1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
|
/*
* Copyright (c) 2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#ifndef ARM_COMPUTE_CPU_GEMM_H
#define ARM_COMPUTE_CPU_GEMM_H
#include "src/cpu/ICpuOperator.h"
#include "arm_compute/core/ITensorPack.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Types.h"
#include "src/cpu/kernels/CpuGemmInterleave4x4Kernel.h"
#include "src/cpu/kernels/CpuGemmMatrixAdditionKernel.h"
#include "src/cpu/kernels/CpuGemmMatrixMultiplyKernel.h"
#include "src/cpu/kernels/CpuGemmTranspose1xWKernel.h"
#include "src/cpu/operators/CpuActivation.h"
#include "src/cpu/operators/CpuAdd.h"
#include "src/cpu/operators/internal/CpuGemmAssemblyDispatch.h"
#include <memory>
namespace arm_compute
{
namespace cpu
{
/** Basic function to execute GEMM. This function calls the following kernels:
*
* If optimized assembly is available:
* -# @ref cpu::CpuGemmAssemblyDispatch
* -# @ref cpu::CpuActivation (if alpha != 1.0)
* Else:
* -# @ref cpu::kernels::CpuGemmInterleave4x4Kernel (if the output tensor is a matrix)
* -# @ref cpu::kernels::CpuGemmTranspose1xWKernel (if the output tensor is a matrix)
* -# @ref cpu::kernels::CpuGemmMatrixMultiplyKernel
* In both cases:
* -# @ref cpu::kernels::CpuGemmMatrixAdditionKernel (if c != nullptr and beta != 0.0 and is not reshaped once)
* Else:
* -# @ref cpu::CpuAdd (if c != nullptr and is reshaped once and not optimized assembly in place)
*
* -# @ref cpu::CpuActivation (if activation is specified in GEMMInfo)
*/
class CpuGemm : public ICpuOperator
{
public:
/** Default constructor */
CpuGemm() = default;
/** Default destructor */
~CpuGemm() = default;
/** Configure operator for a given list of arguments
*
* Valid data layouts:
* - All
*
* Valid data type configurations:
* |src0 |src1 |src2 |dst |
* |:------------|:-----------|:---------|:--------------|
* |F32 |F32 |F32 |F32 |
* |F16 |F16 |F16 |F16 |
* |BFLOAT16 |BFLOAT16 |BFLOAT16 |BFLOAT16 |
*
* @note GEMM: General Matrix Multiply - [alpha * A * B + beta * C].
* @note GEMM: The tensors a, b, c, d must have the same data type. You should not mix data types when calling this function.
*
* @param[in] a First input tensor info (Matrix A or Vector A). Data type supported: BFLOAT16/F16/F32
* @param[in] b Second input tensor info (Matrix B). Data type supported: same as @p a
* @param[in] c Third input tensor info (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
* @param[out] d Output tensor info. Data type supported: same as @p a
* @param[in] alpha Weight of the matrix product
* @param[in] beta Weight of matrix C
* @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
* if the reshape of matrix B should happen only for the first run
*/
void configure(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, ITensorInfo *d,
float alpha, float beta, const GEMMInfo &gemm_info = GEMMInfo());
/** Static function to check if given info will lead to a valid configuration of @ref CpuGemm.
*
* Similar to @ref CpuGemm::configure()
*
* @return a status
*/
static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *d,
float alpha, float beta, const GEMMInfo &gemm_info = GEMMInfo());
// Inherited methods overridden:
void run(ITensorPack &tensors) override;
void prepare(ITensorPack &constants) override;
experimental::MemoryRequirements workspace() const override;
private:
enum AuxTensorIdx
{
AsmGemmWorkspace = 0,
Pretraspose,
InterleavedLHS,
TransposedRHS,
TempResult,
Count
};
std::unique_ptr<kernels::CpuGemmInterleave4x4Kernel> _interleave_kernel{ nullptr };
std::unique_ptr<kernels::CpuGemmTranspose1xWKernel> _transpose_kernel{ nullptr };
std::unique_ptr<kernels::CpuGemmMatrixMultiplyKernel> _mm_kernel{ nullptr };
std::unique_ptr<CpuGemmAssemblyDispatch> _asm_glue{ nullptr };
std::unique_ptr<kernels::CpuGemmMatrixAdditionKernel> _ma_kernel{ nullptr };
std::unique_ptr<CpuActivation> _alpha_scale_func{ nullptr };
std::unique_ptr<CpuAdd> _add_bias{ nullptr };
std::unique_ptr<CpuActivation> _activation_func{ nullptr };
TensorInfo _tmp_a{};
TensorInfo _tmp_b{};
TensorInfo _tmp_d{};
bool _run_vector_matrix_multiplication{ false };
bool _run_alpha_scale{ false };
bool _run_addition{ false };
bool _run_bias_addition{ false };
bool _run_activation{ false };
bool _reshape_b_only_on_first_run{ false };
bool _is_prepared{ false };
experimental::MemoryRequirements _aux_mem{ Count };
};
} // namespace cpu
} // namespace arm_compute
#endif /*ARM_COMPUTE_CPU_GEMM_H */
|
