xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/kernels/ClGemmLowpMatrixMultiplyNativeKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2021 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_MATRIXMULTIPLY_NATIVE_KERNEL_H
25 #define ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_NATIVE_KERNEL_H
26 
27 #include "arm_compute/core/KernelDescriptors.h"
28 #include "src/core/common/Macros.h"
29 #include "src/gpu/cl/ClCompileContext.h"
30 #include "src/gpu/cl/IClKernel.h"
31 
32 namespace arm_compute
33 {
34 namespace opencl
35 {
36 namespace kernels
37 {
38 /** OpenCL kernel to multiply matrices with QASYMM8/QASYMM8_SIGNED data type */
39 class ClGemmLowpMatrixMultiplyNativeKernel : public IClKernel
40 {
41 public:
42     ClGemmLowpMatrixMultiplyNativeKernel();
43     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmLowpMatrixMultiplyNativeKernel);
44     /** Initialise the kernel's input and dst.
45      *
46      * @param[in]  compile_context The compile context to be used.
47      * @param[in]  src0            Source tensor containing the LHS matrix. Data type supported: QASYMM8/QASYMM8_SIGNED
48      * @param[in]  src1            Source tensor containing the RHS matrix. Data type supported: same as @p src0
49      * @param[out] dst             Destination tensor to store the result of matrix multiplication. Data type supported: S32
50      * @param[in]  lhs_info        LHS matrix information used to retrieve the number of rows to be processed by each thread
51      *                             lhs_info.m0: 2,3,4,5,6,7,8
52      *                             lhs_info.k0: 2,3,4,8,16
53      * @param[in]  rhs_info        RHS matrix information used to retrieve the number of columns to be processed by each thread
54      *                             rhs_info.n0: 2,3,4,8,16
55      *                             rhs_info.k0: same as lhs_info.k0
56      * @param[in]  gemm_info       GEMM information used to retrieve the original dimensions of the input matrices
57      */
58     void configure(const CLCompileContext &compile_context, const ITensorInfo *src0, ITensorInfo *src1, ITensorInfo *dst,
59                    const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info);
60     /** Static function to check if given info will lead to a valid configuration
61      *
62      * Similar to @ref ClGemmLowpMatrixMultiplyNativeKernel::configure()
63      *
64      * @return a status
65      */
66     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst,
67                            const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info);
68 
69     // Inherited methods overridden:
70     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
71 
72 private:
73     bool _slide_matrix_b{ true };
74     bool _reinterpret_input_as_3d{ false };
75     bool _reinterpret_output_as_3d{ false };
76     bool _use_dummy_work_items{ false };
77 };
78 } // namespace kernels
79 } // namespace opencl
80 } // namespace arm_compute
81 #endif /* ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_NATIVE_KERNEL_H */
82