1 /*
2  * Copyright (c) 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_GEMMLOWP_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_MMUL_KERNEL_H
25 #define ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_MMUL_KERNEL_H
26 
27 #include "arm_compute/core/KernelDescriptors.h"
28 #include "src/core/common/Macros.h"
29 #include "src/gpu/cl/IClKernel.h"
30 
31 namespace arm_compute
32 {
33 namespace opencl
34 {
35 namespace kernels
36 {
37 /** OpenCL kernel to multiply matrices with QASYMM8/QASYMM8_SIGNED data types when only the input matrix RHS (src1) has been reshaped using the MMUL instruction
38  *
39  * @note The input matrix src1 must be reshaped through @ref opencl::kernels::ClGemmReshapeRhsMatrixKernel
40  * @note For fused output stage, only GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT type is supported
41  */
42 class ClGemmLowpMatrixMultiplyReshapedOnlyRhsMMULKernel : public IClKernel
43 {
44 public:
45     ClGemmLowpMatrixMultiplyReshapedOnlyRhsMMULKernel();
46     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmLowpMatrixMultiplyReshapedOnlyRhsMMULKernel);
47     /** Initialise the kernel's source and destination.
48      *
49      * @param[in]  compile_context    The compile context to be used.
50      * @param[in]  src0               Input tensor containing the LHS matrix. Data type supported: QASYMM8/QASYMM8_SIGNED
51      * @param[in]  src1               Input tensor containing the RHS reshaped matrix. Data type supported: same as @p src0
52      * @param[out] dst                Destination tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/S32.
53      * @param[in]  gemm_info          GEMM information used to retrieve the original dimensions of the input matrices, output stage information and RHS/LHS info.
54      *                                lhs_info.m0: 1,2,4
55      *                                Only the following values are supported for RHS info:
56      *                                rhs_info.n0: 1,4,8
57      *                                rhs_info.k0: same as lhs_info.k0: 4
58      *                                rhs_info.transpose: true
59      * @param[in]  vector_sum_col     (Optional) Input row-vector of sums of all the entries in each column of matrix B.
60      *                                Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: S32
61      * @param[in]  vector_sum_row     (Optional) Input row-vector of sums of all the entries in each row of matrix A.
62      *                                Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: S32
63      * @param[in]  bias               (Optional) Biases tensor. Can be a nullptr if the addition of biases is not required.
64      *                                Biases are 1D tensor with dimensions [OFM] or same dimensionality as dst if gemm_info.broadcast_bias is false. Data type supported: S32.
65      * @param[in]  output_multipliers (Optional) Output multipliers tensor. Supported data types: S32.
66      * @param[in]  output_shifts      (Optional) Output shifts tensor. Supported data types: S32.
67      */
68     void configure(const CLCompileContext &compile_context, const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst, const GEMMKernelInfo &gemm_info,
69                    ITensorInfo *vector_sum_col = nullptr, const ITensorInfo *vector_sum_row = nullptr, ITensorInfo *bias = nullptr,
70                    ITensorInfo *output_multipliers = nullptr, ITensorInfo *output_shifts = nullptr);
71     /** Static function to check if given info will lead to a valid configuration
72      *
73      * Similar to @ref ClGemmLowpMatrixMultiplyReshapedOnlyRhsMMULKernel::configure()
74      *
75      * @return a status
76      */
77     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst, const GEMMKernelInfo &gemm_info,
78                            const ITensorInfo *vector_sum_col = nullptr, const ITensorInfo *vector_sum_row = nullptr, const ITensorInfo *bias = nullptr,
79                            const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr);
80 
81     // Inherited methods overridden:
82     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
83 
84 private:
85     bool       _fuse_output_stage{ false };
86     signed int _m{ 1 };
87     signed int _n{ 1 };
88     signed int _k{ 1 };
89 };
90 } // namespace kernels
91 } // namespace opencl
92 } // namespace arm_compute
93 #endif /* ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_MMULKERNEL_H */