1 /*
2 * Copyright (c) 2020-2022 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
25 #include "arm_compute/core/Helpers.h"
26 #include "arm_compute/core/ITensor.h"
27 #include "arm_compute/core/Types.h"
28 #include "arm_compute/core/utils/misc/Traits.h"
29 #include "src/core/NEON/SVEMath.h"
30 #include "src/core/NEON/wrapper/intrinsics/intrinsics.h"
31 #include <arm_sve.h>
32
33 namespace arm_compute
34 {
35 namespace cpu
36 {
add_qasymm8_signed_sve2(const ITensor * src0,const ITensor * src1,ITensor * dst,const ConvertPolicy & policy,const Window & window)37 void add_qasymm8_signed_sve2(const ITensor *src0, const ITensor *src1, ITensor *dst, const ConvertPolicy &policy, const Window &window)
38 {
39 ARM_COMPUTE_UNUSED(policy);
40
41 // Create input windows
42 Window input1_win = window.broadcast_if_dimension_le_one(src0->info()->tensor_shape());
43 Window input2_win = window.broadcast_if_dimension_le_one(src1->info()->tensor_shape());
44
45 // Clear X Dimension on execution window as we handle manually
46 Window win = window;
47 win.set(Window::DimX, Window::Dimension(0, 1, 1));
48
49 const auto window_start_x = static_cast<int>(window.x().start());
50 const auto window_end_x = static_cast<int>(window.x().end());
51 const bool is_broadcast_across_x = src0->info()->tensor_shape().x() != src1->info()->tensor_shape().x();
52
53 const UniformQuantizationInfo iq1_info = src0->info()->quantization_info().uniform();
54 const UniformQuantizationInfo iq2_info = src1->info()->quantization_info().uniform();
55 const UniformQuantizationInfo oq_info = dst->info()->quantization_info().uniform();
56
57 const auto invvscaleo = svdup_n_f32(1.f / oq_info.scale);
58 const auto voffseto = svdup_n_f32(oq_info.offset);
59
60 if(is_broadcast_across_x)
61 {
62 const bool is_broadcast_input_2 = input2_win.x().step() == 0;
63 Window broadcast_win = is_broadcast_input_2 ? input2_win : input1_win;
64 Window non_broadcast_win = !is_broadcast_input_2 ? input2_win : input1_win;
65 const ITensor *broadcast_tensor = is_broadcast_input_2 ? src1 : src0;
66 const ITensor *non_broadcast_tensor = !is_broadcast_input_2 ? src1 : src0;
67 const auto all_true_pg = svptrue_b8();
68
69 const auto vscale1 = is_broadcast_input_2 ? svdup_n_f32(iq1_info.scale) : svdup_n_f32(iq2_info.scale);
70 const auto vscale2 = is_broadcast_input_2 ? svdup_n_f32(iq2_info.scale) : svdup_n_f32(iq1_info.scale);
71 const auto voffset1 = is_broadcast_input_2 ? svdup_n_s32(iq1_info.offset) : svdup_n_s32(iq2_info.offset);
72 const auto voffset2 = is_broadcast_input_2 ? svdup_n_s32(iq2_info.offset) : svdup_n_s32(iq1_info.offset);
73
74 // Clear X Dimension on execution window as we handle manually
75 non_broadcast_win.set(Window::DimX, Window::Dimension(0, 1, 1));
76
77 Iterator broadcast_input(broadcast_tensor, broadcast_win);
78 Iterator non_broadcast_input(non_broadcast_tensor, non_broadcast_win);
79 Iterator output(dst, win);
80
81 execute_window_loop(win, [&](const Coordinates &)
82 {
83 const auto non_broadcast_input_ptr = reinterpret_cast<const int8_t *>(non_broadcast_input.ptr());
84 const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr());
85
86 const int8_t broadcast_value = *reinterpret_cast<const int8_t *>(broadcast_input.ptr());
87 const auto broadcast_value_vec = svdup_n_s8(broadcast_value);
88
89 int x = window_start_x;
90 svbool_t pg = svwhilelt_b8(x, window_end_x);
91 const auto bf_0 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlb_s16(broadcast_value_vec)), voffset2)), vscale2);
92 const auto bf_1 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlb_s16(broadcast_value_vec)), voffset2)), vscale2);
93 const auto bf_2 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlt_s16(broadcast_value_vec)), voffset2)), vscale2);
94 const auto bf_3 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlt_s16(broadcast_value_vec)), voffset2)), vscale2);
95
96 do
97 {
98 const auto a = svld1_s8(pg, non_broadcast_input_ptr + x);
99 const auto af_0 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlb_s16(a)), voffset1)), vscale1);
100 const auto af_1 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlb_s16(a)), voffset1)), vscale1);
101 const auto af_2 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlt_s16(a)), voffset1)), vscale1);
102 const auto af_3 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlt_s16(a)), voffset1)), vscale1);
103
104 const auto rf_0 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_0, bf_0), invvscaleo));
105 const auto rf_1 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_1, bf_1), invvscaleo));
106 const auto rf_2 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_2, bf_2), invvscaleo));
107 const auto rf_3 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_3, bf_3), invvscaleo));
108
109 const auto pa = svqxtnt_s32(svqxtnb_s32(rf_0), rf_1);
110 const auto pb = svqxtnt_s32(svqxtnb_s32(rf_2), rf_3);
111 const auto res = svqxtnt_s16(svqxtnb_s16(pa), pb);
112
113 svst1_s8(pg, output_ptr + x, res);
114
115 x += svcntb();
116 pg = svwhilelt_b8(x, window_end_x);
117 }
118 while(svptest_any(all_true_pg, pg));
119 },
120 broadcast_input, non_broadcast_input, output);
121 }
122 else
123 {
124 // Clear X Dimension on execution window as we handle manually
125 input1_win.set(Window::DimX, Window::Dimension(0, 1, 1));
126 input2_win.set(Window::DimX, Window::Dimension(0, 1, 1));
127
128 Iterator input1(src0, input1_win);
129 Iterator input2(src1, input2_win);
130 Iterator output(dst, win);
131
132 const auto vscale1 = svdup_n_f32(iq1_info.scale);
133 const auto vscale2 = svdup_n_f32(iq2_info.scale);
134 const auto voffset1 = svdup_n_s32(iq1_info.offset);
135 const auto voffset2 = svdup_n_s32(iq2_info.offset);
136
137 execute_window_loop(win, [&](const Coordinates &)
138 {
139 const auto input1_ptr = reinterpret_cast<const int8_t *>(input1.ptr());
140 const auto input2_ptr = reinterpret_cast<const int8_t *>(input2.ptr());
141 const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr());
142
143 int x = window_start_x;
144 svbool_t pg = svwhilelt_b8(x, window_end_x);
145 do
146 {
147 const auto a = svld1_s8(pg, input1_ptr + x);
148 const auto b = svld1_s8(pg, input2_ptr + x);
149
150 const auto af_0 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlb_s16(a)), voffset1)), vscale1);
151 const auto af_1 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlb_s16(a)), voffset1)), vscale1);
152 const auto af_2 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlt_s16(a)), voffset1)), vscale1);
153 const auto af_3 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlt_s16(a)), voffset1)), vscale1);
154
155 const auto bf_0 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlb_s16(b)), voffset2)), vscale2);
156 const auto bf_1 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlb_s16(b)), voffset2)), vscale2);
157 const auto bf_2 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlb_s32(svmovlt_s16(b)), voffset2)), vscale2);
158 const auto bf_3 = svmul_f32_z(pg, svcvt_f32_s32_z(pg, svsub_s32_z(pg, svmovlt_s32(svmovlt_s16(b)), voffset2)), vscale2);
159
160 const auto rf_0 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_0, bf_0), invvscaleo));
161 const auto rf_1 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_1, bf_1), invvscaleo));
162 const auto rf_2 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_2, bf_2), invvscaleo));
163 const auto rf_3 = svcvt_s32_f32_z(pg, svmla_f32_z(pg, voffseto, svadd_f32_z(pg, af_3, bf_3), invvscaleo));
164
165 const auto pa = svqxtnt_s32(svqxtnb_s32(rf_0), rf_1);
166 const auto pb = svqxtnt_s32(svqxtnb_s32(rf_2), rf_3);
167 const auto res = svqxtnt_s16(svqxtnb_s16(pa), pb);
168
169 svst1_s8(pg, output_ptr + x, res);
170
171 x += svcntb();
172 pg = svwhilelt_b8(x, window_end_x);
173 }
174 while(svptest_any(svptrue_b8(), pg));
175 },
176 input1, input2, output);
177 }
178 }
179 } // namespace cpu
180 } // namespace arm_compute
181