third_party/boost/endian/doc/endian/overview.adoc

////
Copyright 2011-2016 Beman Dawes

Distributed under the Boost Software License, Version 1.0.
(http://www.boost.org/LICENSE_1_0.txt)
////

[#overview]
# Overview
:idprefix: overview_

## Abstract

Boost.Endian provides facilities to manipulate the
<<overview_endianness,endianness>> of integers and user-defined types.

* Three approaches to endianness are supported. Each has a long history of
successful use, and each approach has use cases where it is preferred over the
other approaches.
* Primary uses:
** Data portability. The Endian library supports binary data exchange, via
either external media or network transmission, regardless of platform
endianness.
** Program portability. POSIX-based and Windows-based operating systems
traditionally supply libraries with non-portable functions to perform endian
conversion. There are at least four incompatible sets of functions in common
use. The Endian library is portable across all {cpp} platforms.
* Secondary use: Minimizing data size via sizes and/or alignments not supported
by the standard {cpp} integer types.

[#overview_endianness]
## Introduction to endianness

Consider the following code:

```
int16_t i = 0x0102;
FILE * file = fopen("test.bin", "wb"); // binary file!
fwrite(&i, sizeof(int16_t), 1, file);
fclose(file);
```

On OS X, Linux, or Windows systems with an Intel CPU, a hex dump of the
"test.bin" output file produces:

```
0201
```

On OS X systems with a PowerPC CPU, or Solaris systems with a SPARC CPU, a hex
dump of the "test.bin" output file produces:

```
0102
```

What's happening here is that Intel CPUs order the bytes of an integer with the
least-significant byte first, while SPARC CPUs place the most-significant byte
first. Some CPUs, such as the PowerPC, allow the operating system to choose
which ordering applies.

Most-significant-byte-first ordering is traditionally called "big endian"
ordering and least-significant-byte-first is traditionally called
"little-endian" ordering. The names are derived from
http://en.wikipedia.org/wiki/Jonathan_Swift[Jonathan Swift]'s satirical novel
_http://en.wikipedia.org/wiki/Gulliver's_Travels[Gulliver's Travels]_, where
rival kingdoms opened their soft-boiled eggs at different ends.

See Wikipedia's http://en.wikipedia.org/wiki/Endianness[Endianness] article for
an extensive discussion of endianness.

Programmers can usually ignore endianness, except when reading a core  dump on
little-endian systems. But programmers  have to deal with endianness when
exchanging binary integers and binary floating point values between computer
systems with differing endianness, whether by physical file transfer or over a
network. And programmers may also want to use the library when minimizing either
internal or external data sizes is advantageous.

[#overview_introduction]
## Introduction to the Boost.Endian library

Boost.Endian provides three different approaches to dealing with endianness. All
three approaches support integers and user-define types (UDTs).

Each approach has a long history of successful use, and each approach has use
cases where it is preferred to the other approaches. See
<<choosing,Choosing between Conversion Functions, Buffer Types, and Arithmetic Types>>.

<<conversion,Endian conversion functions>>::
The application uses the built-in integer types to hold values, and calls the
provided conversion functions to convert byte ordering as needed. Both mutating
and non-mutating conversions are supplied, and each comes in unconditional and
conditional variants.

<<buffers, Endian buffer types>>::
The application uses the provided endian buffer types to hold values, and
explicitly converts to and from the built-in integer types. Buffer sizes of 8,
16, 24, 32, 40, 48, 56, and 64 bits (i.e. 1, 2, 3, 4, 5, 6, 7, and 8 bytes) are
provided. Unaligned integer buffer types are provided for all sizes, and aligned
buffer types are provided for 16, 32, and 64-bit sizes. The provided specific
types are typedefs for a generic class template that may be used directly for
less common use cases.

<<arithmetic, Endian arithmetic types>>::
The application uses the provided endian arithmetic types, which supply the same
operations as the built-in {cpp} arithmetic types. All conversions are implicit.
Arithmetic sizes of 8, 16, 24, 32, 40, 48, 56, and 64 bits (i.e. 1, 2, 3, 4, 5,
6, 7, and 8 bytes) are provided. Unaligned integer types are provided for all
sizes and aligned arithmetic types are provided for 16, 32, and 64-bit sizes.
The provided specific types are typedefs for a generic class template that may
be used directly in generic code of for less common use cases.

Boost Endian is a header-only library. {cpp}11 features affecting interfaces,
such as `noexcept`, are  used only if available. See
<<overview_cpp03_support,{cpp}03 support for {cpp}11 features>> for details.

[#overview_intrinsics]
## Built-in support for Intrinsics

Most compilers, including GCC, Clang, and Visual {cpp}, supply  built-in support
for byte swapping intrinsics. The Endian library uses these intrinsics when
available since they may result in smaller and faster generated code,
particularly for optimized builds.

Defining the macro `BOOST_ENDIAN_NO_INTRINSICS` will suppress use of the
intrinsics. This is useful when a compiler has no intrinsic support or fails to
locate the appropriate header, perhaps because it is an older release or has
very limited supporting libraries.

The macro `BOOST_ENDIAN_INTRINSIC_MSG` is defined as either
`"no byte swap intrinsics"` or a string describing the particular set of
intrinsics being used. This is useful for eliminating missing intrinsics as a
source of performance issues.

## Performance

Consider this problem:

### Example 1
Add 100 to a big endian value in a file, then write the result to a file
[%header,cols=2*]
|===
|Endian arithmetic type approach |Endian conversion function approach
a|
----
big_int32_at x;

... read into x from a file ...


x += 100;


... write x to a file ...
----
a|
----
int32_t x;

... read into x from a file ...

big_to_native_inplace(x);
x += 100;
native_to_big_inplace(x);

... write x to a file ...
----
|===

*There will be no performance difference between the two approaches in optimized
builds, regardless of the native endianness of the machine.* That's because
optimizing compilers will generate exactly the same code for each. That
conclusion was confirmed by studying the generated assembly code for GCC and
Visual {cpp}. Furthermore, time spent doing I/O will determine the speed of this
application.

Now consider a slightly different problem:

### Example 2
Add a million values to a big endian value in a file, then write the result to a
file
[%header,cols=2*]
|===
|Endian arithmetic type approach |Endian conversion function approach
a|
----
big_int32_at x;

... read into x from a file ...



for (int32_t i = 0; i < 1000000; ++i)
  x += i;



... write x to a file ...
----
a|
----
int32_t x;

... read into x from a file ...

big_to_native_inplace(x);

for (int32_t i = 0; i < 1000000; ++i)
  x += i;

native_to_big_inplace(x);

... write x to a file ...
----
|===

With the Endian arithmetic approach, on little endian platforms an implicit
conversion from and then back to big endian is done inside the loop. With the
Endian conversion function approach, the user has ensured the conversions are
done outside the loop, so the code may run more quickly on little endian
platforms.

### Timings

These tests were run against release builds on a circa 2012 4-core little endian
X64 Intel Core i5-3570K CPU @ 3.40GHz under Windows 7.

CAUTION: The Windows CPU timer has very high granularity. Repeated runs of the
same tests often yield considerably different results.

See `test/loop_time_test.cpp` for the actual code and `benchmark/Jamfile.v2` for
the build setup.

#### GNU C++ version 4.8.2 on Linux virtual machine
Iterations: 10'000'000'000, Intrinsics: `__builtin_bswap16`, etc.
[%header,cols=3*]
|===
|Test Case |Endian arithmetic type |Endian conversion function
|16-bit aligned big endian |8.46 s |5.28 s
|16-bit aligned little endian |5.28 s |5.22 s
|32-bit aligned big endian |8.40 s |2.11 s
|32-bit aligned little endian |2.11 s |2.10 s
|64-bit aligned big endian |14.02 s |3.10 s
|64-bit aligned little endian |3.00 s |3.03 s
|===

#### Microsoft Visual C++ version 14.0
Iterations: 10'000'000'000, Intrinsics: `<cstdlib>` `_byteswap_ushort`, etc.
[%header,cols=3*]
|===
|Test Case |Endian arithmetic type |Endian conversion function
|16-bit aligned big endian |8.27 s |5.26 s
|16-bit aligned little endian |5.29 s |5.32 s
|32-bit aligned big endian |8.36 s |5.24 s
|32-bit aligned little endian |5.24 s |5.24 s
|64-bit aligned big endian |13.65 s |3.34 s
|64-bit aligned little endian |3.35 s |2.73 s
|===

[#overview_cpp03_support]
## {cpp}03 support for {cpp}11 features

[%header,cols=2*]
|===
|{cpp}11 Feature
|Action with {cpp}03 Compilers
|Scoped enums
|Uses header
http://www.boost.org/libs/core/doc/html/core/scoped_enum.html[boost/core/scoped_enum.hpp]
to emulate {cpp}11 scoped enums.
|`noexcept`
|Uses `BOOST_NOEXCEPT` macro, which is defined as null for compilers not
supporting this {cpp}11 feature.
|{cpp}11 PODs
(http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2342.htm[N2342])
|Takes advantage of {cpp}03 compilers that relax {cpp}03 POD rules, but see
Limitations <<buffers_limitations,here>> and <<arithmetic_limitations,here>>.
Also see macros for explicit POD control <<buffers_compilation,here>> and
<<arithmetic_compilation,here>>
|===

[#overview_faq]
## Overall FAQ

Is the implementation header only?::
Yes.

Are {cpp}03 compilers supported?::
Yes.

Does the implementation use compiler intrinsic built-in byte swapping?::
Yes, if available. See <<overview_intrinsics,Intrinsic built-in support>>.

Why bother with endianness?::
Binary data portability is the primary use case.

Does endianness have any uses outside of portable binary file or network I/O formats?::
Using the unaligned integer types with a size tailored to the application's
needs is a minor secondary use that saves internal or external memory space. For
example, using `big_int40_buf_t` or `big_int40_t` in a large array saves a lot
of space compared to one of the 64-bit types.

Why bother with binary I/O? Why not just use {cpp} Standard Library stream inserters and extractors?::
* Data interchange formats often specify binary integer data. Binary integer
data is smaller and therefore I/O is faster and file sizes are smaller. Transfer
between systems is less expensive.
* Furthermore, binary integer data is of fixed size, and so fixed-size disk
records are possible without padding, easing sorting and allowing random access.
* Disadvantages, such as the inability to use text utilities on the resulting
files, limit usefulness to applications where the binary I/O advantages are
paramount.

Which is better, big-endian or little-endian?::
Big-endian tends to be preferred in a networking environment and is a bit more
of an industry standard, but little-endian may be preferred for applications
that run primarily on x86, x86-64, and other little-endian CPU's. The
http://en.wikipedia.org/wiki/Endian[Wikipedia] article gives more pros and cons.

Why are only big and little native endianness supported?::
These are the only endian schemes that have any practical value today. PDP-11
and the other middle endian approaches are interesting  curiosities but have no
relevance for today's {cpp} developers. The same is true for architectures that
allow runtime endianness switching. The
<<conversion_native_order_specification,specification for native ordering>> has
been carefully crafted to allow support for such orderings in the future, should
the need arise. Thanks to Howard Hinnant for suggesting this.

Why do both the buffer and arithmetic types exist?::
Conversions in the buffer types are explicit. Conversions in the arithmetic
types are implicit. This fundamental difference is a deliberate design feature
that would be lost if the inheritance hierarchy were collapsed.
The original design provided only arithmetic types. Buffer types were requested
during formal review by those wishing total control over when conversion occurs.
They also felt that buffer types would be less likely to be misused by
maintenance programmers not familiar with the implications of performing a lot
of integer operations on the endian arithmetic integer types.

What is gained by using the buffer types rather than always just using the arithmetic types?::
Assurance that hidden conversions are not performed. This is of overriding
importance to users concerned about achieving the ultimate in terms of speed.
"Always just using the arithmetic types" is fine for other users. When the
ultimate in speed needs to be ensured, the arithmetic types can be used in the
same design patterns or idioms that would be used for buffer types, resulting in
the same code being generated for either types.

What are the limitations of integer support?::
Tests have only been performed on machines that  use two's complement
arithmetic. The Endian conversion functions only support 8, 16, 32, and 64-bit
aligned integers. The endian types only support 8, 16, 24, 32, 40, 48, 56, and
64-bit unaligned integers, and 8, 16, 32, and 64-bit aligned integers.

Is there floating point support?::
An attempt was made to support four-byte ``float``s and eight-byte
``double``s, limited to
http://en.wikipedia.org/wiki/IEEE_floating_point[IEEE 754] (also known as
ISO/IEC/IEEE 60559) floating point and further limited to systems where floating
point endianness does not differ from integer endianness. Even with those
limitations, support for floating point types was not reliable and was removed.
For example, simply reversing the endianness of a floating point number can
result in a signaling-NAN.
+
Support for `float` and `double` has since been reinstated for `endian_buffer`,
`endian_arithmetic` and the conversion functions that reverse endianness in place.
The conversion functions that take and return by value still do not support floating
point due to the above issues; reversing the bytes of a floating point number
does not necessarily produce another valid floating point number.