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_RESHAPED_ONLY_RHS_KERNEL_H
25 #define ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_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 data type when only the input matrix RHS (src1) has been reshaped
39  *
40  * @note The input matrix src1 must be reshaped through @ref opencl::kernels::ClGemmReshapeRhsMatrixKernel
41  * @note For fused output stage, only GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT type is supported
42  */
43 class ClGemmLowpMatrixMultiplyReshapedOnlyRhsKernel : public IClKernel
44 {
45 public:
46     ClGemmLowpMatrixMultiplyReshapedOnlyRhsKernel();
47     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClGemmLowpMatrixMultiplyReshapedOnlyRhsKernel);
48     /** Initialise the kernel's source and destination.
49      *
50      * @param[in]  compile_context    The compile context to be used.
51      * @param[in]  src0               Input tensor containing the LHS matrix. Data type supported: QASYMM8/QASYMM8_SIGNED
52      * @param[in]  src1               Input tensor containing the RHS reshaped matrix. Data type supported: same as @p src0
53      * @param[out] dst                Destination tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/S32.
54      * @param[in]  gemm_info          GEMM information used to retrieve the original dimensions of the input matrices, output stage information and RHS/LHS info.
55      *                                Only the following values are supported for LHS info:
56      *                                lhs_info.m0: 2,3,4,5,6,7,8
57      *                                lhs_info.k0: 2,3,4,8,16
58      *                                Only the following values are supported for RHS info:
59      *                                rhs_info.n0: 2,3,4,8,16
60      *                                rhs_info.k0: same as lhs_info.k0
61      *                                rhs_info.transpose: true
62      * @param[in]  vector_sum_col     (Optional) Input row-vector of sums of all the entries in each column of matrix B.
63      *                                Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: S32
64      * @param[in]  vector_sum_row     (Optional) Input row-vector of sums of all the entries in each row of matrix A.
65      *                                Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: S32
66      * @param[in]  bias               (Optional) Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
67      *                                Biases are 1D tensor with dimensions [OFM]. Data type supported: S32.
68      * @param[in]  output_multipliers (Optional) Output multipliers tensor. In case of per-channel quantization, the number of multipliers must be equal to the number of filters (OFM).
69      *                                Supported data types: S32.
70      * @param[in]  output_shifts      (Optional) Output shifts tensor. In case of per-channel quantization, the number of multipliers must be equal to the number of filters (OFM).
71      *                                Supported data types: S32.
72      */
73     void configure(const CLCompileContext &compile_context, const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst, const GEMMKernelInfo &gemm_info,
74                    ITensorInfo *vector_sum_col = nullptr, const ITensorInfo *vector_sum_row = nullptr, ITensorInfo *bias = nullptr,
75                    ITensorInfo *output_multipliers = nullptr, ITensorInfo *output_shifts = nullptr);
76     /** Static function to check if given info will lead to a valid configuration
77      *
78      * Similar to @ref ClGemmLowpMatrixMultiplyReshapedOnlyRhsKernel::configure()
79      *
80      * @return a status
81      */
82     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst, const GEMMKernelInfo &gemm_info,
83                            const ITensorInfo *vector_sum_col = nullptr, const ITensorInfo *vector_sum_row = nullptr, const ITensorInfo *bias = nullptr,
84                            const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr);
85 
86     // Inherited methods overridden:
87     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
88 
89 private:
90     bool _slide_matrix_b{ true };
91     bool _reinterpret_input_as_3d{ false };
92     bool _reinterpret_output_as_3d{ false };
93     bool _use_dummy_work_items{ false };
94     bool _is_quantized_per_channel{ false };
95     bool _fuse_output_stage{ false };
96 };
97 } // namespace kernels
98 } // namespace opencl
99 } // namespace arm_compute
100 #endif /* ARM_COMPUTE_CL_GEMMLOWP_MATRIXMULTIPLY_RESHAPED_ONLY_RHS_KERNEL_H */