xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/core/CPP/ICPPKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2016-2022 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_ICPPKERNEL_H
25 #define ARM_COMPUTE_ICPPKERNEL_H
26 
27 #include "arm_compute/core/CPP/CPPTypes.h"
28 #include "arm_compute/core/IKernel.h"
29 #include "arm_compute/core/Types.h"
30 #include "arm_compute/core/experimental/Types.h"
31 
32 namespace arm_compute
33 {
34 class Window;
35 class ITensor;
36 
37 /** Common interface for all kernels implemented in C++ */
38 class ICPPKernel : public IKernel
39 {
40 public:
41     static constexpr size_t default_mws       = 1; /* Default minimum workload size value  - no impact */
42 
43     /** Default destructor */
44     virtual ~ICPPKernel() = default;
45 
46     /** Execute the kernel on the passed window
47      *
48      * @warning If is_parallelisable() returns false then the passed window must be equal to window()
49      *
50      * @note The window has to be a region within the window returned by the window() method
51      *
52      * @note The width of the window has to be a multiple of num_elems_processed_per_iteration().
53      *
54      * @param[in] window Region on which to execute the kernel. (Must be a region of the window returned by window())
55      * @param[in] info   Info about executing thread and CPU.
56      */
run(const Window & window,const ThreadInfo & info)57     virtual void run(const Window &window, const ThreadInfo &info)
58     {
59         ARM_COMPUTE_UNUSED(window, info);
60         ARM_COMPUTE_ERROR("default implementation of legacy run() virtual member function invoked");
61     }
62 
63     /** legacy compatibility layer for implemantions which do not support thread_locator
64      * In these cases we simply narrow the interface down the legacy version
65      *
66      * @param[in] window         Region on which to execute the kernel. (Must be a region of the window returned by window())
67      * @param[in] info           Info about executing thread and CPU.
68      * @param[in] thread_locator Specifies "where" the current thread is in the multi-dimensional space
69      */
run_nd(const Window & window,const ThreadInfo & info,const Window & thread_locator)70     virtual void run_nd(const Window &window, const ThreadInfo &info, const Window &thread_locator)
71     {
72         ARM_COMPUTE_UNUSED(thread_locator);
73         run(window, info);
74     }
75 
76     /** Execute the kernel on the passed window
77      *
78      * @warning If is_parallelisable() returns false then the passed window must be equal to window()
79      *
80      * @note The window has to be a region within the window returned by the window() method
81      *
82      * @note The width of the window has to be a multiple of num_elems_processed_per_iteration().
83      *
84      * @param[in] tensors A vector containing the tensors to operate on.
85      * @param[in] window  Region on which to execute the kernel. (Must be a region of the window returned by window())
86      * @param[in] info    Info about executing thread and CPU.
87      */
run_op(ITensorPack & tensors,const Window & window,const ThreadInfo & info)88     virtual void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
89     {
90         ARM_COMPUTE_UNUSED(tensors, window, info);
91     }
92 
93     /** Return minimum workload size of the relevant kernel
94      *
95      * @param[in] platform     The CPU platform used to create the context.
96      * @param[in] thread_count Number of threads in the execution.
97      *
98      * @return Minimum workload size for requested configuration.
99      */
get_mws(const CPUInfo & platform,size_t thread_count)100     virtual size_t get_mws(const CPUInfo &platform, size_t thread_count) const
101     {
102         ARM_COMPUTE_UNUSED(platform, thread_count);
103 
104         return default_mws;
105     }
106 
107     /** Name of the kernel
108      *
109      * @return Kernel name
110      */
111     virtual const char *name() const = 0;
112 };
113 } // namespace arm_compute
114 #endif /*ARM_COMPUTE_ICPPKERNEL_H */
115