1 | /** @file
|
---|
2 | * IPRT / No-CRT - Minimal C++ algorithm header.
|
---|
3 | */
|
---|
4 |
|
---|
5 | /*
|
---|
6 | * Copyright (C) 2022-2023 Oracle and/or its affiliates.
|
---|
7 | *
|
---|
8 | * This file is part of VirtualBox base platform packages, as
|
---|
9 | * available from https://www.virtualbox.org.
|
---|
10 | *
|
---|
11 | * This program is free software; you can redistribute it and/or
|
---|
12 | * modify it under the terms of the GNU General Public License
|
---|
13 | * as published by the Free Software Foundation, in version 3 of the
|
---|
14 | * License.
|
---|
15 | *
|
---|
16 | * This program is distributed in the hope that it will be useful, but
|
---|
17 | * WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
19 | * General Public License for more details.
|
---|
20 | *
|
---|
21 | * You should have received a copy of the GNU General Public License
|
---|
22 | * along with this program; if not, see <https://www.gnu.org/licenses>.
|
---|
23 | *
|
---|
24 | * The contents of this file may alternatively be used under the terms
|
---|
25 | * of the Common Development and Distribution License Version 1.0
|
---|
26 | * (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
|
---|
27 | * in the VirtualBox distribution, in which case the provisions of the
|
---|
28 | * CDDL are applicable instead of those of the GPL.
|
---|
29 | *
|
---|
30 | * You may elect to license modified versions of this file under the
|
---|
31 | * terms and conditions of either the GPL or the CDDL or both.
|
---|
32 | *
|
---|
33 | * SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
|
---|
34 | */
|
---|
35 |
|
---|
36 | #ifndef VBOX_INCLUDED_SRC_nocrt_algorithm
|
---|
37 | #define VBOX_INCLUDED_SRC_nocrt_algorithm
|
---|
38 | #ifndef RT_WITHOUT_PRAGMA_ONCE
|
---|
39 | # pragma once
|
---|
40 | #endif
|
---|
41 |
|
---|
42 | #ifdef _MSC_VER
|
---|
43 | # pragma warning(push)
|
---|
44 | # pragma warning(disable: 4643) /* warning C4643: Forward declaring 'ios_base' in namespace std is not permitted by the C++ Standard */
|
---|
45 | #endif
|
---|
46 |
|
---|
47 | namespace std
|
---|
48 | {
|
---|
49 | /**
|
---|
50 | * Swap the values pointed to by the two references.
|
---|
51 | */
|
---|
52 | template<typename a_Type>
|
---|
53 | void swap(a_Type &a_rObj1, a_Type &a_rObj2)
|
---|
54 | {
|
---|
55 | a_Type Tmp(a_rObj1);
|
---|
56 | a_rObj1 = a_rObj2;
|
---|
57 | a_rObj2 = Tmp;
|
---|
58 | }
|
---|
59 |
|
---|
60 | /**
|
---|
61 | * Swap the values pointed to by two forward iterators.
|
---|
62 | */
|
---|
63 | template<typename a_ForwardIt1, typename a_ForwardIt2>
|
---|
64 | void iter_swap(a_ForwardIt1 a_It1, a_ForwardIt2 a_It2)
|
---|
65 | {
|
---|
66 | swap(*a_It1, *a_It2);
|
---|
67 | }
|
---|
68 |
|
---|
69 | template<typename a_RandomIt>
|
---|
70 | void sort(a_RandomIt a_ItBegin, a_RandomIt a_ItEnd)
|
---|
71 | {
|
---|
72 | /* Note! Using shell sort here because it's tiny and we've got code for it. */
|
---|
73 | /** @todo replace with faster code. */
|
---|
74 |
|
---|
75 | /* Anything worth sorting? */
|
---|
76 | std::size_t const cElements = a_ItEnd - a_ItBegin;
|
---|
77 | if (cElements >= 1)
|
---|
78 | {
|
---|
79 | /* Loop on decreasing gap, ending with 1: */
|
---|
80 | std::size_t cGap = (cElements + 1) / 2;
|
---|
81 | while (cGap > 0)
|
---|
82 | {
|
---|
83 | /* Iterate from cGap till the end: */
|
---|
84 | for (std::size_t i = cGap; i < cElements; i++)
|
---|
85 | {
|
---|
86 | /* Find the best suitable location for the item at 'i' comparing
|
---|
87 | backwards in steps of 'cGap', swapping the item at 'i' with the
|
---|
88 | one at '-cGap*j' if it's smaller, stopping if it's larger.
|
---|
89 |
|
---|
90 | Note! Original algorithm would make a copy of the item, this version
|
---|
91 | avoids extra copies of sorted items at the cost of extra copies
|
---|
92 | when dealing with unsorted ones a small cGaps values. */
|
---|
93 | a_RandomIt ItCur = a_ItBegin + i;
|
---|
94 | size_t j = i;
|
---|
95 | do
|
---|
96 | {
|
---|
97 | j -= cGap;
|
---|
98 | a_RandomIt ItAtGap = a_ItBegin + j;
|
---|
99 | if (*ItAtGap < *ItCur)
|
---|
100 | break;
|
---|
101 | std::iter_swap(ItAtGap, ItCur);
|
---|
102 | ItCur = ItAtGap;
|
---|
103 | } while (j >= cGap);
|
---|
104 | }
|
---|
105 |
|
---|
106 | /* This does not generate the most optimal gap sequence, but it has the
|
---|
107 | advantage of being simple and avoid floating point. */
|
---|
108 | cGap /= 2;
|
---|
109 | }
|
---|
110 | }
|
---|
111 | }
|
---|
112 |
|
---|
113 | template<typename a_RandomIt, typename a_FnCompareType>
|
---|
114 | void sort(a_RandomIt a_ItBegin, a_RandomIt a_ItEnd, a_FnCompareType a_fnComp)
|
---|
115 | {
|
---|
116 | /* Note! Using shell sort here because it's tiny and we've got code for it. */
|
---|
117 | /** @todo replace with faster code. */
|
---|
118 |
|
---|
119 | /* Anything worth sorting? */
|
---|
120 | std::size_t const cElements = a_ItEnd - a_ItBegin;
|
---|
121 | if (cElements >= 1)
|
---|
122 | {
|
---|
123 | /* Loop on decreasing gap, ending with 1: */
|
---|
124 | std::size_t cGap = (cElements + 1) / 2;
|
---|
125 | while (cGap > 0)
|
---|
126 | {
|
---|
127 | /* Iterate from cGap till the end: */
|
---|
128 | for (std::size_t i = cGap; i < cElements; i++)
|
---|
129 | {
|
---|
130 | /* Find the best suitable location for the item at 'i' comparing
|
---|
131 | backwards in steps of 'cGap', swapping the item at 'i' with the
|
---|
132 | one at '-cGap*j' if it's smaller, stopping if it's larger.
|
---|
133 |
|
---|
134 | Note! Original algorithm would make a copy of the item, this version
|
---|
135 | avoids extra copies of sorted items at the cost of extra copies
|
---|
136 | when dealing with unsorted ones a small cGaps values. */
|
---|
137 | a_RandomIt ItCur = a_ItBegin + i;
|
---|
138 | size_t j = i;
|
---|
139 | do
|
---|
140 | {
|
---|
141 | j -= cGap;
|
---|
142 | a_RandomIt ItAtGap = a_ItBegin + j;
|
---|
143 | if (a_fnComp(*ItAtGap, *ItCur))
|
---|
144 | break;
|
---|
145 | std::iter_swap(ItAtGap, ItCur);
|
---|
146 | ItCur = ItAtGap;
|
---|
147 | } while (j >= cGap);
|
---|
148 | }
|
---|
149 |
|
---|
150 | /* This does not generate the most optimal gap sequence, but it has the
|
---|
151 | advantage of being simple and avoid floating point. */
|
---|
152 | cGap /= 2;
|
---|
153 | }
|
---|
154 | }
|
---|
155 | }
|
---|
156 |
|
---|
157 | }
|
---|
158 |
|
---|
159 | #ifdef _MSC_VER
|
---|
160 | # pragma warning(pop)
|
---|
161 | #endif
|
---|
162 |
|
---|
163 | #endif /* !VBOX_INCLUDED_SRC_nocrt_algorithm */
|
---|
164 |
|
---|