Compute Library
 21.02
NEGEMMLowpQuantizeDownInt32ScaleKernel.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2020-2021 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_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H
25 #define ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H
26 
28 
29 namespace arm_compute
30 {
31 class ITensor;
32 
33 /** Neon kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED
34  *
35  * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8/QASYMM8_SIGNED value.
36  * The following computations will be performed by the kernel:
37  *
38  * -# Add offset terms to final result
39  * -# Multiply each entry of result by result_mult_int
40  * -# Add bias to final result if bias tensor is not a nullptr
41  * -# Shift the int32 accumulator by result_shift
42  * -# Clamp the value between the specified min and max bounds
43  * -# Clamp the resulting int32 values:
44  * -# -to the [0..255] range and cast to QASYMM8.
45  * -# -to the [-128..127] range and cast to QASYMM8_SIGNED.
46  *
47  */
49 {
50 public:
51  const char *name() const override
52  {
53  return "NEGEMMLowpQuantizeDownInt32ScaleKernel";
54  }
55  /** Constructor */
57  /** Prevent instances of this class from being copied (As this class contains pointers)*/
59  /** Prevent instances of this class from being copied (As this class contains pointers)*/
61  /** Allow instances of this class to be moved */
63  /** Allow instances of this class to be moved */
65  /** Default destructor */
67  /** Initialise the kernel's input and output.
68  *
69  * @param[in] input Input tensor. Data type supported: S32
70  * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
71  * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
72  * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED
73  * @param[out] output_stage GEMMLowp output stage metadata.
74  */
75  void configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo *output_stage);
76  /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ScaleKernel
77  *
78  * @param[in] input Input tensor. Data type supported: S32
79  * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
80  * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
81  * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED
82  * @param[out] output_stage GEMMLowp output stage metadata.
83  *
84  * @return a status
85  */
86  static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage);
87 
88  // Inherited methods overridden:
89  void run(const Window &window, const ThreadInfo &info) override;
90 
91 private:
92  /** Template function to run the NEGEMMLowpQuantizeDownInt32ScaleKernel
93  *
94  * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
95  */
96  template <typename T>
97  void run(const Window &window);
98 
99  /** Common signature for all the specialised NEGEMMLowpQuantizeDownInt32ScaleKernel functions
100  *
101  * @param[in] window Region on which to execute the kernel.
102  */
103  using QuantizeDownFunctionPtr = void (NEGEMMLowpQuantizeDownInt32ScaleKernel::*)(const Window &window);
104 
105  QuantizeDownFunctionPtr _func;
106  const ITensor *_input;
107  const ITensor *_bias;
108  ITensor *_output;
109  const GEMMLowpOutputStageInfo *_output_stage;
110  bool _is_bounded_relu;
111 };
112 } // namespace arm_compute
113 
114 #endif /* ARM_COMPUTE_NEGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H */
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
NEGEMMLowpQuantizeDownInt32ScaleKernel & operator=(const NEGEMMLowpQuantizeDownInt32ScaleKernel &)=delete
Prevent instances of this class from being copied (As this class contains pointers) ...
Neon kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED...
Common interface for all kernels implemented in C++.
Definition: ICPPKernel.h:38
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Static function to check if given info will lead to a valid configuration of NEGEMMLowpQuantizeDownIn...
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
Status class.
Definition: Error.h:52
Interface for Neon tensor.
Definition: ITensor.h:36
Copyright (c) 2017-2021 Arm Limited.
GEMMLowp output stage info.
Definition: Types.h:1952
void configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo *output_stage)
Initialise the kernel&#39;s input and output.
~NEGEMMLowpQuantizeDownInt32ScaleKernel()=default
Default destructor.
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Information about executing thread and CPU.
Definition: CPPTypes.h:235
void run(const Window &window, const ThreadInfo &info) override
Execute the kernel on the passed window.
Describe a multidimensional execution window.
Definition: Window.h:39