xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/kernels/CLFFTScaleKernel.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2021 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/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::F16, 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 } // namespace
54 
CLFFTScaleKernel()55 CLFFTScaleKernel::CLFFTScaleKernel()
56     : _input(nullptr), _output(nullptr), _run_in_place(false)
57 {
58     _type = CLKernelType::ELEMENTWISE;
59 }
60 
configure(ICLTensor * input,ICLTensor * output,const FFTScaleKernelInfo & config)61 void CLFFTScaleKernel::configure(ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
62 {
63     configure(CLKernelLibrary::get().get_compile_context(), input, output, config);
64 }
65 
configure(const CLCompileContext & compile_context,ICLTensor * input,ICLTensor * output,const FFTScaleKernelInfo & config)66 void CLFFTScaleKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
67 {
68     ARM_COMPUTE_ERROR_ON_NULLPTR(input);
69     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr));
70     auto padding_info = get_padding_info({ input, output });
71 
72     _input        = input;
73     _output       = output;
74     _run_in_place = (output == nullptr) || (output == input);
75 
76     // Create kernel
77     CLBuildOptions build_opts;
78     build_opts.add_option_if(_run_in_place, "-DIN_PLACE");
79     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(output != nullptr ? output->info()->num_channels() : input->info()->num_channels()));
80     build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
81     build_opts.add_option_if(config.conjugate, "-DCONJ");
82     std::string kernel_name = "fft_scale_conj";
83     _kernel                 = create_kernel(compile_context, kernel_name, build_opts.options());
84 
85     // Set static arguments
86     unsigned int idx = (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
87     _kernel.setArg<cl_float>(idx, config.scale);
88 
89     // Configure kernel window
90     Window win = calculate_max_window(*input->info(), Steps());
91 
92     if(output != nullptr)
93     {
94         // Output auto inizialitation if not yet initialized
95         auto_init_if_empty(*output->info(), *input->info()->clone());
96     }
97 
98     ICLKernel::configure_internal(win);
99 
100     // Set config_id for enabling LWS tuning
101     _config_id = kernel_name;
102     _config_id += "_";
103     _config_id += lower_string(string_from_data_type(input->info()->data_type()));
104     _config_id += "_";
105     _config_id += support::cpp11::to_string(input->info()->dimension(0));
106     _config_id += "_";
107     _config_id += support::cpp11::to_string(input->info()->dimension(1));
108     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
109 }
110 
validate(const ITensorInfo * input,const ITensorInfo * output,const FFTScaleKernelInfo & config)111 Status CLFFTScaleKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const FFTScaleKernelInfo &config)
112 {
113     ARM_COMPUTE_UNUSED(config);
114     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
115 
116     return Status{};
117 }
118 
run(const Window & window,cl::CommandQueue & queue)119 void CLFFTScaleKernel::run(const Window &window, cl::CommandQueue &queue)
120 {
121     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
122     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
123 
124     Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
125     Window slice     = collapsed.first_slice_window_3D();
126 
127     do
128     {
129         unsigned int idx = 0;
130         add_3D_tensor_argument(idx, _input, slice);
131         if(!_run_in_place)
132         {
133             add_3D_tensor_argument(idx, _output, slice);
134         }
135         enqueue(queue, *this, slice, lws_hint());
136     }
137     while(collapsed.slide_window_slice_3D(slice));
138 }
139 } // namespace arm_compute
140