xref: /aosp_15_r20/external/ComputeLibrary/src/cpu/kernels/CpuGemmLowpMatrixMultiplyKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2022 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_CPU_GEMMLOWP_MATRIXMULTIPLY_KERNEL_H
25 #define ARM_COMPUTE_CPU_GEMMLOWP_MATRIXMULTIPLY_KERNEL_H
26 
27 #include "src/core/common/Macros.h"
28 #include "src/cpu/ICpuKernel.h"
29 
30 namespace arm_compute
31 {
32 namespace cpu
33 {
34 namespace kernels
35 {
36 /** Kernel to multiply matrices
37  *
38  * @note @ref CpuGemmLowpMatrixMultiplyKernel low precision matrix product kernel
39  *  This kernel performs the following computation:
40  *
41  *  -# Convert a values from int8 to int32
42  *  -# Convert b values from int8 to int32
43  *  -# Compute the int32 matrix product of the resulting a * b and store the result as int32
44  *
45  */
46 class CpuGemmLowpMatrixMultiplyKernel : public ICpuKernel<CpuGemmLowpMatrixMultiplyKernel>
47 {
48 public:
49     /** Default constructor */
50     CpuGemmLowpMatrixMultiplyKernel() = default;
51     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(CpuGemmLowpMatrixMultiplyKernel);
52     /** Initialise the kernel's input and output.
53      *
54      * The input matrices @p src0 and @p src1 must be the output of the kernels: @ref CpuGemmInterleave4x4Kernel and @ref CpuGemmTranspose1xWKernel. These two
55      * kernels change the layout of the original matrices to be more cache-friendly.
56      *
57      * @param[in]  src0 Input tensor info containing the interleaved Matrix A. Data type supported: U8/QASYMM8/S8/QASYMM8_SIGNED
58      * @param[in]  src1 Input tensor info containing the transposed1xW Matrix B. Data type supported: U8/QASYMM8/S8/QASYMM8_SIGNED/QSYMM8/QSYMM8_PER_CHANNEL
59      * @param[out] dst  Output tensor info to store the result of matrix multiplication. Data type supported: S32
60      */
61     void configure(const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst);
62     /** Static function to check if given info will lead to a valid configuration
63      *
64      * Similar to CpuGemmLowpMatrixMultiplyKernel::configure()
65      *
66      * @return a status
67      */
68     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst);
69 
70     // Inherited methods overridden:
71     void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override;
72     const char *name() const override;
73 
74 private:
75     bool _slide_matrix_b{ true };
76 };
77 } // namespace kernels
78 } // namespace cpu
79 } // namespace arm_compute
80 #endif /*ARM_COMPUTE_CPU_GEMMLOWP_MATRIXMULTIPLY_KERNEL_H*/
81