linecode / stdgpu

stdgpu: Efficient STL-like Data Structures on the GPU

stdgpu: Efficient STL-like Data Structures on the GPU

Features | Examples | Documentation | Building | Integration | Contributing | License | Contact

Features

stdgpu is an open-source library providing several generic GPU data structures for fast and reliable data management. Multiple platforms such as CUDA, OpenMP, and HIP are supported allowing you to rapidly write highly complex agnostic and native algorithms that look like sequential CPU code but are executed in parallel on the GPU.

• Productivity. Previous libraries such as thrust, VexCL, ArrayFire or Boost.Compute focus on the fast and efficient implementation of various algorithms for contiguously stored data to enhance productivity. stdgpu follows an orthogonal approach and focuses on fast and reliable data management to enable the rapid development of more general and flexible GPU algorithms just like their CPU counterparts.

• Interoperability. Instead of providing yet another ecosystem, stdgpu is designed to be a lightweight container library. Therefore, a core feature of stdgpu is its interoperability with previous established frameworks, i.e. the thrust library, to enable a seamless integration into new as well as existing projects.

• Maintainability. Following the trend in recent C++ standards of providing functionality for safer and more reliable programming, the philosophy of stdgpu is to provide clean and familiar functions with strong guarantees that encourage users to write more robust code while giving them full control to achieve a high performance.

At its heart, stdgpu offers the following GPU data structures and containers:

atomic & atomic_ref Atomic primitive types and references	bitset Space-efficient bit array	deque Dynamically sized double-ended queue
queue & stack Container adapters	unordered_map & unordered_set Hashed collection of unique keys and key-value pairs	vector Dynamically sized contiguous array

In addition, stdgpu also provides commonly required functionality in algorithm, bit, cmath, contract, cstddef, cstlib, functional, iterator, limits, memory, mutex, ranges, utility to complement the GPU data structures and to increase their usability and interoperability.

Examples

In order to reliably perform complex tasks on the GPU, stdgpu offers flexible interfaces that can be used in both agnostic code, e.g. via the algorithms provided by thrust, as well as in native code, e.g. in custom CUDA kernels.

For instance, stdgpu is extensively used in SLAMCast, a scalable live telepresence system, to implement real-time, large-scale 3D scene reconstruction as well as real-time 3D data streaming between a server and an arbitrary number of remote clients.

Agnostic code. In the context of SLAMCast, a simple task is the integration of a range of updated blocks into the duplicate-free set of queued blocks for data streaming which can be expressed very conveniently:

#include <stdgpu/cstddef.h>             // stdgpu::index_t
#include <stdgpu/iterator.h>            // stdgpu::make_device
#include <stdgpu/unordered_set.cuh>     // stdgpu::unordered_set

class stream_set
{
public:
    void
    add_blocks(const short3* blocks,
               const stdgpu::index_t n)
    {
        set.insert(stdgpu::make_device(blocks),
                   stdgpu::make_device(blocks + n));
    }

    // Further functions

private:
    stdgpu::unordered_set<short3> set;
    // Further members
};

Native code. More complex operations such as the creation of the duplicate-free set of updated blocks or other algorithms can be implemented natively, e.g. in custom CUDA kernels with stdgpu's CUDA backend enabled:

#include <stdgpu/cstddef.h>             // stdgpu::index_t
#include <stdgpu/unordered_map.cuh>     // stdgpu::unordered_map
#include <stdgpu/unordered_set.cuh>     // stdgpu::unordered_set

__global__ void
compute_update_set(const short3* blocks,
                   const stdgpu::index_t n,
                   const stdgpu::unordered_map<short3, voxel*> tsdf_block_map,
                   stdgpu::unordered_set<short3> mc_update_set)
{
    // Global thread index
    stdgpu::index_t i = blockIdx.x * blockDim.x + threadIdx.x;
    if (i >= n) return;

    short3 b_i = blocks[i];

    // Neighboring candidate blocks for the update
    short3 mc_blocks[8]
    = {
        short3(b_i.x - 0, b_i.y - 0, b_i.z - 0),
        short3(b_i.x - 1, b_i.y - 0, b_i.z - 0),
        short3(b_i.x - 0, b_i.y - 1, b_i.z - 0),
        short3(b_i.x - 0, b_i.y - 0, b_i.z - 1),
        short3(b_i.x - 1, b_i.y - 1, b_i.z - 0),
        short3(b_i.x - 1, b_i.y - 0, b_i.z - 1),
        short3(b_i.x - 0, b_i.y - 1, b_i.z - 1),
        short3(b_i.x - 1, b_i.y - 1, b_i.z - 1),
    };

    for (stdgpu::index_t j = 0; j < 8; ++j)
    {
        // Only consider existing neighbors
        if (tsdf_block_map.contains(mc_blocks[j]))
        {
            mc_update_set.insert(mc_blocks[j]);
        }
    }
}

More examples can be found in the examples directory.

Documentation

A comprehensive introduction into the design and API of stdgpu can be found here:

• stdgpu API documentation
• thrust algorithms documentation
• Research paper

Since a core feature and design goal of stdgpu is its interoperability with thrust, it offers full support for all thrust algorithms instead of reinventing the wheel. More information about the design can be found in the related research paper.

Building

Before building the library, please make sure that all required tools and dependencies are installed on your system. Newer versions are supported as well.

Required

• C++14 compiler
  • GCC 7
    • (Ubuntu 18.04) sudo apt install g++ make
  • Clang 6
    • (Ubuntu 18.04) sudo apt install clang make
  • MSVC 19.20
    • (Windows) Visual Studio 2019 https://visualstudio.microsoft.com/downloads/
• CMake 3.15
  • (Ubuntu 18.04) https://apt.kitware.com
  • (Windows) https://cmake.org/download
• thrust 1.9.2
  • (Ubuntu/Windows) https://github.com/thrust/thrust
  • May already be installed by backend dependencies

Required for CUDA backend

• CUDA Toolkit 10.0
  • (Ubuntu/Windows) https://developer.nvidia.com/cuda-downloads
  • Includes thrust

Required for OpenMP backend

• OpenMP 2.0
  • GCC 7
    • (Ubuntu 18.04) Already installed
  • Clang 6
    • (Ubuntu 18.04) sudo apt install libomp-dev
  • MSVC 19.20
    • (Windows) Already installed

Required for HIP backend (experimental)

• ROCm 3.1
  • (Ubuntu) https://github.com/RadeonOpenCompute/ROCm
  • Includes thrust

The library can be built as every other project which makes use of the CMake build system.

In addition, we also provide cross-platform scripts to make the build process more convenient. Since these scripts depend on the selected build type, there are scripts for both debug and release builds.

Command	Effect
sh scripts/setup_<build_type>.sh	Performs a full clean build of the project. Removes old build, configures the project (build path: ./build), builds the project, and runs the unit tests.
sh scripts/build_<build_type>.sh	(Re-)Builds the project. Requires that the project is set up.
sh scripts/run_tests_<build_type>.sh	Runs the unit tests. Requires that the project is built.
sh scripts/install_<build_type>.sh	Installs the project at the configured install path (default: ./bin).

Integration

In the following, we show some examples on how the library can be integrated into and used in a project.

CMake Integration. To use the library in your project, you can either install it externally first and then include it using find_package:

find_package(stdgpu 1.0.0 REQUIRED)

add_library(foo ...)

target_link_libraries(foo PUBLIC stdgpu::stdgpu)

Or you can embed it into your project and build it from a subdirectory:

# Exclude the examples from the build
set(STDGPU_BUILD_EXAMPLES OFF CACHE INTERNAL "")

# Exclude the tests from the build
set(STDGPU_BUILD_TESTS OFF CACHE INTERNAL "")

add_subdirectory(stdgpu)

add_library(foo ...)

target_link_libraries(foo PUBLIC stdgpu::stdgpu)

CMake Options. To configure the library, two sets of options are provided. The following build options control the build process:

Build Option	Effect	Default
STDGPU_BACKEND	Device system backend	STDGPU_BACKEND_CUDA
STDGPU_BUILD_SHARED_LIBS	Builds the project as a shared library, if set to ON, or as a static library, if set to OFF	BUILD_SHARED_LIBS
STDGPU_SETUP_COMPILER_FLAGS	Constructs the compiler flags	ON if standalone, OFF if included via add_subdirectory
STDGPU_TREAT_WARNINGS_AS_ERRORS	Treats compiler warnings as errors	OFF
STDGPU_BUILD_EXAMPLES	Build the examples	ON
STDGPU_BUILD_TESTS	Build the unit tests	ON
STDGPU_BUILD_TEST_COVERAGE	Build a test coverage report	OFF
STDGPU_ANALYZE_WITH_CLANG_TIDY	Analyzes the code with clang-tidy	OFF
STDGPU_ANALYZE_WITH_CPPCHECK	Analyzes the code with cppcheck	OFF

In addition, the implementation of some functionality can be controlled via configuration options:

Configuration Option	Effect	Default
STDGPU_ENABLE_AUXILIARY_ARRAY_WARNING	Enable warnings when auxiliary arrays are allocated in memory API (deprecated)	OFF
STDGPU_ENABLE_CONTRACT_CHECKS	Enable contract checks	OFF if CMAKE_BUILD_TYPE equals Release or MinSizeRel, ON otherwise
STDGPU_ENABLE_MANAGED_ARRAY_WARNING	Enable warnings when managed memory is initialized on the host side but should be on device in memory API (deprecated)	OFF
STDGPU_USE_32_BIT_INDEX	Use 32-bit instead of 64-bit signed integer for index_t	ON
STDGPU_USE_FAST_DESTROY	Use fast destruction of allocated arrays (filled with a default value) by omitting destructor calls in memory API (deprecated)	OFF
STDGPU_USE_FIBONACCI_HASHING	Use Fibonacci Hashing instead of Modulo to compute hash bucket indices (deprecated)	ON

Contributing

For detailed information on how to contribute, see CONTRIBUTING.

License

Distributed under the Apache 2.0 License. See LICENSE for more information.

If you use stdgpu in one of your projects, please cite the following publications:

stdgpu: Efficient STL-like Data Structures on the GPU

@UNPUBLISHED{stotko2019stdgpu,
    author = {Stotko, P.},
     title = {{stdgpu: Efficient STL-like Data Structures on the GPU}},
      year = {2019},
     month = aug,
      note = {arXiv:1908.05936},
       url = {https://arxiv.org/abs/1908.05936}
}

SLAMCast: Large-Scale, Real-Time 3D Reconstruction and Streaming for Immersive Multi-Client Live Telepresence

@article{stotko2019slamcast,
    author = {Stotko, P. and Krumpen, S. and Hullin, M. B. and Weinmann, M. and Klein, R.},
     title = {{SLAMCast: Large-Scale, Real-Time 3D Reconstruction and Streaming for Immersive Multi-Client Live Telepresence}},
   journal = {IEEE Transactions on Visualization and Computer Graphics},
    volume = {25},
    number = {5},
     pages = {2102--2112},
      year = {2019},
     month = may
}

Contact

Patrick Stotko - [email protected]