mdds
block_util.hpp
1 /* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
2 
3 // SPDX-FileCopyrightText: 2021 - 2025 Kohei Yoshida
4 //
5 // SPDX-License-Identifier: MIT
6 
7 #pragma once
8 
9 #include "mdds/global.hpp"
10 #include "../types.hpp"
11 
12 namespace mdds { namespace mtv { namespace aos { namespace detail {
13 
14 template<typename Blks, lu_factor_t F>
15 struct adjust_block_positions
16 {
17  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
18  {
19  static_assert(
20  mdds::detail::invalid_static_int<F>, "The loop-unrolling factor must be one of 0, 4, 8, 16, or 32.");
21  }
22 };
23 
24 template<typename Blks>
25 struct adjust_block_positions<Blks, lu_factor_t::none>
26 {
27  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
28  {
29  int64_t n = blocks.size();
30 
31  if (start_block_index >= n)
32  return;
33 
34 #if MDDS_USE_OPENMP
35 #pragma omp parallel for
36 #endif
37  for (int64_t i = start_block_index; i < n; ++i)
38  blocks[i].position += delta;
39  }
40 };
41 
42 template<typename Blks>
43 struct adjust_block_positions<Blks, lu_factor_t::lu4>
44 {
45  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
46  {
47  int64_t n = blocks.size();
48 
49  if (start_block_index >= n)
50  return;
51 
52  // Ensure that the section length is divisible by 4.
53  int64_t len = n - start_block_index;
54  int64_t rem = len & 3; // % 4
55  len -= rem;
56  len += start_block_index;
57 #if MDDS_USE_OPENMP
58 #pragma omp parallel for
59 #endif
60  for (int64_t i = start_block_index; i < len; i += 4)
61  {
62  blocks[i].position += delta;
63  blocks[i + 1].position += delta;
64  blocks[i + 2].position += delta;
65  blocks[i + 3].position += delta;
66  }
67 
68  rem += len;
69  for (int64_t i = len; i < rem; ++i)
70  blocks[i].position += delta;
71  }
72 };
73 
74 template<typename Blks>
75 struct adjust_block_positions<Blks, lu_factor_t::lu8>
76 {
77  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
78  {
79  int64_t n = blocks.size();
80 
81  if (start_block_index >= n)
82  return;
83 
84  // Ensure that the section length is divisible by 8.
85  int64_t len = n - start_block_index;
86  int64_t rem = len & 7; // % 8
87  len -= rem;
88  len += start_block_index;
89 #if MDDS_USE_OPENMP
90 #pragma omp parallel for
91 #endif
92  for (int64_t i = start_block_index; i < len; i += 8)
93  {
94  blocks[i].position += delta;
95  blocks[i + 1].position += delta;
96  blocks[i + 2].position += delta;
97  blocks[i + 3].position += delta;
98  blocks[i + 4].position += delta;
99  blocks[i + 5].position += delta;
100  blocks[i + 6].position += delta;
101  blocks[i + 7].position += delta;
102  }
103 
104  rem += len;
105  for (int64_t i = len; i < rem; ++i)
106  blocks[i].position += delta;
107  }
108 };
109 
110 template<typename Blks>
111 struct adjust_block_positions<Blks, lu_factor_t::lu16>
112 {
113  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
114  {
115  int64_t n = blocks.size();
116 
117  if (start_block_index >= n)
118  return;
119 
120  // Ensure that the section length is divisible by 16.
121  int64_t len = n - start_block_index;
122  int64_t rem = len & 15; // % 16
123  len -= rem;
124  len += start_block_index;
125 #if MDDS_USE_OPENMP
126 #pragma omp parallel for
127 #endif
128  for (int64_t i = start_block_index; i < len; i += 16)
129  {
130  blocks[i].position += delta;
131  blocks[i + 1].position += delta;
132  blocks[i + 2].position += delta;
133  blocks[i + 3].position += delta;
134  blocks[i + 4].position += delta;
135  blocks[i + 5].position += delta;
136  blocks[i + 6].position += delta;
137  blocks[i + 7].position += delta;
138  blocks[i + 8].position += delta;
139  blocks[i + 9].position += delta;
140  blocks[i + 10].position += delta;
141  blocks[i + 11].position += delta;
142  blocks[i + 12].position += delta;
143  blocks[i + 13].position += delta;
144  blocks[i + 14].position += delta;
145  blocks[i + 15].position += delta;
146  }
147 
148  rem += len;
149  for (int64_t i = len; i < rem; ++i)
150  blocks[i].position += delta;
151  }
152 };
153 
154 template<typename Blks>
155 struct adjust_block_positions<Blks, lu_factor_t::lu32>
156 {
157  void operator()(Blks& blocks, int64_t start_block_index, int64_t delta) const
158  {
159  int64_t n = blocks.size();
160 
161  if (start_block_index >= n)
162  return;
163 
164  // Ensure that the section length is divisible by 32.
165  int64_t len = n - start_block_index;
166  int64_t rem = len & 31; // % 32
167  len -= rem;
168  len += start_block_index;
169 #if MDDS_USE_OPENMP
170 #pragma omp parallel for
171 #endif
172  for (int64_t i = start_block_index; i < len; i += 32)
173  {
174  blocks[i].position += delta;
175  blocks[i + 1].position += delta;
176  blocks[i + 2].position += delta;
177  blocks[i + 3].position += delta;
178  blocks[i + 4].position += delta;
179  blocks[i + 5].position += delta;
180  blocks[i + 6].position += delta;
181  blocks[i + 7].position += delta;
182  blocks[i + 8].position += delta;
183  blocks[i + 9].position += delta;
184  blocks[i + 10].position += delta;
185  blocks[i + 11].position += delta;
186  blocks[i + 12].position += delta;
187  blocks[i + 13].position += delta;
188  blocks[i + 14].position += delta;
189  blocks[i + 15].position += delta;
190  blocks[i + 16].position += delta;
191  blocks[i + 17].position += delta;
192  blocks[i + 18].position += delta;
193  blocks[i + 19].position += delta;
194  blocks[i + 20].position += delta;
195  blocks[i + 21].position += delta;
196  blocks[i + 22].position += delta;
197  blocks[i + 23].position += delta;
198  blocks[i + 24].position += delta;
199  blocks[i + 25].position += delta;
200  blocks[i + 26].position += delta;
201  blocks[i + 27].position += delta;
202  blocks[i + 28].position += delta;
203  blocks[i + 29].position += delta;
204  blocks[i + 30].position += delta;
205  blocks[i + 31].position += delta;
206  }
207 
208  rem += len;
209  for (int64_t i = len; i < rem; ++i)
210  blocks[i].position += delta;
211  }
212 };
213 
214 }}}} // namespace mdds::mtv::aos::detail
215 
216 /* vim:set shiftwidth=4 softtabstop=4 expandtab: */
mdds::mtv::aos::detail::adjust_block_positions
Definition: block_util.hpp:15
mdds
Definition: cref_wrapper.hpp:12
Generated on Thu Oct 15 2026 20:13:21 for mdds by   doxygen 1.8.10