• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2021 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_CONV3D_H
25 #define ARM_COMPUTE_CL_DIRECT_CONV3D_H
26 
27 #include "src/gpu/cl/IClKernel.h"
28 #include "src/gpu/cl/IClOperator.h"
29 
30 #include <memory>
31 
32 namespace arm_compute
33 {
34 class CLCompileContext;
35 struct Conv3dInfo;
36 class IClKernel;
37 
38 namespace opencl
39 {
40 /** Basic function to simulate a directly convolution layer with 3 spatial dimensions. This function calls the following OpenCL kernels:
41  *
42  * -# @ref opencl::ClDirectConv3d
43  */
44 class ClDirectConv3d : public IClOperator
45 {
46 public:
47     ClDirectConv3d() = default;
48     /** Set the src and dst tensors.
49      *
50      * Valid data layouts:
51      * - NDHWC
52      *
53      * Valid data type configurations:
54      * |src0           |src1           |src2   |dst            |
55      * |:--------------|:--------------|:------|:--------------|
56      * |F16            |F16            |F16    |F16            |
57      * |F32            |F32            |F32    |F32            |
58      * |QASYMM8        |QASYMM8        |S32    |QASYMM8        |
59      * |QASYMM8_SIGNED |QASYMM8_SIGNED |S32    |QASYMM8_SIGNED |
60      *
61      * @param[in]  compile_context The compile context to be used.
62      * @param[in]  src0            Source tensor. 4 lower dimensions represent a single src [IFM, width, height, depth],
63      *                             while every optional dimension from 5 and above represent a batch of srcs.
64      * @param[in]  src1            Weights tensor. Weights are 5D tensor with dimensions [OFM, IFM, kernel_w, kernel_h, kernel_d].
65      * @param[in]  src2            Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
66      * @param[out] dst             Destination tensor. 4 lower dimensions represent a single dst [OFM, width, height, depth], while the rest represent batch of dsts.
67      * @param[in]  conv3d_info     Contains strides, padding, rounding, activation, dilation and fast math information. Activation and fast math are currently unused.
68      *
69      */
70     void configure(const CLCompileContext &compile_context, const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *src2, ITensorInfo *dst, const Conv3dInfo &conv3d_info);
71 
72     /** Static function to check if given info will lead to a valid configuration
73      *
74      * Similar to ClDirectConv3d::configure()
75      *
76      * @return a status
77      */
78     static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const Conv3dInfo &conv3d_info);
79 
80     // Inherited method overridden
81     void run(ITensorPack &tensors) override;
82 
83 private:
84     std::unique_ptr<IClKernel> _direct_conv3d_kernel{ nullptr };
85 };
86 } // namespace opencl
87 } // namespace arm_compute
88 #endif /* ARM_COMPUTE_CL_DIRECT_CONV3D_H */