xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/operators/ClGemmLowpOutputStage.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2021 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_GEMMLOWP_OUTPUT_STAGE_H
25 #define ARM_COMPUTE_CL_GEMMLOWP_OUTPUT_STAGE_H
26 
27 #include "src/gpu/cl/ClCompileContext.h"
28 #include "src/gpu/cl/IClOperator.h"
29 
30 /** This file contains all available output stages for GEMMLowp on OpenCL.
31  *
32  *  In gemmlowp, the "output stage" is the process that takes a final int32 accumulator value (the output of @ref CLGEMMLowpMatrixMultiplyCore),
33  *  and processes it to obtain the final QASYMM8/QASYMM8_SIGNED value.
34  *
35  *  More information about the GEMMLowp output stage can be found at https://github.com/google/gemmlowp/blob/master/doc/output.md
36  */
37 
38 namespace arm_compute
39 {
40 namespace opencl
41 {
42 /** Basic function to execute GEMMLowpQuantizeDown kernels on CL.
43  *
44  *  This function calls the following CL kernels:
45  *
46  * -# @ref opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleKernel
47  * -# @ref opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFloatKernel
48  * -# @ref opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFixedPointKernel
49 */
50 class ClGemmLowpOutputStage : public IClOperator
51 {
52 public:
53     /** Constructor */
54     ClGemmLowpOutputStage() = default;
55     /** Initialise the kernel's inputs, output
56      *
57      * Valid data layouts:
58      * - All
59      *
60      * Valid data type configurations:
61      * |src0           |src1          |dst           |
62      * |:--------------|:-------------|:-------------|
63      * |S32            |S32           |QASYMM8       |
64      * |S32            |S32           |QASYMM8_SIGNED|
65      * |S32            |S32           |QSYMM16       |
66      *
67      * @param[in]  compile_context The compile context to be used.
68      * @param[in]  src             Source tensor. Data type supported: S32
69      * @param[in]  bias            Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
70      *                             Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p src.
71      * @param[out] dst             Destination tensor. Data type supported: QASYMM8/QASYMM8_SIGNED
72      * @param[in]  info            GEMMLowp output stage metadata.
73      */
74     void configure(const CLCompileContext &compile_context, const ITensorInfo *src, const ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo &info);
75     /** Static function to check if given info will lead to a valid configuration
76      *
77      * Similar to ClGemmLowpOutputStage::configure()
78      *
79      * @return a status
80      */
81     static Status validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo &info);
82 
83     // Inherited methods overridden:
84     void run(ITensorPack &tensors) override;
85 };
86 } // namespace opencl
87 } // namespace arm_compute
88 #endif /* ARM_COMPUTE_CL_GEMMLOWP_OUTPUT_STAGE_H */
89