1 /*
2 * Copyright (c) 2022-2023 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 "ClTemplateCast.h"
25
26 #include "src/core/helpers/WindowHelpers.h"
27 #include "src/dynamic_fusion/sketch/gpu/GpuKernelComponentGroup.h"
28
29 namespace arm_compute
30 {
31 namespace experimental
32 {
33 namespace dynamic_fusion
34 {
ClTemplateCast(ComponentId id,const ArgumentPack<ITensorInfo> & tensors,const Attributes & attributes)35 ClTemplateCast::ClTemplateCast(ComponentId id, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes)
36 : IGpuTemplateComponentWriter{ id, tensors }, _src{}, _dst{}, _attributes{ attributes }
37 {
38 _src = this->tensors().get_const_tensor(TensorType::ACL_SRC_0);
39 _dst = this->tensors().get_const_tensor(TensorType::ACL_DST_0);
40
41 ARM_COMPUTE_ERROR_ON_NULLPTR(_src, _dst);
42 }
43
get_name() const44 std::string ClTemplateCast::get_name() const
45 {
46 const size_t src_size = data_size_from_type(_src->data_type());
47 const size_t dst_size = data_size_from_type(_dst->data_type());
48
49 return (src_size >= dst_size) ? "cast_down" : "cast_up";
50 }
51
get_component_code(const ComponentGroup & comp_group) const52 std::string ClTemplateCast::get_component_code(const ComponentGroup &comp_group) const
53 {
54 ARM_COMPUTE_UNUSED(comp_group);
55
56 const std::string kernel_name = get_name();
57 const auto is_root = (comp_group.get_root_component()->id() == this->id());
58
59 std::string code = R"_(
60 //------------------ START KERNEL {{meta_kernel_id}} CAST ---------------------
61 )_";
62
63 if(is_root)
64 {
65 code += R"_(
66 // IN_0(src) {{src}}
67 // OUT(dst, accum) {{dst}}
68
69 TILE(uint, M0, 1, g_dst_indirect_y);
70 {
71 {{src}}_offset_first_element_in_bytes += get_global_id(2) * {{src}}_stride_z;
72
73 TILE({{DATA_TYPE_IN}}, M0, N0, {{tmp}});
74 T_LOAD({{DATA_TYPE_IN}}, M0, N0, BUFFER, {{src}}, g_ind_0, g_ind_1, 1, {{src}}_stride_y, {{tmp}});
75 )_";
76 }
77
78 code += R"_(
79 LOOP_UNROLLING(int, m0, 0, 1, M0,
80 {
81 )_";
82
83 if(kernel_name == "cast_down" && is_data_type_quantized(_src->data_type()))
84 {
85 code += R"_(
86 {{tmp}}[m0].v ^= (VEC_DATA_TYPE({{DATA_TYPE_IN}}, N0))0x80;
87 )_";
88 }
89
90 if(kernel_name == "cast_down" && (is_data_type_float(_src->data_type()) || _attributes.convert_policy() == ConvertPolicy::SATURATE))
91 {
92 code += R"_(
93 {{dst}}[m0].v = CONVERT_SAT({{tmp}}[m0].v, VEC_DATA_TYPE({{DATA_TYPE_OUT}}, N0));
94 )_";
95 }
96 else
97 {
98 code += R"_(
99 {{dst}}[m0].v = CONVERT({{tmp}}[m0].v, VEC_DATA_TYPE({{DATA_TYPE_OUT}}, N0));
100 )_";
101 }
102
103 code += R"_(
104 })
105 )_";
106
107 if(is_root)
108 {
109 code += R"_(
110 LOOP_UNROLLING(int, i, 0, 1, M0,
111 {
112 g_dst_indirect_y[i].v = (uint)min((int)(g_ind_1 + i), (int)({{arg_dst}}_w) - 1);
113 g_dst_indirect_y[i].v += (int)(g_ind_2 % {{arg_dst}}_h) * (int)({{arg_dst}}_w);
114 g_dst_indirect_y[i].v += (int)(g_ind_2 / {{arg_dst}}_h) * (int)({{arg_dst}}_w * {{arg_dst}}_h);
115 })
116 }
117 )_";
118 }
119
120 code += R"_(
121 //------------------ END KERNEL {{meta_kernel_id}} CAST ---------------------
122 )_";
123
124 return code;
125 }
126
declare_variables(GpuKernelVariableTable & vtable,const ComponentGroup & comp_group) const127 void ClTemplateCast::declare_variables(GpuKernelVariableTable &vtable, const ComponentGroup &comp_group) const
128 {
129 vtable.declare_variable(
130 comp_group,
131 _src,
132 GpuKernelArgumentInfo(GpuKernelArgumentInfo::Type::Tensor_4D_t_Buffer),
133 "src");
134
135 vtable.declare_variable(
136 comp_group,
137 _dst,
138 GpuKernelArgumentInfo(GpuKernelArgumentInfo::Type::Tensor_4D_t_Buffer),
139 "dst");
140 }
141
get_tag_lut(const GpuKernelVariableTable & vtable,const ComponentGroup & comp_group) const142 TagLUT ClTemplateCast::get_tag_lut(const GpuKernelVariableTable &vtable, const ComponentGroup &comp_group) const
143 {
144 const auto is_root = (comp_group.get_root_component()->id() == this->id());
145
146 TagLUT lut{};
147
148 // Arguments and global shared variables
149 lut["src"] = vtable.get_variable(_src);
150 lut["dst"] = vtable.get_variable(_dst);
151 lut["tmp"] = (is_root) ? lut["src"].value + "_in_data" : lut["src"];
152
153 const auto dst_argument = vtable.get_variable(comp_group.get_any_dst_tensor());
154 lut["arg_dst"] = dst_argument.uniq_name;
155
156 // Local build options
157 lut["meta_kernel_id"] = id();
158
159 lut["DATA_TYPE_IN"] = get_cl_type_from_data_type(_src->data_type());
160 lut["DATA_TYPE_OUT"] = get_cl_type_from_data_type(_dst->data_type());
161
162 return lut;
163 }
164
get_build_options(const ComponentGroup & comp_group) const165 CLBuildOptions ClTemplateCast::get_build_options(const ComponentGroup &comp_group) const
166 {
167 ARM_COMPUTE_UNUSED(comp_group);
168
169 const auto root_window = comp_group.get_root_component()->template_writer()->get_window();
170 const unsigned int n0 = root_window.x().step();
171 const unsigned int m0 = root_window.y().step();
172
173 // Set build options
174 CLBuildOptions build_opts{};
175 build_opts.add_option("-DN0=" + support::cpp11::to_string(n0));
176 build_opts.add_option("-DPARTIAL_N0=" + support::cpp11::to_string(_src->dimension(0) % n0));
177 build_opts.add_option("-DM0=" + support::cpp11::to_string(m0));
178
179 return build_opts;
180 }
181
get_config_id() const182 std::string ClTemplateCast::get_config_id() const
183 {
184 std::string config_id{};
185
186 config_id += "_";
187 config_id += lower_string(string_from_data_type(_src->data_type()));
188 config_id += "_";
189 config_id += lower_string(string_from_data_type(_dst->data_type()));
190 config_id += "_";
191 config_id += support::cpp11::to_string(_src->dimension(0));
192 config_id += "_";
193 config_id += support::cpp11::to_string(_src->dimension(1));
194
195 return config_id;
196 }
197
get_headers_list() const198 std::set<std::string> ClTemplateCast::get_headers_list() const
199 {
200 return std::set<std::string>{ "helpers.h", "tile_helpers.h" };
201 }
202
get_window() const203 Window ClTemplateCast::get_window() const
204 {
205 ARM_COMPUTE_ERROR_ON_MSG(_dst->tensor_shape().total_size() == 0U, "Destination tensor is not initialized");
206
207 const unsigned int n0 = adjust_vec_size(16 / _dst->element_size(), _dst->dimension(0));
208 Window win = calculate_max_window(*_dst, Steps(n0));
209 return win.collapse(win, Window::DimZ);
210 }
211
212 } // namespace dynamic_fusion
213 } // namespace experimental
214 } // namespace arm_compute
215