akrolik/rndn

Optimizing GPU compiler and database system for NVIDIA hardware

rNdN

rNdN is a runtime-optimized GPU compiler and database system, focusing on end-to-end performance of short-running queries.

Design

Compiler

Assembler

Runtime

Installation

rNdN is distributed both as source code (this repository) and pre-built Docker images that accompany conference paper artefacts. Instructions for downloading and running the Docker images are included in the paper appendices.

Requirements

1. A recent consumer-grade NVIDIA GPU is required for the complete compiler and assembler:
   • Pascal 10-series (sm_61)
   • Ampere 30-series (sm_86)
   • Assembler support for Maxwell, Turing and Volta requires an updated scheduler profile. Otherwise, --backend=ptxas must be used.
2. CUDA >= 11.3 and associated graphics driver

Note: Ubuntu 18.04 and 20.04 have been extensively tested. Other platforms may require additional setup or code changes.

Build requirements

1. Compiler supporting C++17
2. flex and bison tools
3. LLVM >= 14.0 (both llvm-14 and llvm-14-dev)
4. CMake >= 3.21
5. libpcre2-8 (distributed as part of libpcre2-dev)

Data requirements

For evaluating the TPC-H benchmark suite version 2, data must be generated using the dbgen tool. The queries are included as part of this repository.

Building

# Clone repository
git clone <repository path>
cd rNdN

# Setup build environment
mkdir build
cd build
cmake ..

# Build the project
make -j

Usage examples

rNdN executes queries and programs written in HorseIR, and accepts configuration options for debugging, optimization, and data.

./rNdN [OPTION...] <filename>

Full list of options

./rNdN --help

TPC-H query data loading

./rNdN --data-scale-tpch=<data scale> --data-load-tpch=<data path> <filename>

Other useful options

• --debug-time: Profile compilation and execution
• --debug-print: Output detailed debug logging information
• --debug-options: Output current options, including defaults

Citations

Alexander Krolik, Clark Verbrugge, and Laurie Hendren. r3d3: Optimized query compilation on GPUs. In 2021 IEEE/ACM International Symposium on Code Generation and Optimization (CGO), pages 277–288, 2021. doi:10.1109/CGO51591.2021.9370323.

@inproceedings{Krolik2021:r3d3,
    author    = {Krolik, Alexander and Verbrugge, Clark and Hendren, Laurie},
    booktitle = {2021 IEEE/ACM International Symposium on Code Generation and Optimization (CGO)},
    title     = {r3d3: Optimized Query Compilation on GPUs},
    year      = {2021},
    volume    = {},
    number    = {},
    pages     = {277-288},
    doi       = {10.1109/CGO51591.2021.9370323}
}

Alexander Krolik. rNdN: Optimized query compilation for GPUs. PhD thesis, McGill University, May 2022. URL: TODO.

@phdthesis{Krolik2022:Thesis,
    author = {Krolik, Alexander}, 
    title  = {{rNdN}: Optimized Query Compilation for GPUs},
    school = {McGill University},
    year   = {2022},
    month  = {May},
    url    = {}
}

Acknowledgements

This work was completed at the School of Computer Science, McGill University under supervision of Professor Clark Verbrugge and Professor Laurie Hendren.

We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC).

Nous remercions le Conseil de recherches en sciences naturelles et en génie du Canada (CRSNG) de son soutien.