ref: ed218d25d6e01167f8c3485d44b9d33e98a129c8
dir: /msvc10/Readme.txt/
This directory includes hand-crafted project files for building SoX using the
Microsoft Visual C++ 10.0 compilers (available through Visual Studio 2010 or
by downloading the freely-available Microsoft Windows SDK 7.1). The hand-made
project files may be obsoleted by additional CMAKE support in the future, but
for now, this is the easiest way to build SoX with MS Visual C++.
The resulting sox.exe has support for all SoX features except magic, ffmpeg,
and pulseaudio. LAME (libmp3lame.dll or lame_enc.dll), MAD (libmad.dll or
cygmad-0.dll), libsndfile (libsndfile-1.dll), and AMR support (libamrnb-3.dll,
libamrwb-3.dll) are loaded at runtime if they are available. OpenMP support is
available only via Visual Studio 2010 Professional (it is not availalble if you
build SoX via Visual Studio 2010 Express or via the Microsoft Windows SDK).
How to build:
1. If you don't already have it, install .NET 4.0 (required for msbuild).
If you don't already have Visual Studio 2010 or the Windows SDK 7.1
installed, download and install the Windows SDK 7.1 from Microsoft:
http://www.microsoft.com/downloads/en/details.aspx?FamilyID=6b6c21d2-2006-4afa-9702-529fa782d63b&displaylang=en
Be sure to include at least the following features:
Windows Headers and Libraries - Windows Headers
Windows Headers and Libraries - x86 Libraries
Windows Native Code Development - Tools
Windows Native Code Development - Visual C++ Compilers
2. Check out the SoX git code into a directory named sox.
Extract the source code for the other libraries next to the sox
directory. Remove the version numbers from the directory names.
The following versions were tested and successfully built:
-- flac-1.2.1.tar.gz extracted into directory flac
-- lame-398.4.tar.gz extracted into directory lame
-- libid3tag-0.15.1b.tar.gz extracted into directory libid3tag
-- libmad-0.15.1b.tar.gz extracted into directory libmad
-- libogg-1.2.2.tar.gz extracted into directory libogg
-- libpng-1.5.1.tar.gz extracted into directory libpng
-- libsndfile-1.0.23.tar.gz extracted into directory libsndfile
-- libvorbis-1.3.2.tar.gz extracted into directory libvorbis
-- speex-1.2rc1.tar.gz extracted into directory speex
-- wavpack-4.60.1.tar.bz2 extracted into directory wavpack
-- zlib-1.2.5.tar.gz extracted into directory zlib
3. If using Visual Studio, open the sox\msvc10\SoX.sln solution in VS2010.
If using the Windows SDK, open a normal command prompt, then run:
"c:\Program Files\Microsoft SDKs\Windows\v7.1\bin\SetEnv.cmd" /x86 /Release /xp
then CD to the sox\msvc10 folder.
4. If any of the above libraries are not available or not wanted, adjust the
corresponding HAVE_* settings in the soxconfig.h file and remove the
corresponding project from the SoX.sln solution.
If using Visual Studio, you will find the soxconfig.h file in the LibSox
project's Config Files folder.
If using the Windows SDK, you'll have to use a text editor (i.e. notepad) to
edit the soxconfig.h file (sox\msvc10\sox\soxconfig.h), and you'll have to
manually remove the entries for the unwanted projects.
5. If using Visual Studio Professional or above and you want OpenMP support,
enable it in the project settings for the LibSox and SoX projects
(Configuration Properties, C/C++, Language, Open MP Support, set to Yes).
6. If using Visual Studio, build the solution using the GUI.
If using the Windows SDK, run: msbuild SoX.sln
7. The resulting executable files will be in sox\msvc10\Debug or
sox\msvc10\Release. The resulting sox.exe will dynamically link to
libmp3lame.dll, libmad.dll, libsndfile-1.dll, libamrnb-3.dll, and
libamrwb-3.dll if they are available, but will run without them (though the
corresponding features will be unavailable if they are not present).
Points to note:
- The libsndfile-1.0.20.tar.gz package does not include the sndfile.h header
file. Normally, before compiling libsndfile, you would create sndfile.h
(either by processing it via autoconf, by downloading a copy, or by renaming
sndfile.h.in). However, this SoX solution includes its own version of
sndfile.h, so you should not create a sndfile.h under the libsndfile folder.
To repeat: you should extract a clean copy of libsndfile-1.0.20.tar.gz, and
should not add, process, or rename any files.
- The solution includes an experimental effect called speexdsp that uses the
speex DSP library. This does not yet enable any support for the speex file
format or speex codec. The speexdsp effect is simply an experimental effect
to make use of the automatic gain control and noise filtering components that
are part of the speex codec package. Support for the speex codec may be added
later.
- If you enable Open MP support, you will need vcomp100.dll either installed on
your machine or copied into the directory next to sox.exe. If you have Open
MP support in your copy of Visual Studio, this file can be found here:
c:\Program Files\Microsoft Visual Studio 10.0\
vc\redist\x86\Microsoft.VC100.OPENMP
- The included projects do not enable SSE2. You can enable this in the project
properties under Configuration Properties, C/C++, Code Generation, Enable
Enhanced Instruction Set. Note that some editions of Visual Studio might
not include Enhanced Instruction Set support.
- The included projects set the floating-point model to "fast". This means
that the compiler is free to optimize floating-point operations. For
example, the compiler might optimize the expression (14.0 * x / 7.0) into
(x * 2.0). In addition, the compiler is allowed to leave expression results
in floating-point registers to store temporary values instead of rounding
each intermediate result to a 32-bit or 64-bit value. In some cases, these
optimizations can change the results of floating-point calculations. If you
need more precise results, you can change this optimization setting can be
changed to one of the other values. The "precise" setting avoids any
optimization that might change the result (preserves the order of all
operations) but keeps optimizations that might give more accurate results
(such as using more precision than necessary for intermediate values if doing
so results in faster code). The "strict" setting avoids any optimization that
might change the result in any way contrary to the C/C++ standard and rounds
every intermediate result to the requested precision according to standard
floating-point rounding rules. You can change this setting in the project
properties under Configuration Properties, C/C++, Code Generation, Floating
Point Model.