ZHW - ZFP Hardware Implementation

ZFP is a floating-point compression format gaining traction in high-performance computing applications. A software implementation has demonstrated the ability to reduce data movement across communication channels and to reduce the footprint of floating-point arrays in memory. Nevertheless, any benefit in performance is limited to the spare compute cycles available before reaching bandwidth limitations. A hardware implementation of ZFP has the potential to substantially improve HPC application performance.

The hardware implementation of ZFP is sourced in SystemC to facilitate its evaluation in various architectures. The encode pipeline consists of several modules. Uncompressed blocks of floating-point numbers in IEEE format flow into the pipeline in a stream and a compressed bitstream flows out from the pipeline. The decode pipeline is the inverse of the encode pipeline. A modular design enables number formats other than IEEE (such as posits) to be considered with minor adaptations. Best performance will be realized from the hardware ZFP unit when batches of blocks are processed at a time. The implementation supports 1-D, 2-D and 3-D blocks of floating-point numbers. C++ template parameters are used to specify the bit width of floating-point numbers and the array dimension of the encoder.

Getting Started

• Build and install the SystemC library with c++ 11 support
• Clone or copy the zhw repository to your system
• Set your build toolchain to clang-12 or higher

Building Using CMake

System configuration can be delicate, so we provide an example recipe to build ZHW on a clean Ubuntu 20.04 install. To build on other operating systems, one might adapt the example to account for the idiosyncracies of a particular software stack.

sudo apt install build-essential
sudo apt install cmake
sudo apt install clang-12 --install-suggests
sudo apt install wget
cd ${HOME}
wget https://www.accellera.org/images/downloads/standards/systemc/systemc-2.3.3.tar.gz
mkdir systemc
mv systemc*tar.gz systemc
cd systemc
tar -xvf systemc*tar.gz
mkdir build
cd build
cmake ../systemc-2.3.3 -DCMAKE_INSTALL_PREFIX:PATH=${HOME}/local -DCMAKE_INSTALL_DOCDIR:PATH=${HOME}/local/doc/systemc -DCMAKE_CXX_STANDARD:STRING=11
make
make install
export CC=/usr/bin/clang-12
export CPP=/usr/bin/clang-cpp-12
export CXX=/usr/bin/clang++-12
export LD=/usr/bin/ld.lld-12
cd ${HOME}
mkdir zhw_proj
cd zhw_proj
git clone https://github.com/LLNL/zhw.git
mkdir build
cd build
export PATH=$PATH:${HOME}/local
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${HOME}/local/lib
cmake -G"Unix Makefiles" -DCMAKE_BUILD_TYPE=Debug ../zhw
make

Testing

Test bench programs have been created with several test cases from the SDRbench data set. ZHW configuration can be modified in these test benches via #define settings, or command line parameters. see test/encode_test/encode_test.cpp or test/decode_test/decode_test.cpp for compiler parameters, or, run decode_test -h / encode_test -h to review command line parameters. We also provide the parameter sweeping scripts sweepencoder.sh and sweepdecoder.sh which can be used to validate a matrix of ZHW parameters on SDRBench sample data sets.

References

1. Michael Barrow, Zhuanhao Wu, Scott Lloyd, Maya Gokhale, Hiren Patel, and Peter Lindstrom, "ZHW: A Numerical CODEC for Big Data Scientific Computation", IEEE International Conference on Field Programmable Technology, December 2022.

   • note: SystemC clang (synthesizable) compilations of the ZHW encoder and decoder pertaining to this manuscript are also available.

2. Peter Lindstrom, "Fixed-Rate Compressed Floating-Point Arrays," IEEE Transactions on Visualization and Computer Graphics, 20(12):2674-2683, December 2014. doi:10.1109/TVCG.2014.2346458.

3. ZFP on GitHub, http://github.com/LLNL/zfp

4. SDRBench on GitHub, [https://sdrbench.github.io/]

Dependencies

The open source SystemC library is needed for simulation of ZHW. http://workspace.accellera.org/downloads/standards/systemc

Contributing

Bug fixes, new features, applications, or hardware modules can be submitted as pull requests. All new contributions must be open source and not contain any proprietary code.

License

ZHW is distributed under the terms of the BSD 3-Clause license. See LICENSE and NOTICE for details. SPDX-License-Identifier: BSD-3-Clause

LLNL-CODE-811758