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/CLHeightConcatenateLayerKernel.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/Utils.h"
31 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
32 #include "src/core/CL/CLValidate.h"
33 #include "src/core/helpers/WindowHelpers.h"
34 #include "support/Cast.h"
35
36 #include "support/StringSupport.h"
37
38 namespace arm_compute
39 {
40 namespace
41 {
validate_arguments(const ITensorInfo * input,unsigned int height_offset,const ITensorInfo * output)42 Status validate_arguments(const ITensorInfo *input, unsigned int height_offset, const ITensorInfo *output)
43 {
44 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
45 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
46 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
47 ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(Window::DimY) + height_offset > output->dimension(Window::DimY));
48
49 ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != output->dimension(0));
50 for(size_t i = 2; i < Coordinates::num_max_dimensions; ++i)
51 {
52 ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != output->dimension(i));
53 }
54 ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
55
56 return Status{};
57 }
58 } // namespace
59
CLHeightConcatenateLayerKernel()60 CLHeightConcatenateLayerKernel::CLHeightConcatenateLayerKernel()
61 : _height_offset(0)
62 {
63 }
64
validate(const ITensorInfo * input,unsigned int height_offset,const ITensorInfo * output)65 Status CLHeightConcatenateLayerKernel::validate(const ITensorInfo *input, unsigned int height_offset, const ITensorInfo *output)
66 {
67 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, height_offset, output));
68 return Status{};
69 }
70
configure(const CLCompileContext & compile_context,ITensorInfo * input,unsigned int height_offset,ITensorInfo * output)71 void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int height_offset, ITensorInfo *output)
72 {
73 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
74 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, height_offset, output));
75
76 auto padding_info = get_padding_info({ input, output });
77
78 _height_offset = height_offset;
79
80 // Add build options
81 const unsigned int num_elems_processed_per_iteration = adjust_vec_size(4, input->dimension(0));
82
83 CLBuildOptions build_opts;
84 build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->element_size()));
85 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
86 build_opts.add_option("-DHEIGHT_OFFSET=" + support::cpp11::to_string(_height_offset));
87 build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->dimension(2)));
88 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->dimension(0) % num_elems_processed_per_iteration));
89
90 if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
91 {
92 const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
93 const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
94
95 build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
96 build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
97 build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq_info.scale));
98 build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
99 }
100
101 // Create kernel
102 _kernel = create_kernel(compile_context, "concatenate_height", build_opts.options());
103 // Configure kernel window
104
105 // The window needs to be based on input as we copy all the heights of input
106 Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
107 ICLKernel::configure_internal(win.collapse(win, Window::DimZ));
108
109 // Set output valid region
110 output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
111
112 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
113 }
114
run_op(ITensorPack & tensors,const Window & window,cl::CommandQueue & queue)115 void CLHeightConcatenateLayerKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
116 {
117 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
118 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
119
120 const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
121 auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
122
123 unsigned int idx = 0;
124 add_4D_tensor_argument(idx, src, window);
125 add_4D_tensor_argument(idx, dst, window);
126 enqueue(queue, *this, window, lws_hint());
127 }
128 } // namespace arm_compute
129