• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2017-2020 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 
25 #ifndef ARM_COMPUTE_GCGEMM_H
26 #define ARM_COMPUTE_GCGEMM_H
27 
28 #include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.h"
29 #include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.h"
30 #include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.h"
31 #include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMTranspose1xWKernel.h"
32 #include "arm_compute/runtime/GLES_COMPUTE/GCTensor.h"
33 #include "arm_compute/runtime/IFunction.h"
34 #include "arm_compute/runtime/MemoryGroup.h"
35 
36 namespace arm_compute
37 {
38 class IGCTensor;
39 
40 /** Basic function to execute GEMM on OpenGLES Compute. This function calls the following kernels:
41  *
42  *  -# @ref GCGEMMInterleave4x4Kernel (if the output tensor is a matrix)
43  *  -# @ref GCGEMMTranspose1xWKernel (if the output tensor is a matrix)
44  *  -# @ref GCGEMMMatrixMultiplyKernel
45  *  -# @ref GCGEMMMatrixAdditionKernel (if c != nullptr and beta != 0.0)
46  *
47  * @deprecated This function is deprecated and is intended to be removed in 21.05 release
48  *
49  */
50 class GCGEMM : public IFunction
51 {
52 public:
53     /** Default constructor. */
54     GCGEMM(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
55     /** Prevent instances of this class from being copied (As this class contains pointers) */
56     GCGEMM(const GCGEMM &) = delete;
57     /** Default move constructor */
58     GCGEMM(GCGEMM &&) = default;
59     /** Prevent instances of this class from being copied (As this class contains pointers) */
60     GCGEMM &operator=(const GCGEMM &) = delete;
61     /** Default move assignment operator */
62     GCGEMM &operator=(GCGEMM &&) = default;
63     /** Initialise the kernel's inputs and output
64      *
65      * @note GEMM: General Matrix Multiply - [alpha * A * B + beta * C].
66      *
67      * @note All tensors must have the same data type.
68      *
69      * @note Whilst the first input tensor can be a vector, the second input tensor must be at least a matrix
70      *
71      * @param[in]  a         First input tensor  (Matrix or Vector A). Data types supported: F32
72      * @param[in]  b         Second input tensor (Matrix B). Data type supported: same as @p a.
73      * @param[in]  c         Third input tensor  (Matrix C). It can be a nullptr if just the multiplication between @p a and @p b is needed. Data type supported: same as @p a.
74      * @param[out] output    Output tensor. Data type supported: same as @p a
75      * @param[in]  alpha     Weight of the matrix product
76      * @param[in]  beta      Weight of matrix C
77      * @param[in]  gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
78      *                       if the reshape of matrix B should happen only for the first run
79      */
80     void configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *c, IGCTensor *output, float alpha, float beta, const GEMMInfo &gemm_info = GEMMInfo());
81     /** Static function to check if given info will lead to a valid configuration of @ref GCGEMM.
82      *
83      * @param[in]  a         First input tensor  (Matrix or Vector A). Data types supported: F16/F32
84      * @param[in]  b         Second input tensor (Matrix B). Data type supported: same as @p a.
85      * @param[in]  c         Third input tensor  (Matrix C). It can be a nullptr if just the multiplication between @p a and @p b is needed. Data type supported: same as @p a.
86      * @param[out] output    Output tensor. Data type supported: same as @p a
87      * @param[in]  alpha     Weight of the matrix product
88      * @param[in]  beta      Weight of matrix C
89      * @param[in]  gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
90      *                       if the reshape of matrix B should happen only for the first run
91      *
92      * @return a status
93      */
94     static Status validate(const ITensorInfo *a, const ITensorInfo *b, const IGCTensor *c, const ITensorInfo *output, const float alpha, const float beta, const GEMMInfo &gemm_info = GEMMInfo());
95 
96     // Inherited methods overridden:
97     void run() override;
98     void prepare() override;
99 
100 private:
101     MemoryGroup                _memory_group;
102     GCGEMMInterleave4x4Kernel  _interleave_kernel;
103     GCGEMMTranspose1xWKernel   _transpose_kernel;
104     GCGEMMMatrixMultiplyKernel _mm_kernel;
105     GCGEMMMatrixAdditionKernel _ma_kernel;
106     GCTensor                   _tmp_a;
107     GCTensor                   _tmp_b;
108     const IGCTensor           *_original_b;
109     bool                       _is_interleaved_transposed;
110     bool                       _run_addition;
111     bool                       _reshape_b_only_on_first_run;
112     bool                       _is_prepared;
113 };
114 }
115 
116 #endif /* ARM_COMPUTE_GCGEMM_H */
117