xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2022 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_CL_GEMM_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_KERNEL_H
25 #define ARM_COMPUTE_CL_GEMM_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_KERNEL_H
26 
27 #include "src/core/common/Macros.h"
28 #include "src/gpu/cl/ClCompileContext.h"
29 #include "src/gpu/cl/IClKernel.h"
30 
31 #include "arm_compute/core/KernelDescriptors.h"
32 
33 namespace arm_compute
34 {
35 namespace opencl
36 {
37 namespace kernels
38 {
39 /** OpenCL kernel to multiply matrices when only the input matrix RHS (src1) has been reshaped
40  *
41  * @note The input matrix src1 must be reshaped through @ref ClGemmReshapeRhsMatrixKernel
42  */
43 class ClGemmMatrixMultiplyReshapedOnlyRhsKernel : public ICLKernel
44 {
45 public:
46     ClGemmMatrixMultiplyReshapedOnlyRhsKernel();
47     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmMatrixMultiplyReshapedOnlyRhsKernel);
48     /** Initialise the kernel's input and output.
49      *
50      * @note If rhs_info.export_to_cl_image = true, this OpenCL kernel will fetch the RHS data using the OpenCL read_image built-in function.
51      *       Reading from the OpenCL image object can increase the performance. However, since the OpenCL image object is created importing the OpenCL buffer,
52      *       the following conditions are required:
53      *       -# rhs_info.n0 can only be 4, 8 and 16
54      *       -# rhs_info.k0 can only be 4, 8 and 16
55      *       -# Data type can only be F32
56      *       -# The platform should support the OpenCL cl_khr_image2d_from_buffer extension
57      *       -# The stride Y for the src1 should satisfy the OpenCL pitch alignment requirement
58      *       -# src1 width should be less or equal to (CL_DEVICE_IMAGE2D_MAX_WIDTH * 4)
59      *       -# src1 (height * depth) should be less or equal to CL_DEVICE_IMAGE2D_MAX_HEIGHT
60      *
61      * @param[in]  compile_context The compile context to be used.
62      * @param[in]  src0            Input tensor containing the LHS matrix. Data type supported: F16/F32 (only F32 if rhs_info.export_to_cl_image = true).
63      *                             The number of dimensions for the LHS matrix must be less or equal than 4.
64      * @param[in]  src1            Input tensor containing the RHS reshaped matrix. Data type supported: same as @p src0. The number of dimensions for the RHS matrix must be less or equal than 3.
65      * @param[in]  src2            Input tensor containing the bias matrix. Data type supported: same as @p src0.
66      * @param[out] dst             Output tensor to store the result of matrix multiplication. Data type supported: same as @p src0
67      * @param[in]  alpha           Weight of the matrix product
68      * @param[in]  beta            Weight of the matrix bias
69      * @param[in]  lhs_info        LHS matrix information used to retrieve the number of rows to be processed by each thread. Only the following values are supported:
70      *                             lhs_info.m0: 1,2,3,4,5,6,7,8
71      * @param[in]  rhs_info        RHS matrix information used for reshaping the src1 tensor.  Only the following values are supported:
72      *                             rhs_info.k0: 2,3,4,8,16
73      *                             rhs_info.n0: 2,3,4,8,16
74      *                             rhs_info.transpose: true,false
75      * @param[in]  gemm_info       GEMM information used to retrieve the original dimensions of the input matrices
76      */
77     void configure(const ClCompileContext &compile_context,
78                    const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *src2, ITensorInfo *dst, float alpha, float beta,
79                    const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info);
80     /** Static function to check if given info will lead to a valid configuration
81      *
82      * Similar to @ref ClGemmMatrixMultiplyReshapedOnlyRhsKernel::configure()
83      *
84      * @return a status
85      */
86     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, float alpha, float beta,
87                            const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info);
88 
89     // Inherited methods overridden:
90     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
91 
92 private:
93     bool         _slide_matrix_b{ true };
94     bool         _reinterpret_input_as_3d{ false };
95     bool         _reinterpret_output_as_3d{ false };
96     bool         _use_dummy_work_items{ false };
97     bool         _add_bias{ false };
98     bool         _export_to_cl_image{ false };
99     bool         _has_pad_y{ false };
100     signed int   _m{ 1 };
101     signed int   _n{ 1 };
102     signed int   _k{ 1 };
103     unsigned int _num_post_op_args{ 0 }; // (EXPERIMENTAL_POST_OPS) total number of post op arguments
104 };
105 } // namespace kernels
106 } // namespace opencl
107 } // namespace arm_compute
108 #endif /* ARM_COMPUTE_CL_GEMM_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_KERNEL_H */
109