CMake-based LLVM MinGW-W64 Cross Toolchain

This thing’s primary use is to build Windows binaries of apps.

Prerequisites

• You should also install Ninja and use CMake’s Ninja build file generator.

• As a build environment, any modern glibc Linux distribution should work.

Setup Build Environment

Manjaro / Arch Linux

These packages need to be installed first before compiling mpv:

pacman -S git gyp mercurial ninja cmake meson ragel yasm nasm asciidoc enca gperf unzip p7zip clang lld libc++ libc++abi python-pip curl mimalloc ccache go

pip3 install mako jsonschema

Ubuntu Linux / WSL (Windows 10)

apt-get install build-essential checkinstall bison flex gettext git mercurial ninja-build gyp cmake yasm nasm automake pkgconf libtool libtool-bin clang lld libc++1 libc++abi1 libgmp-dev libmpfr-dev libmpc-dev libgcrypt-dev gperf ragel texinfo autopoint re2c asciidoc python3-pip docbook2x unzip p7zip-full curl mimalloc ccache golang

pip3 install meson mako jsonschema

Note:

• It is advised to use bash over dash. Set sudo ln -sf /bin/bash /bin/sh. Revert back by sudo ln -sf /bin/dash /bin/sh.
• To update package installed by pip, run pip3 install <package> --upgrade.
• Distributions clang is usually problematic, it is highly recommended to install Fuchsia Clang as an alternative:

go install go.chromium.org/luci/cipd/client/cmd/...@latest
cipd install fuchsia/third_party/clang/linux-amd64 latest -root /usr/local/fuchsia-clang
PATH="/usr/local/fuchsia-clang/bin:$PATH"

Compiling with Clang

Supported target architecture (TARGET_ARCH) with clang is: x86_64-w64-mingw32, aarch64-w64-mingw32.

Example:

cmake -DTARGET_ARCH=x86_64-w64-mingw32 \
-DCMAKE_INSTALL_PREFIX="$PWD/clang_root" \
-DMARCH=x86-64-v3 \
-DSINGLE_SOURCE_LOCATION="$PWD/src_packages" \
-G Ninja -B build_x86_64_v3 -S mpv-winbuild-cmake

The cmake command will create clang_root as clang sysroot where llvm tools installed. build_x86_64 is build directory to compiling packages.

cd build_x86_64
ninja toolchains-download # download llvm
ninja llvm       # build LLVM (take around ~2 hours)
ninja llvm-clang # build clang on specified target
ninja download   # download packages
ninja fullclean && ninja download # workaround to avoid problems
ninja mpv        # build mpv and all its dependencies

If you want add another target (ex. aarch64-w64-mingw32), change TARGET_ARCH and build folder.

cmake -DTARGET_ARCH=aarch64-w64-mingw32 \
-DCMAKE_INSTALL_PREFIX="$PWD/clang_root" \
-DSINGLE_SOURCE_LOCATION="$PWD/src_packages" \
-G Ninja -B build_aarch64 -S mpv-winbuild-cmake
cd build_aarch64
ninja llvm-clang # same as above

If you've changed MARCH option, you need to run:

ninja rebuild_cache

to update flags which will pass on gcc, g++ and etc.

Building Software (Second Time)

To build mpv for a second time:

ninja update # perform git pull on all packages that used git

After that, build mpv as usual:

ninja mpv

CMake Options

Options	Description	Default
TARGET_ARCH	the target to build. available: x86_64-w64-mingw32, aarch64-w64-mingw32	x86_64-w64-mingw32
MARCH	for a specific microarchitecture. For x86_64, this means -march, for aarch64, this means -mcpu	x86_64:x86-64-v3 aarch64:cortex-x1
CMAKE_INSTALL_PREFIX	location of the host toolchain installation	${CMAKE_CURRENT_BINARY_DIR}/install
SINGLE_SOURCE_LOCATION	location of the packages source	${CMAKE_CURRENT_BINARY_DIR}/src_packages
MINGW_INSTALL_PREFIX	location of the MinGW sysroot	${CMAKE_INSTALL_PREFIX}/${TARGET_ARCH}
ENABLE_CCACHE	ccache integration	OFF
CCACHE_MAXSIZE	ccache size	500M
MALLOC	malloc for LD_PRELOAD	/usr/lib/libmimalloc.so /usr/lib/libjemalloc.so
CLANG_PACKAGES_PGO	IRPGO for packages	OFF
	GEN: build packages with profile instrumentation
	USE: build packages with profdata
CLANG_PACKAGES_PROFDATA_FILE	path to your profdata
CLANG_FLAGS	FLAGS passed to clang
LLD_FLAGS	FLAGS passed to lld
TOOLCHAIN_FLAGS	CFLAGS passed to host compiler when build LLVM
LLVM_ENABLE_LTO	Build LLVM with LTO	OFF
	Thin: use ThinLTO
	Full: use FullLTO
LLVM_ENABLE_PGO	Build LLVM with PGO	OFF
	GEN: build LLVM with IR profile instrumentation
	CSGEN: build LLVM with CSIR profile instrumentation and profdata
	USE: build LLVM with profdata
LLVM_PROFDATA_FILE	path to your LLVM profdata
CUSTOM_LIBCXX	use bundled libcxx instead of default cxx stdlib to build LLVM, Don't enable it unless you know what it does	OFF
CUSTOM_COMPILER_RT	use bundled compiler-rt builtins instead of default compiler-rt to build LLVM, Don't enable it unless you know what it does	OFF

Available Commands

Commands	Description
ninja package	compile a package
ninja clean	remove all stamp files in all packages.
ninja fullclean	remove all stamp and build files in all packages
ninja download	Download all packages' sources at once without compiling.
ninja patch	patch all packages
ninja update	Update all git repos. When a package pulls new changes, all of its stamp files will be deleted and will be forced rebuild. If there is no change, it will not remove the stamp files and no rebuild occur. Use this instead of ninja clean if you don't want to rebuild everything in the next run.
ninja toolchains-fullclean	remove all stamp and build files in all toolchain packages
ninja toolchains-download	Download all toolchain packages' sources at once without compiling.
ninja toolchains-update	Update all toolchain git repos. When a package pulls new changes, all of its stamp files will be deleted and will be forced rebuild. If there is no change, it will not remove the stamp files and no rebuild occur. Use this instead of ninja clean if you don't want to rebuild everything in the next run.
ninja package-fullclean	Remove all stamp files of a package.
ninja package-removebuild	Remove 'build' directory of a package.
ninja package-removeprefix	Remove 'prefix' directory.
ninja package-force-update	Update a package. Only git repo will be updated.
ninja ccache-recomp	Recompress ccache cache.

package is package's name found in packages folder.

Information about packages

• Apps

  • mpv
  • FFmpeg
  • qBittorrent-Enhanced-Edition
  • curl
  • svtav1-psy
  • mediainfo

• Git deps

  • amf-headers
  • ANGLE
  • aom
  • avisynth-headers
  • brotli
  • bzip2
  • c-ares
  • curl
  • dav1d
  • fast_float
  • FFmpeg
  • fontconfig
  • freetype2
  • fribidi
  • game-music-emu
  • glad
  • glslang
  • graphengine
  • harfbuzz
  • highway
  • lcms2
  • libarchive
  • libaribcaption
  • libass
  • libbluray
  • libdovi
  • libdvdcss
  • libdvdnav
  • libdvdread
  • libhdr10plus
  • libjpeg
  • libjxl
  • liblc3
  • libmediainfo
  • libmediainfo
  • libmodplug
  • libmysofa
  • libplacebo
  • libpng
  • libpsl
  • libsamplerate
  • libsdl2
  • libsixel
  • libsoxr
  • libsrt
  • libssh
  • libtorrent
  • libudfread
  • libunibreak
  • libva
  • libvidstab
  • libvpl
  • libvpx
  • libwebp
  • libxml2
  • libzimg
  • luajit
  • mimalloc
  • mpv
  • mpv-debug-plugin
  • mpv-menu-plugin
  • mujs
  • nghttp2
  • nghttp3
  • ngtcp2
  • nvcodec-headers
  • ogg
  • openal-soft
  • opencl
  • opencl-header
  • openssl
  • opus
  • qt-base
  • qt-svg
  • qt-tools
  • rubberband
  • shaderc
  • sleef
  • speex
  • spirv-cross
  • spirv-headers
  • spirv-tools
  • svtav1
  • uchardet
  • vorbis
  • vulkan
  • vulkan-header
  • vvenc
  • x265
  • xxhash
  • xz
  • zenlib
  • zenlib
  • zlib-ng
  • zstd

• Zip deps

  • libiconv (1.18)
  • lzo (2.10)
  • vapoursynth (R70)
  • boost (1.87.0)

WSL workaround

Place the file on specified location to limit ram & cpu usage to avoid getting stuck while building mpv.

Recommended 3GB per CPU

# /etc/wsl.conf
[interop]
enabled=false
appendWindowsPath=false
---------------------------------------
# C:\Users\<UserName>\.wslconfig
[wsl2]
processors=8
memory=24GB
swap=0
[experimental]
autoMemoryReclaim=dropcache
sparseVhd=true