xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuSub.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2023 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_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUSUB
25 #define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUSUB
26 
27 #include "arm_compute/core/Error.h"
28 
29 namespace arm_compute
30 {
31 /** Forward declaration */
32 class ITensorInfo;
33 
34 namespace experimental
35 {
36 namespace dynamic_fusion
37 {
38 /** Forward declaration */
39 class GpuWorkloadContext;
40 class GpuWorkloadSketch;
41 
42 /** Operator interface. */
43 class GpuSub final
44 {
45 public:
46     /** Create an operator and fuse it into the workload sketch.
47      *    @note If @ref validate_op() fails, the creation also fails and may throw an error.
48      *    @note If @ref validate_op() fails, @p sketch remains unchanged and valid.
49      *
50      * Valid data type configurations:
51      * |lhs            |rhs            |dst           |
52      * |:--------------|:--------------|:-------------|
53      * |F16            |F16            |F16           |
54      * |F32            |F32            |F32           |
55      * |S32            |S32            |S32           |
56      * |S16            |S16            |S16           |
57      * |U8             |U8             |U8            |
58      *
59      * Valid data layouts:
60      * - Any
61      *
62      * @param[in,out] sketch Workload sketch into which the operator will be fused
63      * @param[in]     lhs    Left hand side tensor info. Data types supported: U8/S16/S32/F16/F32.
64      * @param[in]     rhs    Right hand side tensor info. Same as @p lhs.
65      *
66      * @return Pointer for the destination tensor info
67      */
68     static ITensorInfo *create_op(GpuWorkloadSketch &sketch,
69                                   ITensorInfo       *lhs,
70                                   ITensorInfo       *rhs);
71 
72     /** Check if the operator configuration is supported, irrespective of fusion
73      *
74      * @param[in] context Workload context within which the operator is running
75      * @param[in] lhs     Left hand side tensor info. Data types supported: U8/S16/S32/F16/F32.
76      * @param[in] rhs     Right hand side tensor info. Same as @p lhs.
77      *
78      * @return Status
79      */
80     static Status is_supported_op(const GpuWorkloadContext &context,
81                                   const ITensorInfo        *lhs,
82                                   const ITensorInfo        *rhs);
83 
84     /** Validate the operator and check if its configuration is supported and if it can be fused into the workload sketch.
85      *
86      * Parameters are similar to @ref GpuSub::create_op()
87      *
88      * @return Status
89      */
90     static Status validate_op(const GpuWorkloadSketch &sketch,
91                               const ITensorInfo       *rhs,
92                               const ITensorInfo       *lhs);
93 };
94 } // namespace dynamic_fusion
95 } // namespace experimental
96 } // namespace arm_compute
97 #endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUSUB */
98