1 /*
2 * Copyright (c) 2019-2020 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/CLFFTScaleKernel.h"
25
26 #include "arm_compute/core/CL/CLKernelLibrary.h"
27 #include "arm_compute/core/CL/ICLTensor.h"
28 #include "arm_compute/core/TensorInfo.h"
29 #include "src/core/CL/CLValidate.h"
30 #include "src/core/helpers/AutoConfiguration.h"
31 #include "src/core/helpers/WindowHelpers.h"
32 #include "support/StringSupport.h"
33
34 namespace arm_compute
35 {
36 namespace
37 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output)38 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
39 {
40 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
41 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32);
42
43 // Checks performed when output is configured
44 if((output != nullptr) && (output->total_size() != 0))
45 {
46 ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 1 && output->num_channels() != 2);
47 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
48 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
49 }
50
51 return Status{};
52 }
53
validate_and_configure_window(ITensorInfo * input,ITensorInfo * output)54 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
55 {
56 // Configure kernel window
57 Window win = calculate_max_window(*input, Steps());
58
59 if(output != nullptr)
60 {
61 // Output auto inizialitation if not yet initialized
62 auto_init_if_empty(*output, *input->clone());
63
64 // CLFFTScaleKernel doesn't need padding so update_window_and_padding() can be skipped
65 Coordinates coord;
66 coord.set_num_dimensions(output->num_dimensions());
67 output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
68 }
69
70 return std::make_pair(Status{}, win);
71 }
72 } // namespace
73
CLFFTScaleKernel()74 CLFFTScaleKernel::CLFFTScaleKernel()
75 : _input(nullptr), _output(nullptr), _run_in_place(false)
76 {
77 }
78
configure(ICLTensor * input,ICLTensor * output,const FFTScaleKernelInfo & config)79 void CLFFTScaleKernel::configure(ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
80 {
81 configure(CLKernelLibrary::get().get_compile_context(), input, output, config);
82 }
83
configure(const CLCompileContext & compile_context,ICLTensor * input,ICLTensor * output,const FFTScaleKernelInfo & config)84 void CLFFTScaleKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
85 {
86 ARM_COMPUTE_ERROR_ON_NULLPTR(input);
87 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr));
88
89 _input = input;
90 _output = output;
91 _run_in_place = (output == nullptr) || (output == input);
92
93 // Create kernel
94 CLBuildOptions build_opts;
95 build_opts.add_option_if(_run_in_place, "-DIN_PLACE");
96 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(output != nullptr ? output->info()->num_channels() : input->info()->num_channels()));
97 build_opts.add_option_if(config.conjugate, "-DCONJ");
98 std::string kernel_name = "fft_scale_conj";
99 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
100
101 // Set static arguments
102 unsigned int idx = (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
103 _kernel.setArg<cl_float>(idx, config.scale);
104
105 // Configure kernel window
106 auto win_config = validate_and_configure_window(input->info(), _run_in_place ? nullptr : output->info());
107 ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
108 ICLKernel::configure_internal(win_config.second);
109
110 // Set config_id for enabling LWS tuning
111 _config_id = kernel_name;
112 _config_id += "_";
113 _config_id += lower_string(string_from_data_type(input->info()->data_type()));
114 _config_id += "_";
115 _config_id += support::cpp11::to_string(input->info()->dimension(0));
116 _config_id += "_";
117 _config_id += support::cpp11::to_string(input->info()->dimension(1));
118 }
119
validate(const ITensorInfo * input,const ITensorInfo * output,const FFTScaleKernelInfo & config)120 Status CLFFTScaleKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const FFTScaleKernelInfo &config)
121 {
122 ARM_COMPUTE_UNUSED(config);
123 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
124 ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
125
126 return Status{};
127 }
128
run(const Window & window,cl::CommandQueue & queue)129 void CLFFTScaleKernel::run(const Window &window, cl::CommandQueue &queue)
130 {
131 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
132 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
133
134 Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
135 Window slice = collapsed.first_slice_window_3D();
136
137 do
138 {
139 unsigned int idx = 0;
140 add_3D_tensor_argument(idx, _input, slice);
141 if(!_run_in_place)
142 {
143 add_3D_tensor_argument(idx, _output, slice);
144 }
145 enqueue(queue, *this, slice, lws_hint());
146 }
147 while(collapsed.slide_window_slice_3D(slice));
148 }
149 } // namespace arm_compute
150