Compiler-Based Graph Representations for Deep Learning Models of Code

Alexander Brauckmann, Andrés Goens, Sebastian Ertel, Jeronimo Castrillon

Research output: Chapter in Book/Report/Conference proceedingConference contribution


In natural language processing, novel methods in deep learning, like recurrent neural networks (RNNs) on sequences of words, have been very successful. In contrast to natural languages, programming languages usually have a well-defined structure. With this structure compilers can reason about programs, using graphs such as abstract syntax trees (ASTs) or control-data flow graphs (CDFGs). In this paper, we argue that we should use these graph structures instead of sequences for learning compiler optimization tasks. To this end, we use graph neural networks (GNNs) for learning predictive compiler tasks on two representations based on ASTs and CDFGs. Experiments show that this improves upon the state-of-the-art in the task of heterogeneous OpenCL mapping, while providing orders of magnitude faster inference times, crucial for compiler optimizations. When testing on benchmark suites not included for training, our AST-based model significantly outperforms the state-of-the-art by over 12 percentage points in terms of accuracy. It is the only one to perform clearly better than a random mapping. On the task of predicting thread coarsening factors, we show that all of the methods fail to produce an overall speedup.
Original languageEnglish
Title of host publicationProceedings of the 29th International Conference on Compiler Construction
EditorsLouis-Noël Pouchet, Alexandra Jimborean
Place of PublicationNew York, NY, USA
PublisherACM Association for Computing Machinery
Number of pages11
ISBN (Print)9781450371209
Publication statusPublished - 22 Feb 2020
Event 29th International Conference on Compiler Construction - San Diego, United States
Duration: 22 Feb 202026 Feb 2020
Conference number: 29

Publication series

NameCC 2020
PublisherAssociation for Computing Machinery


Conference 29th International Conference on Compiler Construction
Abbreviated titleCC 2020
Country/TerritoryUnited States
CitySan Diego
Internet address


  • LLVM
  • Deep Learning
  • Graphs
  • Compilers


Dive into the research topics of 'Compiler-Based Graph Representations for Deep Learning Models of Code'. Together they form a unique fingerprint.

Cite this