Advanced Build Configuration

Building With Build Script

The simplest way to build USD is to run the supplied build_usd.py script. This script will download required dependencies and build and install them along with USD in a given directory.

See instructions and examples in README.md.

Building With CMake

Users may specify libraries to build USD against and other build options by passing arguments when running cmake. Documentation for these arguments are below.

Some examples:

On Linux

cmake                                       \
-DTBB_ROOT_DIR=/path/to/tbb                 \
-DOPENSUBDIV_ROOT_DIR=/path/to/opensubdiv   \
/path/to/USD/source

cmake --build . --target install -- -j <NUM_CORES>

On macOS

The following will generate an Xcode project that can be used to build USD.

cmake                                       \
-G "Xcode"                                  \
-DTBB_ROOT_DIR=/path/to/tbb                 \
-DOPENSUBDIV_ROOT_DIR=/path/to/opensubdiv   \
/path/to/USD/source

cmake --build . --target install -- -j <NUM_CORES>

On Windows

The following will generate a Visual Studio 2017 solution that can be used to build USD.

"C:\Program Files\CMake\bin\cmake.exe"      ^
-G "Visual Studio 15 2017 Win64"            ^
-DTBB_ROOT_DIR=C:\path\to\tbb               ^
-DOPENSUBDIV_ROOT_DIR=C:\path\to\opensubdiv ^
\path\to\USD\source

cmake --build . --target install -- /m:%NUMBER_OF_PROCESSORS%

For other versions of Visual Studio, use the following cmake arguments:

For VS2019: -G "Visual Studio 16 2019" -A x64
For VS2022: -G "Visual Studio 17 2022" -A x64

For more information on Visual Studio generators for cmake, see Visual Studio Generators.

Optional Components

USD contains several optional components that are enabled by default but may be disabled when invoking cmake. Disabling these components removes the need for their dependencies when building USD.

Python

Some optional USD components use Python:

The USD Toolset
Third Party Plugins
Python language bindings for the USD C++ API
Unit tests using Python

Please refer to VERSIONS.md for supported Python versions.

Support for Python can optionally be disabled by specifying the cmake flag PXR_ENABLE_PYTHON_SUPPORT=FALSE.

OpenGL

Support for OpenGL can optionally be disabled by specifying the cmake flag PXR_ENABLE_GL_SUPPORT=FALSE. This will skip components and libraries that depend on GL, including:

usdview
Hydra GL imaging

Metal

Building USD with Metal enabled requires macOS Mojave (10.14) or newer. Support for Metal can optionally be disabled by specifying the cmake flag PXR_ENABLE_METAL_SUPPORT=FALSE. This will skip components and libraries that depend on Metal, including:

Hydra imaging

Vulkan

Vulkan support for USD is currently experimental and disabled by default.

Building USD with Vulkan enabled requires the Vulkan SDK and glslang to be installed. If you wish to build a debug build on Windows, please make sure you have the optional component "Shader Toolchain Debug Symbols" installed as part of the Vulkan SDK.

The VULKAN_SDK environment variable must point to the location of the SDK. The glslang compiler headers must be locatable during the build process.

Support for Vulkan can optionally be enabled by specifying the cmake flag PXR_ENABLE_VULKAN_SUPPORT=TRUE.

MaterialX

Enable MaterialX support in the build by specifying the cmake flag PXR_ENABLE_MATERIALX_SUPPORT=TRUE when invoking cmake. Note that MaterialX with shared library support is required.

When building via build_usd.py, MaterialX support is enabled by default. The default can be overriden using the --materialx and --no-materialx flags.

The additional dependencies that must be supplied when invoking cmake are:

Dependency Name	Description
MaterialX_DIR	Path to the CMake package config of a MaterialX SDK install.

See 3rd Party Library and Application Versions for version information.

OSL (OpenShadingLanguage)

Support for OSL is disabled by default, and can optionally be enabled by specifying the cmake flag PXR_ENABLE_OSL_SUPPORT=TRUE. This will enable components and libraries that depend on OSL.

Enabling OSL suport allows the Shader Definition Registry (sdr) to parse metadata from OSL shaders.

Documentation

Doxygen documentation can optionally be generated by specifying the cmake flag PXR_BUILD_DOCUMENTATION=TRUE.

There are two documentation sub-components, controlled by setting PXR_BUILD_HTML_DOCUMENTATION and PXR_BUILD_PYTHON_DOCUMENTATION. See the sections below for descriptions of precisely what they control. Neither sub- component will have any effect if PXR_BUILD_DOCUMENTATION is not TRUE.

The additional dependencies that must be supplied for enabling documentation generation are:

Dependency Name	Description
DOXYGEN_EXECUTABLE	The location of Doxygen

See 3rd Party Library and Application Versions for version information, including supported Doxygen versions.

HTML Documentation

If PXR_BUILD_HTML_DOCUMENTATION evaluates TRUE then building of the HTML documentation is enabled. This includes USD overview and general concepts, and the C++ API documentation. It defaults to TRUE, and so is built by default when PXR_BUILD_DOCUMENTATION is also enabled (but has no effect if PXR_BUILD_DOCUMENTATION is not set.)

The additional dependencies that must be supplied for enabling HTML documentation generation are:

Dependency Name	Description
DOT_EXECUTABLE	The location of Dot(from GraphViz).

See 3rd Party Library and Application Versions for version information, including supported GraphViz versions.

Python Documentation

Python docstrings for Python entities can be generated by specifying the cmake flag PXR_BUILD_PYTHON_DOCUMENTATION. This process requires that Python support (PXR_ENABLE_PYTHON_SUPPORT) and documentation (PXR_BUILD_DOCUMENTATION) are enabled. It defaults to FALSE.

This process uses the scripts in the docs/python subdirectory. Relevant documentation from generated doxygen XML data is extracted and matched with associated Python classes, functions, and properties in the built Python modules. A __DOC.py file is generated and installed in each of the directories of the installed Python modules. The __DOC.py file adds the docstrings to the Python entities when the module is loaded.

Imaging

This component contains Hydra, a high-performance graphics rendering engine.

Disable this component by specifying the cmake flag PXR_BUILD_IMAGING=FALSE when invoking cmake. Disabling this component will also disable the USD Imaging component and any Imaging Plugins.

USD Imaging

This component provides the USD imaging delegates for Hydra, as well as usdview, a standalone native viewer for USD files.

Disable this component by specifying the cmake flag PXR_BUILD_USD_IMAGING=FALSE when invoking cmake. usdview may also be disabled independently by specifying the cmake flag PXR_BUILD_USDVIEW=FALSE.

Command-line Tools

USD by default builds several helpful command-line tools for validating and manipulating USD files. For more information on the tools, see USD Toolset.

Disable building the command-line tools by specifying the cmake flag PXR_BUILD_USD_TOOLS=FALSE when invoking cmake.

Examples

USD by default builds several example projects that demonstrate how to develop various extensions and plugins.

Disable building the examples by specifying the cmake flag PXR_BUILD_EXAMPLES=FALSE when invoking cmake.

Tutorials

USD by default builds USD and Python files used for the USD Tutorials.

Disable building the tutorial support files by specifying the cmake flag PXR_BUILD_TUTORIALS=FALSE when invoking cmake.

Imaging Plugins

Hydra's rendering functionality can be extended with these optional plugins.

Ptex

Support for Ptex is disabled by default and can be enabled by specifying the cmake flag PXR_ENABLE_PTEX_SUPPORT=TRUE.

OpenImageIO

This plugin can optionally be enabled by specifying the cmake flag PXR_BUILD_OPENIMAGEIO_PLUGIN=TRUE. When enabled, OpenImageIO provides broader support for reading and writing different image formats as textures. If OpenImageIO is disabled, imaging by default supports the image formats bmp, jpg, png, tga, and hdr. With OpenImageIO enabled, support extends to exr, tif, zfile, and tx file formats, which allows for the use of more advanced features like subimages and mipmaps.

OpenColorIO

This plugin can optionally be enabled by specifying the cmake flag PXR_BUILD_OPENCOLORIO_PLUGIN=TRUE. When enabled, OpenColorIO provides color management for Hydra viewports.

Embree Rendering

This component contains an example rendering backend for Hydra and usdview, based on the embree raycasting library. Enable the plugin in the build by specifying the cmake flag PXR_BUILD_EMBREE_PLUGIN=TRUE when invoking cmake.

The additional dependencies that must be supplied when invoking cmake are:

Dependency Name	Description
EMBREE_LOCATION	The root path to an embree library install.

See 3rd Party Library and Application Versions for version information.

RenderMan Rendering

This plugin uses Pixar's RenderMan as a rendering backend for Hydra and usdview. Enable the plugin in the build by specifying the cmake flag PXR_BUILD_PRMAN_PLUGIN=TRUE when invoking cmake.

The additional dependencies that must be supplied when invoking cmake are:

Dependency Name	Description
RENDERMAN_LOCATION	The root path to an RenderMan install.

See 3rd Party Library and Application Versions for version information.

Third Party Plugins

USD provides several plugins for integration with third-party software packages. There is additional documentation on each plugin here. These plugins are not built by default and must be enabled via the instructions below.

The USD Maya plugins can be found in the Autodesk-supported repo available here.

The USD Katana plugins can be found in the Foundry-supported repo available here.

Alembic Plugin

Enable the Alembic plugin in the build by specifying the cmake flag PXR_BUILD_ALEMBIC_PLUGIN=TRUE when invoking cmake.

The additional dependencies that must be supplied when invoking cmake are:

Dependency Name	Description
ALEMBIC_DIR	The location of Alembic
OPENEXR_LOCATION	The location of OpenEXR
Imath_DIR (If not using OpenEXR)	Path to the CMake package config of a Imath SDK install. (With OpenEXR 3+, Imath can be used explicitly instead of OpenEXR.)

Either OpenEXR or Imath is required depending on which library is used by the Alembic library specified in ALEMBIC_DIR.

See 3rd Party Library and Application Versions for version information.

Support for Alembic files using the HDF5 backend is enabled by default but can be disabled by specifying the cmake flag PXR_ENABLE_HDF5_SUPPORT=FALSE. HDF5 support requires the following dependencies:

Dependency Name	Description
HDF5_LOCATION	The location of HDF5

For further information see the documentation on the Alembic plugin here.

Draco Plugin

Enable the Draco plugin in the build by specifying the cmake flag PXR_BUILD_DRACO_PLUGIN=TRUE when invoking cmake. This plugin is compatible with Draco 1.3.4. The additional dependencies that must be supplied when invoking cmake are:

Dependency Name	Description	Version
DRACO_ROOT	The root path to a Draco SDK install.	1.3.4

Tests

Tests are built by default but can be disabled by specifying the cmake flag PXR_BUILD_TESTS=FALSE when invoking cmake.

Running Tests

Run tests by invoking ctest from the build directory, which is typically the directory in which cmake was originally invoked. For example, to run all tests in a release build with verbose output:

ctest -C Release -V

The "-R" argument may be used to specify a regular expression matching the names of tests to be run. For example, to run all tests in a release build matching "testUsdShade" with verbose output:

ctest -C Release -R testUsdShade -V

See the ctest documentation for more options.

Test Run Directories

Each test is run out of an automatically-created temporary directory containing any additional files required by the test. Set the cmake option PXR_TEST_RUN_TEMP_DIR_PREFIX to prepend a prefix string to the name of these directories. For example, setting this option to "foo-" will create test run directories named "foo-"

Diagnosing Failed Tests

In order to aid with diagnosing of failing tests, test generated files for failing test are explicitly put in the following directories, where <ctest_run_timestamp> (formatted as "%Y-%m-%dT%H.%M.%S") represents the timestamp when ctest was run for the failing test.

${CMAKE_BINARY_DIR}/Testing/Failed-Diffs/<ctest_run_timestamp>/${TEST_NAME}/${filename}.result.${ext}
${CMAKE_BINARY_DIR}/Testing/Failed-Diffs/<ctest_run_timestamp>/${TEST_NAME}/${filename}.baseline.${ext}

Other Build Options

Plugin Metadata Location

Each library in the USD core generally has an associated file named 'plugInfo.json' that contains metadata about that library, such as the schema types provided by that library. These files are consumed by USD's internal plugin system to lazily load libraries when needed.

The plugin system requires knowledge of where these metadata files are located. The cmake build will ensure this is set up properly based on the install location of the build. However, if you plan to relocate these files to a new location after the build, you must inform the build by setting the cmake variable PXR_INSTALL_LOCATION to the intended final directory where these files will be located. This variable may be a ':'-delimited list of paths.

Another way USD is locating plugins is the PXR_PLUGINPATH_NAME environment variable. This variable may be a list of paths. If you do not want your USD build to use this default variable name, you can override the name of the environment variable using the following CMake option:

-DPXR_OVERRIDE_PLUGINPATH_NAME=CUSTOM_USD_PLUGINPATHS

By doing this, USD will check the CUSTOM_USD_PLUGINPATHS environment variable for paths, instead of the default PXR_PLUGINPATH_NAME one.

The values specified in PXR_PLUGINPATH_NAME or PXR_INSTALL_LOCATION have the following characteristics:

Values may contain any number of paths.
Paths ending with slash ('/') have 'plugInfo.json' appended automatically.
'*' may be used anywhere to match any character except slash.
'**' may be used anywhere to match any character including slash.
Paths follow Unix '$PATH'-like conventions; when duplicate definitions exist in the path, the first one found is used.

Shared library prefix

By default shared libraries will have the prefix 'lib'. This means, for a given component such as usdGeom, the build will generate a corresponding libusdGeom object (libusdGeom.so on Linux, libusdGeom.dll on Windows and libusdGeom.dylib on Mac). You can change the prefix (or remove it) through PXR_LIB_PREFIX. For example,

-DPXR_LIB_PREFIX=pxr

Will generate pxrusdGeom.so on Linux, pxrusdGeom.dll on Windows and pxrusdGeom.dylib on Mac for the usdGeom component.

Note: This prefix does not apply to shared objects used for Python bindings.

USD Developer Options

C++ Namespace Configuration

USD comes with options to enable and customize C++ namespaces via the following flags:

Option Name	Description	Default
PXR_SET_EXTERNAL_NAMESPACE	The outer namespace identifier	`pxr`
PXR_SET_INTERNAL_NAMESPACE	The internal namespace identifier	`pxrInternal_v_x_y` (for version x.y.z)
PXR_ENABLE_NAMESPACES	Enable namespaces	`ON`

When enabled, there are a set of macros provided in a generated header, pxr/pxr.h, which facilitates using namespaces:

Macro Name	Description
PXR_NAMESPACE_OPEN_SCOPE	Opens the namespace scope.
PXR_NAMESPACE_CLOSE_SCOPE	Closes the namespace.
PXR_NS	Explicit qualification on items, e.g. `PXR_NS::TfToken foo = ...`
PXR_NAMESPACE_USING_DIRECTIVE	Enacts a using-directive, e.g. `using namespace PXR_NS;`

ASCII Parser Editing/Validation

There is an ASCII parser for the USD file format, which can be found in sdf. Most users will not have a need to edit the parser, but for the adventurous ones, there are a couple additional requirements.

If you choose to edit the ASCII parsers, make sure PXR_VALIDATE_GENERATED_CODE is set to TRUE. This flag enables tests that check the generated code in sdf and gf.

Dependency Name	Description
FLEX_EXECUTABLE	Path to flex executable
BISON_EXECUTABLE	Path to bison executable

See 3rd Party Library and Application Versions for version information.

USD Schema Generation

USD generates some code through a process called [schema generation]. This process requires the following python modules be installed and available on the syspath. You can learn more about Schemas and why you might want to generate them here.

Python Module Name	Description
Jinja2	Jinja is the core code generator of usdGenSchema
Argparse	Argparse is used for basic command line arguments

See 3rd Party Library and Application Versions for version information.

Optimization Options

There are certain optimizations that can be enabled in the build.

Malloc Library

We've found that USD performs best with allocators such as Jemalloc. In support of this, you can specify your own allocator through PXR_MALLOC_LIBRARY. This variable should be set to a path to a shared object for the allocator. For example,

-DPXR_MALLOC_LIBRARY:path=/usr/local/lib/libjemalloc.so

If none are specified, the default allocator will be used. More information on getting the most out of USD can be found Getting the Best Performance with USD.

Linker Options

There are four ways to link USD controlled by the following options:

Option Name	Default	Description
BUILD_SHARED_LIBS	`ON`	Build shared or static libraries
PXR_BUILD_MONOLITHIC	`OFF`	Build single or several libraries
PXR_MONOLITHIC_IMPORT		CMake file defining usd_ms import library

Shared Libraries

The default creates several shared libraries. This option allows loading just the libraries necessary for a given task.

Option Name	Value
BUILD_SHARED_LIBS	`ON`
PXR_BUILD_MONOLITHIC	`OFF`
PXR_MONOLITHIC_IMPORT

cmake -DBUILD_SHARED_LIBS=ON ...

Static Libraries

This mode builds several static libraries. This option allows embedding just the libraries necessary for a given task. However, it does not allow USD plugins or Python modules since that would necessarily cause multiple symbol definitions; for any given symbol we'd have an instance in the main application and another in each plugin/module.

Option Name	Value
BUILD_SHARED_LIBS	`OFF`
PXR_BUILD_MONOLITHIC	`OFF`
PXR_MONOLITHIC_IMPORT

cmake -DBUILD_SHARED_LIBS=OFF ...

Internal Monolithic Library

This mode builds the core libraries (i.e. everything under pxr/) into a single archive library, 'usd_m', and from that it builds a single shared library, 'usd_ms'. It builds plugins outside of pxr/ and Python modules as usual except they link against 'usd_ms' instead of the individual libraries of the default mode. Plugins inside of pxr/ are compiled into 'usd_m' and 'usd_ms'. plugInfo.json files under pxr/ refer to 'usd_ms'.

This mode is useful to reduce the number of installed files and simplify linking against USD.

Option Name	Value
BUILD_SHARED_LIBS	Don't care
PXR_BUILD_MONOLITHIC	`ON`
PXR_MONOLITHIC_IMPORT

cmake -DPXR_BUILD_MONOLITHIC=ON ...

External Monolithic Library

This mode is similar to the Internal Monolithic Library except the client has control of building the monolithic shared library. This mode is useful to embed USD into another shared library. The build steps are significantly more complicated and are described below.

Option Name	Value
BUILD_SHARED_LIBS	Don't care
PXR_BUILD_MONOLITHIC	`ON`
PXR_MONOLITHIC_IMPORT	Path-to-import-file

To build in this mode:

Choose a path where the import file will be. You'll be creating a cmake file with add_library(usd_ms SHARED IMPORTED) and one or more set_property calls. The file doesn't need to exist. If it does exist it should be empty or valid cmake code.
Configure the build in the usual way but with PXR_BUILD_MONOLITHIC=ON and PXR_MONOLITHIC_IMPORT set to the path in step 1.
Build the usual way except the target is monolithic.
Create your shared library. If using cmake you can include the file pxr/usd-targets-<CONFIG> under the USD binary (build) directory, where <CONFIG> is the configuration you built in step 3. Then you can link your library against 'usd_m'. However, this isn't as simple as target_link_libraries(mylib PUBLIC usd_m) because you must get everything from 'usd_m'. See Linking Whole Archives for more details.

Edit the import file to describe your library. Your cmake build may be able to generate the file directly via export(). The USD build will include this file and having done so must be able to link against your library by adding 'usd_ms' as a target link library. The file should look something like this:

add_library(usd_ms SHARED IMPORTED)
set_property(TARGET usd_ms PROPERTY IMPORTED_LOCATION ...)
# The following is necessary on Windows.
#set_property(TARGET usd_ms PROPERTY IMPORTED_IMPLIB ...)
set_property(TARGET usd_ms PROPERTY INTERFACE_COMPILE_DEFINITIONS ...)
set_property(TARGET usd_ms PROPERTY INTERFACE_INCLUDE_DIRECTORIES ...)
set_property(TARGET usd_ms PROPERTY INTERFACE_LINK_LIBRARIES ...)

Complete the USD build by building the usual way, either with the default target or the 'install' target.

Two notes:

Your library does not need to be named usd_ms. That's simply the name given to it by the import file. The IMPORTED_LOCATION has the real name and path to your library.
USD currently only supports installations where your library is in the same directory the USD library/libraries would have been relative to the other installed USD files. Specifically, the location of your library will be used to find plugInfo.json files using the relative paths ../share/usd/plugins and ../plugin/usd.

Linking Whole Archives

Normally when linking against a static library the linker will only pull in object files that provide a needed symbol. USD has many files that have static global objects with constructors with side effects. If nothing uses any visible symbol from those object files then a normal link would not include them. The side effects will not occur and USD will not work.

To include everything you need to tell the linker to include the whole archive. That's platform dependent and you'll want code something like this:

if(MSVC)
    target_link_libraries(mylib -WHOLEARCHIVE:$<TARGET_FILE:usd_m> usd_m)
elseif(CMAKE_COMPILER_IS_GNUCXX)
    target_link_libraries(mylib -Wl,--whole-archive usd_m -Wl,--no-whole-archive)
elseif("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
    target_link_libraries(mylib -Wl,-force_load usd_m)
endif()

On Windows cmake cannot recognize 'usd_m' as a library when appended to -WHOLEARCHIVE: because it's not a word to itself so we use TARGET_FILE to get the path to the library. We also link 'usd_m' separately so cmake will add usd_m's interface link libraries, etc. This second instance doesn't increase the resulting file size because all symbols will be found in the first (-WHOLEARCHIVE) instance.

Avoiding linking statically to Python

The default build with python support will link to the python static lib for your interpreter. This is to support running python code from C++. If that is not desirable, python static linking can be disabled using the flag

-DPXR_PY_UNDEFINED_DYNAMIC_LOOKUP=ON

The primary motivating case for this is generating wheel packages for PyPI, but the parameter was made more generic in case it has other uses in the future. It is useful when we want to take advantage of python's approach to ABI compatibility.

Note that this flag has no effect on Windows, see here for more info

Spline Options

Splines (keyframe animation) are implemented by the library pxr/base/ts.

Default Anti-Regression Authoring Mode

A Ts spline provides a function from time to attribute value. Bezier math permits long tangents to create shapes that go backwards in time, resulting in non-functions. This is typically prevented at authoring time, and there are several strategies available. See pxr/base/ts/doxygen/regression.md for details of the different choices.

The hard-coded default is TsAntiRegressionKeepRatio. To set a different default:

With build_usd.py: --build-args USD,"-DPXR_TS_DEFAULT_ANTI_REGRESSION_AUTHORING_MODE=TsAntiRegression..."
With cmake: -DPXR_TS_DEFAULT_ANTI_REGRESSION_AUTHORING_MODE=TsAntiRegression...

Client code can also override the default as needed.

Build Issues FAQ

Windows and Python 3.8+ (non-Anaconda) Python 3.8 and later on Windows will no longer search PATH for DLL dependencies. Instead, clients can call os.add_dll_directory(p) to set paths to search. By default on that platform USD will iterate over PATH and add all paths using os.add_dll_directory() when importing Python modules. Users may override this by setting the environment variable PXR_USD_WINDOWS_DLL_PATH to a PATH-like string. If this is set, USD will use these paths instead.

Note that the above does not apply to Anaconda python 3.8+ interpreters, as they are modified to behave like pre-3.8 python interpreters, and so continue to use the PATH for DLL dependencies. When running under Anaconda users should configure their system the same way they did for pre-python 3.8.

Files

BUILDING.md

Latest commit

History

BUILDING.md

File metadata and controls

Advanced Build Configuration

Table of Contents

Building With Build Script

Building With CMake

On Linux

On macOS

On Windows

Optional Components

Python

OpenGL

Metal

Vulkan

MaterialX

OSL (OpenShadingLanguage)

Documentation

HTML Documentation

Python Documentation

Imaging

USD Imaging

Command-line Tools

Examples

Tutorials

Imaging Plugins

Ptex

OpenImageIO

OpenColorIO

Embree Rendering

RenderMan Rendering

Third Party Plugins

Alembic Plugin

Draco Plugin

Tests

Running Tests

Test Run Directories

Diagnosing Failed Tests

Other Build Options

Plugin Metadata Location

Shared library prefix

USD Developer Options

C++ Namespace Configuration

ASCII Parser Editing/Validation

USD Schema Generation

Optimization Options

Malloc Library

Linker Options

Shared Libraries

Static Libraries

Internal Monolithic Library

External Monolithic Library

Linking Whole Archives

Avoiding linking statically to Python

Spline Options

Default Anti-Regression Authoring Mode

Build Issues FAQ