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
|
/*
* Copyright (c) 2023 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 ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
#define ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
#include "arm_compute/core/ActivationLayerInfo.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/runtime/IFunction.h"
#include <memory>
namespace arm_compute
{
// Forward declarations for used types instead of including their header, that could minimize compile time
class CLCompileContext;
class ICLTensor;
class ITensorInfo;
class MatMulInfo;
class Status;
/** Settings for MatMul OpenCL implementation */
class GpuMatMulSettings
{
public:
/* Placeholder for operator parity between CPU/GPU */
};
/** Basic function to execute MatMul (Matrix Multiplication) on OpenCL */
class CLMatMul : public IFunction
{
public:
/** Default constructor.*/
CLMatMul();
/** Default destructor */
~CLMatMul();
/** Prevent instances of this class from being copied (As this class contains pointers) */
CLMatMul(const CLMatMul &) = delete;
/** Default move constructor */
CLMatMul(CLMatMul &&);
/** Prevent instances of this class from being copied (As this class contains pointers) */
CLMatMul &operator=(const CLMatMul &) = delete;
/** Default move assignment operator */
CLMatMul &operator=(CLMatMul &&);
/** Initialise the kernel's inputs and output
*
* Valid data layouts:
* - All
*
* Valid data type configurations:
* |lhs |rhs |dst |
* |:--------------|:--------------|:--------------|
* |F32 |F32 |F32 |
* |F16 |F16 |F16 |
* |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
* |QASYMM8 |QASYMM8 |QASYMM8 |
*
* @note BatchMatMul: Batched Matrix Multiply - [A * B], Multiplies all slices (slice is an element of a batch) of Tensors A and B
* and stores the result in the dst tensor of the same batch size.
* Batch here is number of slices from A and B multiplied at a time, do not confuse with the batch dimension 'N' of NHWC/NCHW
* For NHWC for example: the batch is the higher dimensions H * N, and in general it is H * all higher dimensions.
* @note All tensors must have the same data type.
*
* @param[in] compile_context The compile context to be used.
* @param[in] lhs Left-hand side tensor info containing the input activations as Matrix A. Data types supported: F16/F32/QASYMM8_SIGNED/QASYMM8.
* @param[in] rhs Right-hand side tensor info containing the input weights as Matrix B. Data types supported: same as @p lhs.
* @param[out] dst Output tensor to store the result of the batched matrix multiplication. Data types supported: same as @p lhs.
* @param[in] matmul_info Contains MatMul operation information described in @ref MatMulInfo.
* @param[in] settings Contains flags for function level settings
* @param[in] act_info (Optional) Contains activation function and lower and upper bound values for bounded activation functions.
*/
void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *dst, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{}, const
ActivationLayerInfo &act_info = ActivationLayerInfo{});
/** Initialise the kernel's inputs and output
*
* Similar to @ref CLMatMul::configure()
*/
void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *dst, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{}, const ActivationLayerInfo &act_info =
ActivationLayerInfo{});
/** Static function to check if given info will lead to a valid configuration of @ref CLMatMul.
*
*
* @note All tensors must have the same data type.
*
* @param[in] lhs Left-hand side (Matrix A) tensor info. Data types supported: F16/F32/QASYMM8_SIGNED/QASYMM8.
* @param[in] rhs Right-hand side (Matrix B) tensor info. Data types supported: same as @p lhs.
* @param[out] output Output tensor info to store the result of the batched matrix multiplication. Data types supported: same as @p lhs.
* @param[in] matmul_info Contains MatMul operation information described in @ref MatMulInfo.
* @param[in] act_info (Optional) Contains activation function and lower and upper bound values for bounded activation functions.
*/
static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulInfo &matmul_info, const ActivationLayerInfo &act_info = ActivationLayerInfo{});
// Inherited methods overridden:
void run() override;
private:
struct Impl;
std::unique_ptr<Impl> _impl;
};
} // namespace arm_compute
#endif /* ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL */
|
