1 /*
2 * Copyright (c) 2018-2022 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/ClWidthConcatenate2TensorsKernel.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/Utils.h"
30 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
31 #include "src/core/CL/CLValidate.h"
32 #include "src/core/helpers/WindowHelpers.h"
33 #include "src/core/utils/helpers/tensor_info.h"
34 #include "support/Cast.h"
35
36 #include "support/StringSupport.h"
37
38 namespace arm_compute
39 {
40 namespace opencl
41 {
42 namespace kernels
43 {
44 namespace
45 {
validate_arguments(const ITensorInfo * src1,const ITensorInfo * src2,const ITensorInfo * dst)46 Status validate_arguments(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
47 {
48 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
49 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src1);
50 ARM_COMPUTE_RETURN_ERROR_ON(src1->data_type() == DataType::UNKNOWN);
51 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src1, src2, dst);
52 ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(0) + src2->dimension(0) > dst->dimension(0));
53
54 for(size_t i = 1; i < Coordinates::num_max_dimensions; ++i)
55 {
56 ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(i) != dst->dimension(i));
57 ARM_COMPUTE_RETURN_ERROR_ON(src2->dimension(i) != dst->dimension(i));
58 }
59 ARM_COMPUTE_RETURN_ERROR_ON(src1->num_dimensions() > 4);
60
61 return Status{};
62 }
63 } // namespace
64
validate(const ITensorInfo * src1,const ITensorInfo * src2,const ITensorInfo * dst)65 Status ClWidthConcatenate2TensorsKernel::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
66 {
67 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src1, src2, dst));
68 return Status{};
69 }
70
ClWidthConcatenate2TensorsKernel()71 ClWidthConcatenate2TensorsKernel::ClWidthConcatenate2TensorsKernel()
72 {
73 _type = CLKernelType::ELEMENTWISE;
74 }
75
configure(const CLCompileContext & compile_context,ITensorInfo * src1,ITensorInfo * src2,ITensorInfo * dst)76 void ClWidthConcatenate2TensorsKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
77 {
78 ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
79 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src1, src2, dst));
80
81 auto padding_info = get_padding_info({ src1, src2, dst });
82
83 const unsigned int min_dimension = std::min(src1->dimension(0), src2->dimension(0));
84 const unsigned int num_elems_processed_per_iteration = adjust_vec_size(8, min_dimension);
85 const unsigned int vec_size_leftover = dst->dimension(0) % num_elems_processed_per_iteration;
86
87 // Add build options
88 CLBuildOptions build_opts;
89 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src1->data_type()));
90 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
91 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_leftover));
92 build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(src1->element_size()));
93 build_opts.add_option("-DINPUT1_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) - vec_size_leftover) % num_elems_processed_per_iteration));
94
95 // If input have different quantization info set quantization parameters needed for the re-quantization process
96 const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(dst, src1, src2);
97 if(is_data_type_quantized_asymmetric(src1->data_type()) && have_different_qinfo)
98 {
99 const UniformQuantizationInfo iq1_info = src1->quantization_info().uniform();
100 const UniformQuantizationInfo iq2_info = src2->quantization_info().uniform();
101 const UniformQuantizationInfo oq_info = dst->quantization_info().uniform();
102
103 build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
104 build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
105 build_opts.add_option("-DOFFSET_IN2=" + float_to_string_with_full_precision(iq2_info.offset));
106 build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2_info.scale));
107 build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
108 build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
109 }
110
111 _depth = src1->dimension(2);
112 _input1_width = src1->dimension(0);
113
114 std::string kernel_name = "concatenate_width_x2";
115
116 // A macro guard to compile ONLY the kernel of interest
117 build_opts.add_option("-D" + upper_string(kernel_name));
118
119 // Create kernel
120 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
121
122 // Configure kernel window
123 Window win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration));
124 ICLKernel::configure_internal(win.collapse(win, Window::DimZ));
125
126 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
127
128 // Set config_id for enabling LWS tuning
129 _config_id = "concatenate_width_x2_";
130 _config_id += lower_string(string_from_data_type(src1->data_type()));
131 _config_id += "_";
132 _config_id += support::cpp11::to_string(src1->dimension(0));
133 _config_id += "_";
134 _config_id += support::cpp11::to_string(src1->dimension(1));
135 _config_id += "_";
136 _config_id += support::cpp11::to_string(src2->dimension(0));
137 _config_id += "_";
138 _config_id += support::cpp11::to_string(src2->dimension(1));
139 }
140
run_op(ITensorPack & tensors,const Window & window,::cl::CommandQueue & queue)141 void ClWidthConcatenate2TensorsKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
142 {
143 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
144 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
145
146 Window slice = window.first_slice_window_4D();
147
148 const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC));
149 const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 1));
150 auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
151
152 do
153 {
154 unsigned int idx = 0;
155 add_4D_tensor_argument(idx, src0, slice);
156 add_4D_tensor_argument(idx, src1, slice);
157 add_4D_tensor_argument(idx, dst, slice);
158 _kernel.setArg<cl_int>(idx++, _depth);
159 _kernel.setArg<cl_int>(idx++, _input1_width);
160 enqueue(queue, *this, window, lws_hint());
161 }
162 while(window.slide_window_slice_4D(slice));
163 }
164 } // namespace kernels
165 } // namespace opencl
166 } // namespace arm_compute
167