1/*
2 * Copyright (C) 2017 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#ifndef LIBTEXTCLASSIFIER_UTIL_BASE_CASTS_H_
18#define LIBTEXTCLASSIFIER_UTIL_BASE_CASTS_H_
19
20#include <string.h>  // for memcpy
21
22namespace libtextclassifier {
23
24// lang_id_bit_cast<Dest,Source> is a template function that implements the
25// equivalent of "*reinterpret_cast<Dest*>(&source)".  We need this in
26// very low-level functions like the protobuf library and fast math
27// support.
28//
29//   float f = 3.14159265358979;
30//   int i = lang_id_bit_cast<int32>(f);
31//   // i = 0x40490fdb
32//
33// The classical address-casting method is:
34//
35//   // WRONG
36//   float f = 3.14159265358979;            // WRONG
37//   int i = * reinterpret_cast<int*>(&f);  // WRONG
38//
39// The address-casting method actually produces undefined behavior
40// according to ISO C++ specification section 3.10 -15 -.  Roughly, this
41// section says: if an object in memory has one type, and a program
42// accesses it with a different type, then the result is undefined
43// behavior for most values of "different type".
44//
45// This is true for any cast syntax, either *(int*)&f or
46// *reinterpret_cast<int*>(&f).  And it is particularly true for
47// conversions between integral lvalues and floating-point lvalues.
48//
49// The purpose of 3.10 -15- is to allow optimizing compilers to assume
50// that expressions with different types refer to different memory.  gcc
51// 4.0.1 has an optimizer that takes advantage of this.  So a
52// non-conforming program quietly produces wildly incorrect output.
53//
54// The problem is not the use of reinterpret_cast.  The problem is type
55// punning: holding an object in memory of one type and reading its bits
56// back using a different type.
57//
58// The C++ standard is more subtle and complex than this, but that
59// is the basic idea.
60//
61// Anyways ...
62//
63// lang_id_bit_cast<> calls memcpy() which is blessed by the standard,
64// especially by the example in section 3.9 .  Also, of course,
65// lang_id_bit_cast<> wraps up the nasty logic in one place.
66//
67// Fortunately memcpy() is very fast.  In optimized mode, with a
68// constant size, gcc 2.95.3, gcc 4.0.1, and msvc 7.1 produce inline
69// code with the minimal amount of data movement.  On a 32-bit system,
70// memcpy(d,s,4) compiles to one load and one store, and memcpy(d,s,8)
71// compiles to two loads and two stores.
72//
73// I tested this code with gcc 2.95.3, gcc 4.0.1, icc 8.1, and msvc 7.1.
74//
75// WARNING: if Dest or Source is a non-POD type, the result of the memcpy
76// is likely to surprise you.
77//
78// Props to Bill Gibbons for the compile time assertion technique and
79// Art Komninos and Igor Tandetnik for the msvc experiments.
80//
81// -- mec 2005-10-17
82
83template <class Dest, class Source>
84inline Dest bit_cast(const Source &source) {
85  static_assert(sizeof(Dest) == sizeof(Source), "Sizes do not match");
86
87  Dest dest;
88  memcpy(&dest, &source, sizeof(dest));
89  return dest;
90}
91
92}  // namespace libtextclassifier
93
94#endif  // LIBTEXTCLASSIFIER_UTIL_BASE_CASTS_H_
95