bensepanski / bricklib

Distributed Performance-portable Stencil Compuitation

Brick Layout for C++

Distributed Performance-portable Stencil Compuitation - [email protected]

Compilation requirements

• C++14 compatible compiler
• OpenMP
• MPI library
• CMake
• Optional backends
  • CUDA
  • OpenCL
  • SYCL/DPC++
  • HIP

We uses CMake to find libraries. If some library fails to load, be sure to check the find module for it in CMake and set the corresponding paths. Sometimes, libraries may not adhere to standards or CMake may not be up-to-date. In such cases, setting CMake variables directly in the top-most CMakeLists.txt is recommended.

Building and running

1. Clone the repository
2. Create a build directory inside the source tree mkdir build
3. Create build configuration cd build && cmake .. -DCMAKE_BUILD_TYPE=Release
4. Build different test cases using make <testname>

For description of the test cases see here.

For the list of configuration variables, use cmake -L .. or cmake-gui in the build directory.

This library can be used as either a submodule or be installed and linked to. examples/external provide an example of linking to this library from an external project.

Using the brick template

The brick template consists of 3 part:

• Brick: declare brick data structure
• BrickInfo: an adjacency list that describes the relations between bricks
• BrickStorage: a chunk of memory for storing bricks

The behavior of such templated data structures are as normal: they do not require the use of code generator to function; provide a fallback way of writing code for compute & data movement.

Stencil Expression Description

Stencil expression for code generator are specified using Python library. Code generator provide optimization and vectorization support for different backend.

The code generation are carried out by CMake wrapper automatically. For details, see Codegen Integration.

Dimension Ordering

Template arguments & code ordering is contiguous dimension last. Dimension arrays are contiguous at 0 (contiguous first).

Directory & Files

• include and src contains the brick library headers and library files.
• docs various documents
• cmake CMake module file
• Included test cases are split into 4 folders:
  • stencils contains different stencils and related initialization code used by all tests as needed
  • single for single node (no MPI)
  • weak for weak scaling or strong scaling with one-level decomposition (one subdomain per rank)
  • strong for strong scaling with two-level decomposition (multiple fixed-sized subdomains per rank)

A large portion of the brick library is entirely based on templates and can be included as a header only library.

Acknowledgements

• This research was supported by the Exascale Computing Project (17-SC-20-SC), a joint project of the U.S. Department of Energy's Office of Science and National Nuclear Security Administration.
• This research used resources of the Oak Ridge Leadership Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.
• This research used resources of the Argonne Leadership Computing Facility at Argonne National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under contract DE-AC02-06CH11357.
• This research used resources in Lawrence Berkeley National Laboratory and the National Energy Research Scientific Computing Center, which are supported by the U.S. Department of Energy Office of Science’s Advanced Scientific Computing Research program under contract number DE-AC02-05CH11231.

Publications

@cite zhao2018 Zhao, Tuowen, Samuel Williams, Mary Hall, and Hans Johansen. 2018. Delivering Performance-Portable Stencil Computations on CPUs and GPUs Using Bricks. In 2018 IEEE/ACM International Workshop on Performance, Portability and Productivity in HPC (P3HPC). 59-70. DOI:https://doi.org/10.1109/P3HPC.2018.00009

@cite zhao2019 Tuowen Zhao, Protonu Basu, Samuel Williams, Mary Hall, and Hans Johansen. 2019. Exploiting reuse and vectorization in blocked stencil computations on CPUs and GPUs. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC '19). Association for Computing Machinery, New York, NY, USA, Article 52, 1–44. DOI:https://doi.org/10.1145/3295500.3356210

@cite zhao2021 Tuowen Zhao, Mary Hall, Hans Johansen, and Samuel Williams. 2021. Improving communication by optimizing on-node data movement with data layout. In Proceedings of the 26th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP '21). Association for Computing Machinery, New York, NY, USA, 304–317. DOI:https://doi.org/10.1145/3437801.3441598