *                            If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255
107      * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype
108      * @param[in] rounding_policy Rounding policy.
109      *
110      * @return a status
111      */
112     static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy);
113 
114     // Inherited methods overridden
115     void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override;
116 
117 private:
118     /** Common signature for all the specialised multiplication functions with integer scaling factor
119      *
120      * @param[in]  in1    Input1 tensor object.
121      * @param[in]  in2    Input2 tensor object.
122      * @param[out] out    Output tensor object.
123      * @param[in]  window Region on which to execute the kernel
124      * @param[in]  scale  Integer scale factor.
125      */
126     using MulFunctionInt = void(const ITensor *in1, const ITensor *in2, ITensor *out, const Window &window, int scale);
127     /** Common signature for all the specialised multiplication functions with float scaling factor
128      *
129      * @param[in]  in1    Input1 tensor object.
130      * @param[in]  in2    Input2 tensor object.
131      * @param[out] out    Output tensor object.
132      * @param[in]  window Region on which to execute the kernel
133      * @param[in]  scale  Float scale factor.
134      */
135     using MulFunctionFloat = void(const ITensor *in1, const ITensor *in2, ITensor *out, const Window &window, float scale);
136     /** Common signature for all the specialised QASYMM8 multiplication functions with float scaling factor
137      *
138      * @param[in]  in1    Input1 tensor object.
139      * @param[in]  in2    Input2 tensor object.
140      * @param[out] out    Output tensor object.
141      * @param[in]  window Region on which to execute the kernel
142      * @param[in]  scale  Float scale factor.
143      *
144      */
145     using MulFunctionQuantized = void(const ITensor *in1, const ITensor *in2, ITensor *out, const Window &window, float scale);
146 
147     MulFunctionFloat     *_func_float;
148     MulFunctionInt       *_func_int;
149     MulFunctionQuantized *_func_quantized;
150 
151 private:
152     float _scale;
153     int   _scale_exponent;
154 };
155 
156 /** Interface for the complex pixelwise multiplication kernel. */
157 class NEComplexPixelWiseMultiplicationKernel : public INEKernel
158 {
159 public:
name()160     const char *name() const override
161     {
162         return "NEComplexPixelWiseMultiplicationKernel";
163     }
164     /** Initialise the kernel's input, output and border mode.
165      *
166      * @param[in]  input1 An input tensor. Data types supported: F32. Number of channels supported: 2 (complex tensor).
167      * @param[in]  input2 An input tensor. Data types supported: same as @p input1. Number of channels supported: same as @p input1.
168      * @param[out] output The output tensor, Data types supported: same as @p input1.  Number of channels supported: same as @p input1.
169      */
170     void configure(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
171     /** Static function to check if given info will lead to a valid configuration of @ref NEComplexPixelWiseMultiplicationKernel
172      *
173      * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2 (complex tensor).
174      * @param[in] input2 An input tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1.
175      * @param[in] output The output tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1.
176      *
177      * @return a status
178      */
179     static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
180 
181     // Inherited methods overridden:
182     void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override;
183 };
184 
185 } // namespace arm_compute
186 #endif /*ARM_COMPUTE_NEPIXELWISEMULTIPLICATIONKERNEL_H */
187