xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/kernels/ClIndirectConv2dKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 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_CL_DIRECT_CONV2D_KERNEL_H
25 #define ARM_COMPUTE_CL_DIRECT_CONV2D_KERNEL_H
26 
27 #include "src/core/common/Macros.h"
28 #include "src/gpu/cl/ClCompileContext.h"
29 #include "src/gpu/cl/IClKernel.h"
30 
31 namespace arm_compute
32 {
33 // Forward declaration
34 struct DirectConvComputeKernelInfo;
35 
36 namespace opencl
37 {
38 namespace kernels
39 {
40 /** Interface for the  indirect convolution kernel. */
41 class ClIndirectConv2dKernel : public IClKernel
42 {
43 public:
44     ClIndirectConv2dKernel();
45     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClIndirectConv2dKernel);
46     /** Set the src, offset, weights, biases and dst tensors info.
47      *
48      * @param[in]  compile_context The compile context to be used.
49      * @param[in]  src             The src tensor info to convolve. 3 lower dimensions represent a single src [IFM, width, height],
50      *                             while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F16/F32.
51      * @param[in]  off             The indirect buffer tensor info. Data types supported: S32.
52      * @param[in]  weights         Weights tensor info. Weights are 4D tensor with dimensions [IFM, kernel_x, kernel_y, OFM].
53      *                             Data type supported: Same as @p src.
54      * @param[in]  biases          Biases tensor info. Biases are 1D tensor with dimension [OFM].
55      *                             Data type supported: Same as @p src.
56      * @param[out] dst             Output tensor info.
57      *                             The 3rd dimension must be equal to the 4th dimension of the @p weights tensor. Data types supported: Same as @p src.
58      * @param[in]  conv_info       Contains padding and stride information described in @ref PadStrideInfo.
59      * @param[in]  act_info        Contains activaton information described in @ref ActivationLayerInfo.
60      * @param[in]  desc            Direct convolution descriptor used to build the NHWC indirect convolution kernel.
61      */
62     void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *off, ITensorInfo *weights, ITensorInfo *biases, ITensorInfo *dst,
63                    const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info, const DirectConvComputeKernelInfo &desc);
64     /** Static function to check if given info will lead to a valid configuration
65      *
66      * Similar to ClIndirectConv2dKernel::configure()
67      *
68      * @return a status
69      */
70     static Status validate(const ITensorInfo *src, const ITensorInfo *off, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *dst,
71                            const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info, const DirectConvComputeKernelInfo &desc);
72 
73     // Inherited methods overridden:
74     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
75 
76 public:
77     bool _export_to_cl_image{ false };
78 };
79 } // namespace kernels
80 } // namespace opencl
81 } // namespace arm_compute
82 #endif /* ARM_COMPUTE_CL_DIRECT_CONV2D_KERNEL_H */
83