Instructions for building this repository on Linux, Windows, and MacOS.
- Build Instructions
- Table Of Contents
- Contributing to the Repository
- Repository Content
- Repository Set-Up
- Building On Windows
- Building On Linux
- Building on MacOS
- Building on Fuchsia
- Building on QNX
- Cross Compilation
- Tests
If you intend to contribute, the preferred work flow is for you to develop your contribution in a fork of this repository in your GitHub account and then submit a pull request. Please see the CONTRIBUTING.md file in this repository for more details.
This repository contains the source code necessary to build the desktop Vulkan loader and its tests.
The install
target installs the following files under the directory
indicated by install_dir:
- install_dir
/lib
: The Vulkan loader library - install_dir
/bin
: The Vulkan loader library DLL (Windows)
This repository does not contain a Vulkan-capable driver. You will need to obtain and install a Vulkan driver from your graphics hardware vendor or from some other suitable source if you intend to run Vulkan applications.
To create your local git repository:
git clone https://github.com/KhronosGroup/Vulkan-Loader.git
This repository attempts to resolve some of its dependencies by using components found from the following places, in this order:
- CMake or Environment variable overrides (e.g., -DVULKAN_HEADERS_INSTALL_DIR)
- System-installed packages, mostly applicable on Linux
Dependencies that cannot be resolved by the SDK or installed packages must be resolved with the "install directory" override and are listed below. The "install directory" override can also be used to force the use of a specific version of that dependency.
This repository has a required dependency on the Vulkan Headers repository.
You must clone the headers repository and build its install
target before
building this repository. The Vulkan-Headers repository is required because it
contains the Vulkan API definition files (registry) that are required to build
the loader. You must also take note of the headers install directory and pass
it on the CMake command line for building this repository, as described below.
The loader tests depend on the Google Test library and on Windows platforms depends on the Microsoft Detours library.
To build the tests, pass the -DUPDATE_DEPS=ON
and -DBUILD_TESTS=ON
options when generating the project:
cmake ... -DUPDATE_DEPS=ON -DBUILD_TESTS=ON ...
This will ensure googletest and detours is downloaded and the appropriate version is used.
A common convention is to place the build
directory in the top directory of
the repository and place the install
directory as a child of the build
directory. The remainder of these instructions follow this convention,
although you can place these directories in any location.
There is a Python utility script, scripts/update_deps.py
, that you can use
to gather and build the dependent repositories mentioned above.
This program also uses information stored in the scripts/known-good.json
file
to checkout dependent repository revisions that are known to be compatible with
the revision of this repository that you currently have checked out.
You can choose to do this manually or automatically. The first step to either is cloning the Vulkan-Loader repo and stepping into that newly cloned folder:
git clone [email protected]:KhronosGroup/Vulkan-Loader.git
cd Vulkan-Loader
To manually update the dependencies you now must create the build folder, and run the update deps script followed by the necessary CMake build commands:
mkdir build
cd build
../scripts/update_deps.py
cmake -C helper.cmake ..
cmake --build .
- You may need to adjust some of the CMake options based on your platform. See the platform-specific sections later in this document.
- The
update_deps.py
script fetches and builds the dependent repositories in the current directory when it is invoked. In this case, they are built in thebuild
directory. - The
build
directory is also being used to build this (Vulkan-ValidationLayers) repository. But there shouldn't be any conflicts inside thebuild
directory between the dependent repositories and the build files for this repository. - The
--dir
option forupdate_deps.py
can be used to relocate the dependent repositories to another arbitrary directory using an absolute or relative path. - The
update_deps.py
script generates a file namedhelper.cmake
and places it in the same directory as the dependent repositories (build
in this case). This file contains CMake commands to set the CMake*_INSTALL_DIR
variables that are used to point to the install artifacts of the dependent repositories. You can use this file with thecmake -C
option to set these variables when you generate your build files with CMake. This lets you avoid entering several*_INSTALL_DIR
variable settings on the CMake command line. - If using "MINGW" (Git For Windows), you may wish to run
winpty update_deps.py
in order to avoid buffering all of the script's "print" output until the end and to retain the ability to interrupt script execution. - Please use
update_deps.py --help
to list additional options and read the internal documentation inupdate_deps.py
for further information.
On the other hand, if you choose to let the CMake scripts do all the heavy-lifting, you may just trigger the following CMake commands:
cmake -S. -Bbuild -DUPDATE_DEPS=On
cmake --build build
- You may need to adjust some of the CMake options based on your platform. See the platform-specific sections later in this document.
- The
build
directory is also being used to build this (Vulkan-ValidationLayers) repository. But there shouldn't be any conflicts inside thebuild
directory between the dependent repositories and the build files for this repository.
This repository contains generated source code in the loader/generated
directory which is not intended to be modified directly. Instead, changes should be
made to the corresponding generator in the scripts
directory. The source files can
then be regenerated using scripts/generate_source.py
.
Run python scripts/generate_source.py --help
to see how to invoke it.
A helper CMake target loader_codegen
is also provided to simplify the invocation of scripts/generate_source.py
.
# Note by default this helper target is disabled
cmake -S . -B build -D LOADER_CODEGEN=ON
cmake --build . --target loader_codegen
When generating native platform build files through CMake, several options can be specified to customize the build. Some of the options are binary on/off options, while others take a string as input. The following is a table of all on/off options currently supported by this repository:
Option | Platform | Default | Description |
---|---|---|---|
BUILD_TESTS | All | OFF |
Controls whether or not the loader tests are built. |
BUILD_WSI_XCB_SUPPORT | Linux | ON |
Build the loader with the XCB entry points enabled. Without this, the XCB headers should not be needed, but the extension VK_KHR_xcb_surface won't be available. |
BUILD_WSI_XLIB_SUPPORT | Linux | ON |
Build the loader with the Xlib entry points enabled. Without this, the X11 headers should not be needed, but the extension VK_KHR_xlib_surface won't be available. |
BUILD_WSI_WAYLAND_SUPPORT | Linux | ON |
Build the loader with the Wayland entry points enabled. Without this, the Wayland headers should not be needed, but the extension VK_KHR_wayland_surface won't be available. |
BUILD_WSI_DIRECTFB_SUPPORT | Linux | OFF |
Build the loader with the DirectFB entry points enabled. Without this, the DirectFB headers should not be needed, but the extension VK_EXT_directfb_surface won't be available. |
BUILD_WSI_SCREEN_QNX_SUPPORT | QNX | OFF |
Build the loader with the QNX Screen entry points enabled. Without this the extension VK_QNX_screen_surface won't be available. |
ENABLE_WIN10_ONECORE | Windows | OFF |
Link the loader to the OneCore umbrella library, instead of the standard Win32 ones. |
USE_GAS | Linux | ON |
Controls whether to build assembly files with the GNU assembler, else fallback to C code. |
USE_MASM | Windows | ON |
Controls whether to build assembly files with MS assembler, else fallback to C code |
BUILD_STATIC_LOADER | macOS | OFF |
This allows the loader to be built as a static library on macOS. Not tested, use at your own risk. |
LOADER_ENABLE_ADDRESS_SANITIZER | Linux & macOS | OFF |
Enables Address Sanitizer in the loader and tests. |
LOADER_ENABLE_THREAD_SANITIZER | Linux & macOS | OFF |
Enables Thread Sanitizer in the loader and tests. |
LOADER_DISABLE_DYNAMIC_LIBRARY_UNLOADING | All | OFF |
Causes the loader to not unload dynamic libraries. Example use case. This option allows leak sanitizers to have full stack traces. |
LOADER_USE_UNSAFE_FILE_SEARCH | All | OFF |
Disables security policies that prevent unsecure locations from being used when running with elevated permissions. |
LOADER_CODEGEN | All | OFF |
Creates a helper CMake target to generate code. |
NOTE: LOADER_USE_UNSAFE_FILE_SEARCH
should NOT be enabled except in very specific contexts (like isolated test environments)!
The following is a table of all string options currently supported by this repository:
Option | Platform | Default | Description |
---|---|---|---|
FALLBACK_CONFIG_DIRS | Linux/MacOS | /etc/xdg |
Configuration path(s) to use instead of XDG_CONFIG_DIRS if that environment variable is unavailable. The default setting is freedesktop compliant. |
FALLBACK_DATA_DIRS | Linux/MacOS | /usr/local/share:/usr/share |
Configuration path(s) to use instead of XDG_DATA_DIRS if that environment variable is unavailable. The default setting is freedesktop compliant. |
BUILD_DLL_VERSIONINFO | Windows | "" (empty string) |
Allows setting the Windows specific version information for the Loader DLL. Format is "major.minor.patch.build". |
These variables should be set using the -D
option when invoking CMake to generate the native platform files.
- Windows
- Any Personal Computer version supported by Microsoft
- Microsoft Visual Studio
- Versions
- The Community Edition of each of the above versions is sufficient, as well as any more capable edition.
- CMake 3.17.2 is recommended.
- Use the installer option to add CMake to the system PATH
- Git Client Support
- Git for Windows is a popular solution for Windows
- Some IDEs (e.g., Visual Studio, GitHub Desktop) have integrated Git client support
The general approach is to run CMake to generate the Visual Studio project
files. Then either run CMake with the --build
option to build from the
command line or use the Visual Studio IDE to open the generated solution and
work with the solution interactively.
Open a developer command prompt and enter:
cd Vulkan-Loader
mkdir build
cd build
cmake -A x64 -DUPDATE_DEPS=ON ..
cmake --build .
The above commands instruct CMake to find and use the default Visual Studio installation to generate a Visual Studio solution and projects for the x64 architecture. The second CMake command builds the Debug (default) configuration of the solution.
Change your current directory to the top of the cloned repository directory, create a build directory and generate the Visual Studio project files:
cd Vulkan-Loader
mkdir build
cd build
cmake -DUPDATE_DEPS=ON -G "Visual Studio 16 2019" -A x64 ..
Note: The
..
parameter tellscmake
the location of the top of the repository. If you place your build directory someplace else, you'll need to specify the location of the repository differently.
The -G
option is used to select the generator
Supported Visual Studio generators:
Visual Studio 17 2022
Visual Studio 16 2019
Visual Studio 15 2017
The -A
option is used to select either the "Win32", "x64", or "ARM64 architecture.
When generating the project files, the absolute path to a Vulkan-Headers
install directory must be provided. This can be done automatically by the
-DUPDATE_DEPS=ON
option, by directly setting the
VULKAN_HEADERS_INSTALL_DIR
environment variable, or by setting the
VULKAN_HEADERS_INSTALL_DIR
CMake variable with the -D
CMake option. In
either case, the variable should point to the installation directory of a
Vulkan-Headers repository built with the install target.
The above steps create a Windows solution file named Vulkan-Loader.sln
in
the build directory.
At this point, you can build the solution from the command line or open the generated solution with Visual Studio.
While still in the build directory:
cmake --build .
to build the Debug configuration (the default), or:
cmake --build . --config Release
to make a Release build.
Launch Visual Studio and open the "Vulkan-Loader.sln" solution file in the build folder. You may select "Debug" or "Release" from the Solution Configurations drop-down list. Start a build by selecting the Build->Build Solution menu item.
The CMake project also generates an "install" target that you can use to copy the primary build artifacts to a specific location using a "bin, include, lib" style directory structure. This may be useful for collecting the artifacts and providing them to another project that is dependent on them.
The default location is $CMAKE_CURRENT_BINARY_DIR\install
, but can be changed
with the CMAKE_INSTALL_PREFIX
variable when first generating the project build
files with CMake.
You can build the install target from the command line with:
cmake --build . --config Release --target install
or build the INSTALL
target from the Visual Studio solution explorer.
Vulkan programs must be able to find and use the Vulkan loader
(vulkan-1.dll
) library as well as any other libraries the program requires.
One convenient way to do this is to copy the required libraries into the same
directory as the program. The projects in this solution copy the Vulkan loader
library and the "googletest" libraries to the build\tests\Debug
or the
build\tests\Release
directory, which is where the
vk_loader_validation_test.exe
executable is found, depending on what
configuration you built.
Other techniques include placing the library in a system folder
(C:\Windows\System32) or in a directory that appears in the PATH
environment
variable.
See the documentation in the docs
folder in this
repository for more information on how the loader finds driver libraries and
layer libraries. The document also describes both how ICDs and layers should
be packaged, and how developers can point to ICDs and layers within their
builds.
This repository has been built and tested on the two most recent Ubuntu LTS versions. Currently, the oldest supported version is Ubuntu 18.04, meaning that the minimum officially supported C++11 compiler version is GCC 5.4.0, although earlier versions may work. It should be straightforward to adapt this repository to other Linux distributions.
CMake 3.17.2 is recommended.
sudo apt-get install git build-essential libx11-xcb-dev \
libxkbcommon-dev libwayland-dev libxrandr-dev
The general approach is to run CMake to generate make files. Then either run
CMake with the --build
option or make
to build from the command line.
cd Vulkan-Loader
mkdir build
cd build
cmake -DUPDATE_DEPS=ON ..
make
See below for the details.
Change your current directory to the top of the cloned repository directory, create a build directory and generate the make files.
cd Vulkan-Loader
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Debug \
-DVULKAN_HEADERS_INSTALL_DIR=absolute_path_to_install_dir \
-DCMAKE_INSTALL_PREFIX=install ..
Note: The
..
parameter tellscmake
the location of the top of the repository. If you place yourbuild
directory someplace else, you'll need to specify the location of the repository top differently.
Use -DCMAKE_BUILD_TYPE
to specify a Debug or Release build.
When generating the project files, the absolute path to a Vulkan-Headers
install directory must be provided. This can be done automatically by the
-DUPDATE_DEPS=ON
option, by directly setting the
VULKAN_HEADERS_INSTALL_DIR
environment variable, or by setting the
VULKAN_HEADERS_INSTALL_DIR
CMake variable with the -D
CMake option. In
either case, the variable should point to the installation directory of a
Vulkan-Headers repository built with the install target.
Note: For Linux, the default value for
CMAKE_INSTALL_PREFIX
is/usr/local
, which would be used if you do not specifyCMAKE_INSTALL_PREFIX
. In this case, you may need to usesudo
to install to system directories later when you runmake install
.
You can just run make
to begin the build.
To speed up the build on a multi-core machine, use the -j
option for make
to specify the number of cores to use for the build. For example:
make -j4
You can also use
cmake --build .
If you want to run a Vulkan application that is not built within this repository with the loader you just built from this repository, you can direct the application to load it from your build directory:
export LD_LIBRARY_PATH=<path to your repository root>/build/loader
By default, the Vulkan Loader is built with support for the Vulkan-defined WSI
display servers: Xcb, Xlib, and Wayland. It is recommended to build the
repository components with support for these display servers to maximize their
usability across Linux platforms. If it is necessary to build these modules
without support for one of the display servers, the appropriate CMake option
of the form BUILD_WSI_xxx_SUPPORT
can be set to OFF
.
Installing the files resulting from your build to the systems directories is optional since environment variables can usually be used instead to locate the binaries. There are also risks with interfering with binaries installed by packages. If you are certain that you would like to install your binaries to system directories, you can proceed with these instructions.
Assuming that you've built the code as described above and the current
directory is still build
, you can execute:
sudo make install
This command installs files to /usr/local
if no CMAKE_INSTALL_PREFIX
is
specified when creating the build files with CMake:
/usr/local/lib
: Vulkan loader library and package config files
You may need to run ldconfig
in order to refresh the system loader search
cache on some Linux systems.
You can further customize the installation location by setting additional
CMake variables to override their defaults. For example, if you would like to
install to /tmp/build
instead of /usr/local
, on your CMake command line
specify:
-DCMAKE_INSTALL_PREFIX=/tmp/build
Then run make install
as before. The install step places the files in
/tmp/build
. This may be useful for collecting the artifacts and providing
them to another project that is dependent on them.
Using the CMAKE_INSTALL_PREFIX
to customize the install location also
modifies the loader search paths to include searching for layers in the
specified install location. In this example, setting CMAKE_INSTALL_PREFIX
to
/tmp/build
causes the loader to search
/tmp/build/etc/vulkan/explicit_layer.d
and
/tmp/build/share/vulkan/explicit_layer.d
for the layer JSON files. The
loader also searches the "standard" system locations of
/etc/vulkan/explicit_layer.d
and /usr/share/vulkan/explicit_layer.d
after
searching the two locations under /tmp/build
.
You can further customize the installation directories by using the CMake
variables CMAKE_INSTALL_SYSCONFDIR
to rename the etc
directory and
CMAKE_INSTALL_DATADIR
to rename the share
directory.
See the CMake documentation for more details on using these variables to further customize your installation.
Also see the LoaderAndLayerInterface
document in the loader
folder in this
repository for more information about loader operation.
Note that some executables in this repository (e.g.,
vk_loader_validation_tests
) use the RPATH linker directive to load the
Vulkan loader from the build directory, build
in this example. This means
that even after installing the loader to the system directories, these
executables still use the loader from the build directory.
Usage of this repository's contents in 32-bit Linux environments is not officially supported. However, since this repository is supported on 32-bit Windows, these modules should generally work on 32-bit Linux.
Here are some notes for building 32-bit targets on a 64-bit Ubuntu "reference" platform:
If not already installed, install the following 32-bit development libraries:
gcc-multilib g++-multilib libx11-dev:i386
This list may vary depending on your distribution and which windowing systems you are building for.
Set up your environment for building 32-bit targets:
export ASFLAGS=--32
export CFLAGS=-m32
export CXXFLAGS=-m32
export PKG_CONFIG_LIBDIR=/usr/lib/i386-linux-gnu
Again, your PKG_CONFIG configuration may be different, depending on your distribution.
Finally, rebuild the repository using cmake
and make
, as explained above.
Tested on OSX version 10.12
NOTE: To force the OSX version set the environment variable MACOSX_DEPLOYMENT_TARGET when building VVL and it's dependencies.
Setup Homebrew and components
-
Ensure Homebrew is at the beginning of your PATH:
export PATH=/usr/local/bin:$PATH
-
Add packages with the following (may need refinement)
brew install python python3 git
Clone the Vulkan-ValidationLayers repository:
git clone https://github.com/KhronosGroup/Vulkan-ValidationLayers.git
CMake 3.17.2 is recommended.
This generator is the default generator.
When generating the project files, the absolute path to a Vulkan-Headers
install directory must be provided. This can be done automatically by the
-DUPDATE_DEPS=ON
option, by directly setting the
VULKAN_HEADERS_INSTALL_DIR
environment variable, or by setting the
VULKAN_HEADERS_INSTALL_DIR
CMake variable with the -D
CMake option. In
either case, the variable should point to the installation directory of a
Vulkan-Headers repository built with the install target.
mkdir build
cd build
cmake -DUPDATE_DEPS=ON -DVULKAN_HEADERS_INSTALL_DIR=absolute_path_to_install_dir -DCMAKE_BUILD_TYPE=Debug ..
make
To speed up the build on a multi-core machine, use the -j
option for make
to specify the number of cores to use for the build. For example:
make -j4
To create and open an Xcode project:
mkdir build-xcode
cd build-xcode
cmake -GXcode ..
open Vulkan-Loader.xcodeproj
Within Xcode, you can select Debug or Release builds in the project's Build Settings.
If you want to test a Vulkan application with the loader you just built, you can direct the application to load it from your build directory:
export DYLD_LIBRARY_PATH=<path to your repository>/build/loader
Fuchsia uses the project's GN build system to integrate with the Fuchsia platform build.
QNX is using its own build system. The proper build environment must be set under the QNX host development system (Linux, Win64, MacOS) by invoking the shell/batch script provided with QNX installation.
Then change working directory to the "build-qnx" in this project and type "make". It will build the ICD loader for all CPU targets supported by QNX.
The Vulkan Loader is a component of the Fuchsia SDK, so it must explicitly declare its exported symbols in the file vulkan.symbols.api; see SDK.
While this repo is capable of cross compilation, there are a handful of caveats.
Unknown function handling is only fully supported on select platforms due to the
need for assembly in the implementation. Other platforms will need to disable
assembly by setting USE_GAS
or USE_MASM
to OFF
.
- 64 bit Windows (x64)
- 32 bit Windows (x86)
- 64 bit Linux (x64)
- 32 bit Linux (x86)
- 64 bit Arm (aarch64)
Platforms not listed will use a fallback C Code path that relies on tail-call optimization to work. No guarantees are made about the use of the fallback code paths.
When cross compiling, the use of Link Time Optimization (LTO) and unknown function handling is not supported. Either LTO needs to be turned off, or the assembly should be disabled.
To build tests, make sure that the BUILD_TESTS
option is set to true. Using
the command line, this looks like -DBUILD_TESTS=ON
.
This project is configured to run with ctest
, which makes it easy to run the
tests. To run the tests, change the directory to that of the build direction, and
execute ctest
.
More details can be found in the README.md for the tests directory of this project.