• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016-2020 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 #include "src/core/CL/kernels/CLBitwiseOrKernel.h"
25 
26 #include "arm_compute/core/CL/CLKernelLibrary.h"
27 #include "arm_compute/core/CL/ICLTensor.h"
28 #include "arm_compute/core/CL/OpenCL.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/Validate.h"
31 #include "src/core/helpers/WindowHelpers.h"
32 
33 using namespace arm_compute;
34 
CLBitwiseOrKernel()35 CLBitwiseOrKernel::CLBitwiseOrKernel()
36     : _input1(nullptr), _input2(nullptr), _output(nullptr)
37 {
38 }
39 
configure(const ICLTensor * input1,const ICLTensor * input2,ICLTensor * output)40 void CLBitwiseOrKernel::configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output)
41 {
42     configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output);
43 }
44 
configure(const CLCompileContext & compile_context,const ICLTensor * input1,const ICLTensor * input2,ICLTensor * output)45 void CLBitwiseOrKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output)
46 {
47     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8);
48     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8);
49     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8);
50 
51     _input1 = input1;
52     _input2 = input2;
53     _output = output;
54 
55     // Create kernel
56     _kernel = create_kernel(compile_context, "bitwise_or");
57 
58     // Configure kernel window
59     constexpr unsigned int num_elems_processed_per_iteration = 16;
60 
61     Window win = calculate_max_window(*input1->info(), Steps(num_elems_processed_per_iteration));
62 
63     AccessWindowHorizontal input1_access(input1->info(), 0, num_elems_processed_per_iteration);
64     AccessWindowHorizontal input2_access(input2->info(), 0, num_elems_processed_per_iteration);
65     AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
66 
67     update_window_and_padding(win, input1_access, input2_access, output_access);
68 
69     ValidRegion valid_region = intersect_valid_regions(input1->info()->valid_region(),
70                                                        input2->info()->valid_region());
71 
72     output_access.set_valid_region(win, valid_region);
73 
74     ICLKernel::configure_internal(win);
75 }
76 
run(const Window & window,cl::CommandQueue & queue)77 void CLBitwiseOrKernel::run(const Window &window, cl::CommandQueue &queue)
78 {
79     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
80     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
81 
82     Window slice = window.first_slice_window_2D();
83 
84     do
85     {
86         unsigned int idx = 0;
87         add_2D_tensor_argument(idx, _input1, slice);
88         add_2D_tensor_argument(idx, _input2, slice);
89         add_2D_tensor_argument(idx, _output, slice);
90         enqueue(queue, *this, slice, lws_hint());
91     }
92     while(window.slide_window_slice_2D(slice));
93 }
94