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/CLRangeKernel.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/ICLTensor.h"
28 #include "arm_compute/core/Utils.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 {
38 constexpr unsigned int vector_size_byte_opencl = 16;
39
validate_arguments(const ITensorInfo * output,const float start,const float end,const float step)40 Status validate_arguments(const ITensorInfo *output, const float start, const float end, const float step)
41 {
42 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
43 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output,
44 1,
45 DataType::U8, DataType::S8, DataType::QASYMM8,
46 DataType::U16, DataType::S16,
47 DataType::U32, DataType::S32,
48 DataType::F16, DataType::F32);
49 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(output);
50
51 ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
52 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start < end) && (step <= 0)), "step must be greater than 0 when start < end");
53 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start > end) && (step >= 0)), "step must be less than 0 when start > end");
54
55 ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
56
57 ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(start, output->data_type(), output->quantization_info()), "start value is outside the range of the data type");
58 ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(end, output->data_type(), output->quantization_info()), "end value is outside the range of the data type");
59 ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(step, output->data_type(), output->quantization_info()), "step value is outside the range of the data type");
60
61 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->num_dimensions() != 1, "Output has to be a 1-D tensor");
62 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->tensor_shape().total_size() < num_of_elements_in_range(start, end, step), "Output tensor size is incorrect");
63
64 return Status{};
65 }
66 } // namespace
67
CLRangeKernel()68 CLRangeKernel::CLRangeKernel()
69 : _start(0), _end(1), _step(1), _output(nullptr)
70 {
71 _type = CLKernelType::ELEMENTWISE;
72 }
73
configure(ICLTensor * output,const float start,const float end,const float step)74 void CLRangeKernel::configure(ICLTensor *output, const float start, const float end, const float step)
75 {
76 configure(CLKernelLibrary::get().get_compile_context(), output, start, end, step);
77 }
78
configure(const CLCompileContext & compile_context,ICLTensor * output,const float start,const float end,const float step)79 void CLRangeKernel::configure(const CLCompileContext &compile_context, ICLTensor *output, const float start, const float end, const float step)
80 {
81 ARM_COMPUTE_ERROR_ON_NULLPTR(output);
82 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(output->info(), start, end, step));
83
84 // Configure kernel window
85 unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / output->info()->element_size(), output->info()->dimension(0));
86 Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
87
88 auto padding_info = get_padding_info({ output });
89
90 _start = start;
91 _end = end;
92 _step = step;
93 _output = output;
94
95 std::string kernel_name = "range";
96
97 // Set build options
98 CLBuildOptions build_opts;
99 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
100 build_opts.add_option("-DVECTOR_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
101 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(output->info()->dimension(0) % num_elems_processed_per_iteration));
102 build_opts.add_option("-DSTART=" + support::cpp11::to_string(start));
103 build_opts.add_option("-DSTEP=" + support::cpp11::to_string(step));
104 if(is_data_type_quantized_asymmetric(output->info()->data_type()))
105 {
106 const UniformQuantizationInfo qinfo = output->info()->quantization_info().uniform();
107 build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(qinfo.offset));
108 build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(qinfo.scale));
109 kernel_name += "_quantized";
110 }
111 // Create kernel
112 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
113 ICLKernel::configure_internal(win);
114
115 // Set config_id for enabling LWS tuning
116 _config_id = kernel_name;
117 _config_id += "_";
118 _config_id += lower_string(string_from_data_type(output->info()->data_type()));
119 _config_id += "_";
120 _config_id += support::cpp11::to_string(output->info()->dimension(0));
121 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
122 }
123
validate(const ITensorInfo * output,const float start,const float end,const float step)124 Status CLRangeKernel::validate(const ITensorInfo *output, const float start, const float end, const float step)
125 {
126 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(output, start, end, step));
127 return Status{};
128 }
129
run(const Window & window,cl::CommandQueue & queue)130 void CLRangeKernel::run(const Window &window, cl::CommandQueue &queue)
131 {
132 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
133 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
134 unsigned int idx = 0;
135 add_1D_tensor_argument(idx, _output, window);
136
137 enqueue(queue, *this, window, lws_hint());
138 }
139 } // namespace arm_compute