1 /*
2 * Copyright (c) 2018-2021 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 #include "src/core/CL/kernels/CLReverseKernel.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/TensorInfo.h"
31 #include "src/core/CL/CLValidate.h"
32 #include "src/core/helpers/AutoConfiguration.h"
33 #include "src/core/helpers/WindowHelpers.h"
34 #include "support/StringSupport.h"
35
36 namespace arm_compute
37 {
38 namespace
39 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,const ITensorInfo * axis)40 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *axis)
41 {
42 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, axis);
43 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
44 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
45 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(axis, 1, DataType::U32);
46 ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis->num_dimensions() > 1, "Axis must be a 1D tensor");
47 ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis->dimension(0) > 4, "Only up to 4 dimensions can be reversed");
48
49 // Checks performed when output is configured
50 if(output->total_size() != 0)
51 {
52 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
53 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
54 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
55 }
56
57 return Status{};
58 }
59 } // namespace
60
CLReverseKernel()61 CLReverseKernel::CLReverseKernel()
62 : _input(nullptr), _output(nullptr), _axis(nullptr)
63 {
64 _type = CLKernelType::ELEMENTWISE;
65 }
66
configure(const ICLTensor * input,ICLTensor * output,const ICLTensor * axis)67 void CLReverseKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *axis)
68 {
69 configure(CLKernelLibrary::get().get_compile_context(), input, output, axis);
70 }
71
configure(const CLCompileContext & compile_context,const ICLTensor * input,ICLTensor * output,const ICLTensor * axis)72 void CLReverseKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *axis)
73 {
74 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, axis);
75 auto padding_info = get_padding_info({ input, output, axis });
76
77 _input = input;
78 _output = output;
79 _axis = axis;
80
81 // Output tensor auto initialization if not yet initialized
82 auto_init_if_empty(*output->info(), *input->info()->clone());
83
84 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), axis->info()));
85
86 // Set kernel build options
87 CLBuildOptions build_opts;
88 build_opts.add_option("-DNUM_REVERSE_DIMS=" + support::cpp11::to_string(axis->info()->dimension(0)));
89 build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->info()->element_size()));
90
91 // Create kernel
92 _kernel = create_kernel(compile_context, "reverse", build_opts.options());
93
94 // Set static kernel arguments
95 unsigned int idx = 2 * num_arguments_per_4D_tensor() + num_arguments_per_1D_tensor();
96 add_argument<cl_uint>(idx, input->info()->dimension(0));
97 add_argument<cl_uint>(idx, input->info()->dimension(1));
98 add_argument<cl_uint>(idx, input->info()->dimension(2));
99 add_argument<cl_uint>(idx, input->info()->dimension(3));
100
101 // Configure kernel window
102 Window win = calculate_max_window(*output->info(), Steps());
103 ICLKernel::configure_internal(win);
104
105 // Set config_id for enabling LWS tuning
106 _config_id += "reverse_";
107 _config_id += lower_string(string_from_data_type(input->info()->data_type()));
108 _config_id += "_";
109 _config_id += support::cpp11::to_string(input->info()->dimension(0));
110 _config_id += "_";
111 _config_id += support::cpp11::to_string(input->info()->dimension(1));
112 _config_id += "_";
113 _config_id += support::cpp11::to_string(input->info()->dimension(2));
114 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
115 }
116
validate(const ITensorInfo * input,const ITensorInfo * output,const ITensorInfo * axis)117 Status CLReverseKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *axis)
118 {
119 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, axis));
120 return Status{};
121 }
122
run(const Window & window,cl::CommandQueue & queue)123 void CLReverseKernel::run(const Window &window, cl::CommandQueue &queue)
124 {
125 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
126 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
127
128 Window collapsed = window.collapse(ICLKernel::window(), Window::DimZ);
129 Window slice = collapsed.first_slice_window_4D();
130 Window axis_slice = collapsed.first_slice_window_1D();
131
132 do
133 {
134 unsigned int idx = 0;
135 add_4D_tensor_argument(idx, _input, slice);
136 add_1D_tensor_argument(idx, _axis, axis_slice);
137 add_4D_tensor_argument(idx, _output, slice);
138 enqueue(queue, *this, slice, lws_hint());
139 }
140 while(collapsed.slide_window_slice_4D(slice));
141 }
142 } // namespace arm_compute
143