Omniverse PhysX vcpkg port
The newest version of the open source NVIDIA Omniverse PhysX, much like its predecessors, supports a large variety of compiler switches, configuration presets and dependency management infrastructure. Sometimes this complexity can be quite intimidating/frustrating to newcomers willing to experiment with it (maybe when considering adoption in a bigger physics project).
Luckily users can now use a rather involved contributed vcpkg port to compile and link Omniverse PhysX with GPU acceleration support directly into their projects with just a few lines of CMake code after installing vcpkg.
Here’s a simple hello_physx_vcpkg project stub:
- CMakeLists.txt
# CMakeLists.txt
cmake_minimum_required(VERSION 3.15)
# CUSTOMIZE THESE IF NEEDED BEFORE IMPORTING THE PHYSX PACKAGE
##############################################################
# set(VCPKG_TARGET_TRIPLET x64-windows)
# set(VCPKG_CRT_LINKAGE dynamic)
# set(VCPKG_TARGET_TRIPLET x64-windows-static)
# set(VCPKG_CRT_LINKAGE static)
set(VCPKG_TARGET_TRIPLET x64-linux)
##############################################################
# IMPORTANT: set the following variable to the location of your vcpkg local installation
set(CMAKE_TOOLCHAIN_FILE "$ENV{HOME}/vcpkg/scripts/buildsystems/vcpkg.cmake")
project(HelloPhysX)
add_executable(hello_physx main.cpp)
# Note: if the package cannot be found here, check that you're using the right triplet
find_package(unofficial-omniverse-physx-sdk CONFIG REQUIRED)
target_link_libraries(hello_physx PRIVATE unofficial::omniverse-physx-sdk::sdk)
# Optional: import the defined target to copy over the GPU acceleration libraries
# (3rd party provided by NVIDIA)
if(TARGET unofficial::omniverse-physx-sdk::gpu-library)
if(UNIX)
# Add rpath setting to find so libraries on unix based systems
set_target_properties(hello_physx PROPERTIES
BUILD_WITH_INSTALL_RPATH TRUE
INSTALL_RPATH "$ORIGIN"
)
endif()
add_custom_command(TARGET hello_physx POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different
$<TARGET_FILE:unofficial::omniverse-physx-sdk::gpu-library>
$<TARGET_FILE_DIR:hello_physx>)
if(WIN32)
add_custom_command(TARGET hello_physx POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different
$<TARGET_FILE:unofficial::omniverse-physx-sdk::gpu-device-library>
$<TARGET_FILE_DIR:hello_physx>)
endif()
else()
message(WARNING "\GPU acceleration library target not defined
- GPU acceleration will NOT be available!\
")
endif()
- main.cpp
#include "PxConfig.h"
#include "PxPhysicsAPI.h"
#include <iostream>
using namespace physx;
int main(int argc, char** argv) {
std::cout << "Starting PhysX up.." << std::endl;
PxDefaultAllocator allocator;
PxDefaultErrorCallback error_callback;
auto foundation = PxCreateFoundation(PX_PHYSICS_VERSION, allocator, error_callback);
// Create a CUDA context manager
PxCudaContextManagerDesc cudaContextManagerDesc;
auto cudaContextManager = PxCreateCudaContextManager(
*foundation, cudaContextManagerDesc, PxGetProfilerCallback()
);
// Check if CUDA context manager is valid
if(cudaContextManager && cudaContextManager->contextIsValid()) {
std::cout << "GPU acceleration is available" << std::endl;
} else {
std::cout << "GPU acceleration is not available" << std::endl;
}
// Create a physics SDK object with GPU acceleration
PxTolerancesScale tolerancesScale;
PxSceneDesc sceneDesc(tolerancesScale);
sceneDesc.gravity = PxVec3(0.0f, -9.81f, 0.0f);
auto dispatcher = PxDefaultCpuDispatcherCreate(4);
sceneDesc.cpuDispatcher = dispatcher;
sceneDesc.filterShader = PxDefaultSimulationFilterShader;
sceneDesc.cudaContextManager = cudaContextManager;
sceneDesc.flags |= PxSceneFlag::eENABLE_GPU_DYNAMICS;
sceneDesc.broadPhaseType = PxBroadPhaseType::eGPU;
auto physics_sdk = PxCreatePhysics(
PX_PHYSICS_VERSION, *foundation, tolerancesScale, true, nullptr
);
auto scene = physics_sdk->createScene(sceneDesc);
std::cout << "PhysX set up" << std::endl;
// Release resources
PX_RELEASE(scene);
PX_RELEASE(dispatcher);
PX_RELEASE(physics_sdk);
PX_RELEASE(cudaContextManager);
PX_RELEASE(foundation);
std::cout << "Shutting down.." << std::endl;
}
After that it’s as simple as
# Git-checkout, compile and install newest PhysX locally
# One could also use 'vcpkg install physx --triplet linux-x64' here
~/hello_physx_vcpkg$ vcpkg install physx
...
~/hello_physx_vcpkg$ mkdir build
~/hello_physx_vcpkg$ cd build
~/hello_physx_vcpkg/build$ cmake -DCMAKE_BUILD_TYPE=Release .. && make
~/hello_physx_vcpkg/build$./hello_physx
Starting PhysX up..
or similarly on Windows
# optionally specify vcpkg-root if cannot be found in PATH
# --triplet x64-windows/x64-windows-static --vcpkg-root C:\Users\user\vcpkg
~/hello_physx_vcpkg$ vcpkg install physx
...
~/hello_physx_vcpkg$ mkdir build
~/hello_physx_vcpkg$ cd build
~/hello_physx_vcpkg/build$ cmake -DCMAKE_BUILD_TYPE=Release ..
~/hello_physx_vcpkg/build$ cmake --build . --config Release
~/hello_physx_vcpkg/build$./Release/hello_physx
Starting PhysX up..
CPU version
To install and build a CPU-only version of the PhysX engine (one which does not use CUDA), you can create a custom triplet file (e.g. create a text file vcpkg/triplets/community/x64-linux-nocuda.cmake) which inherits your desired triplet and ensures a CPU version is installed and built:
# Inherit all flags but also add a PX_GENERATE_GPU_PROJECTS as FALSE
# to compile a CPU-only version of the PhysX engine
include(${CMAKE_CURRENT_LIST_DIR}/../x64-linux.cmake)
set(PX_GENERATE_GPU_PROJECTS FALSE)
If you build a CPU-only version, remember to drop the unofficial::omniverse-physx-sdk::gpu-library target from the targets above.
Support
The port is rather involved and will be community maintained and updated (this is NOT an official NVIDIA supported port). For any issues please open a GitHub issue to the vcpkg official repo and hopefully someone will take care of it.
If you need an older version of the PhysX SDK you can still use versioning to checkout it directly