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/gpu/cl/kernels/ClCopyKernel.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 "arm_compute/core/Utils.h"
32 #include "arm_compute/core/Validate.h"
33 #include "src/core/CL/CLValidate.h"
34 #include "src/core/helpers/AutoConfiguration.h"
35 #include "src/core/helpers/WindowHelpers.h"
36 #include "support/Cast.h"
37 #include "support/StringSupport.h"
38
39 namespace arm_compute
40 {
41 namespace opencl
42 {
43 namespace kernels
44 {
45 namespace
46 {
validate_arguments(const ITensorInfo * src,const ITensorInfo * dst,Window * dst_window=nullptr)47 Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, Window *dst_window = nullptr)
48 {
49 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
50
51 // Validate dst if initialized
52 if(dst->total_size() != 0)
53 {
54 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
55 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
56 if(dst_window == nullptr)
57 {
58 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(src->tensor_shape(), dst->tensor_shape());
59 }
60 else
61 {
62 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(src->tensor_shape(), dst_window->shape());
63 }
64 }
65
66 return Status{};
67 }
68
69 } // namespace
70
ClCopyKernel()71 ClCopyKernel::ClCopyKernel()
72 {
73 _type = CLKernelType::ELEMENTWISE;
74 }
75
configure(const CLCompileContext & compile_context,const ITensorInfo * src,ITensorInfo * dst,Window * dst_window)76 void ClCopyKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Window *dst_window)
77 {
78 ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
79 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, dst_window));
80
81 auto padding_info = get_padding_info({ src, dst });
82
83 // Create kernel
84 CLBuildOptions build_opts;
85 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));
86
87 // Output auto inizialitation if not yet initialized
88 auto_init_if_empty(*dst, *src);
89
90 // Configure window
91 const unsigned int vec_size_x = adjust_vec_size(16 / src->element_size(), src->dimension(0));
92
93 const Window win_config = calculate_max_window(*src, Steps(vec_size_x));
94
95 if(dst_window != nullptr)
96 {
97 _has_dst_window = true;
98 _dst_window = Window(*dst_window);
99 const int width_x = dst_window->num_iterations(0);
100 const int vec_size_x_leftover = width_x % vec_size_x;
101 const bool multi_access_x = width_x >= static_cast<int32_t>(vec_size_x);
102
103 if(multi_access_x)
104 {
105 _dst_window.set(Window::DimX, Window::Dimension(dst_window->x().start(), ceil_to_multiple(dst_window->x().end(), vec_size_x), vec_size_x));
106 }
107
108 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
109 }
110 else
111 {
112 const int width_x = src->tensor_shape().x();
113 const int vec_size_x_leftover = width_x % vec_size_x;
114
115 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
116 }
117
118 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
119
120 // Build kernel
121 _kernel = create_kernel(compile_context, "copy_tensor", build_opts.options());
122
123 // Validate and set the window
124 ICLKernel::configure_internal(win_config);
125
126 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
127 }
128
validate(const arm_compute::ITensorInfo * src,const arm_compute::ITensorInfo * dst,Window * dst_window)129 Status ClCopyKernel::validate(const arm_compute::ITensorInfo *src, const arm_compute::ITensorInfo *dst, Window *dst_window)
130 {
131 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, dst_window));
132
133 return Status{};
134 }
135
run_op(ITensorPack & tensors,const Window & window,cl::CommandQueue & queue)136 void ClCopyKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
137 {
138 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
139 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
140
141 const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
142 auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
143
144 Window slice;
145
146 if(_has_dst_window)
147 {
148 slice = window.first_slice_window_3D();
149 Window out_slice = _dst_window.first_slice_window_3D();
150 do
151 {
152 unsigned int idx = 0;
153 add_3D_tensor_argument(idx, src, slice);
154 add_3D_tensor_argument(idx, dst, out_slice);
155 enqueue(queue, *this, slice, lws_hint());
156 }
157 while(window.slide_window_slice_3D(slice) && _dst_window.slide_window_slice_3D(out_slice));
158 }
159 else
160 {
161 Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
162 slice = collapsed.first_slice_window_3D();
163 do
164 {
165 unsigned int idx = 0;
166 add_3D_tensor_argument(idx, src, slice);
167 add_3D_tensor_argument(idx, dst, slice);
168 enqueue(queue, *this, slice, lws_hint());
169 }
170 while(collapsed.slide_window_slice_3D(slice));
171 }
172 }
173 } // namespace kernels
174 } // namespace opencl
175 } // namespace arm_compute
176