grdfs is an RDFS reasoner that uses OpenCL for fine-grained parallelism. It can thus exploit massively parallel hardware such as modern GPUs but also runs well on CPUs. Currently it supports the following features:

• OpenCL implementation of most used RDFS entailment rules.
• In-memory storage of entailed triples.
• Writing all or only entailed triples to stdout.
• Including RDFS axiomatic triples
• On-device removal of global and local duplicates (local duplicate removal currently only works on GPU devices).

• grdfs uses the turtle parser from RDF3-X, it thus is used as follows:
• grdfs -i <turtle_file>
• By default, grdfs will use the GPU and print some info to stdout.

• -i or --input-file <file.ttl> -- source turtle file (required)
• --device gpu|cpu -- OpenCL device to use (gpu, cpu)
• -l or --no-local-dedup -- disable local deduplication
• -g or --no-global-dedup -- disable global deduplication
• -a or --axioms -- include RDFS axiomatic triples
• -t or --time -- print profiling information
• -p or --print-triples -- write triples to stdout (in NTriples format)

• For building grdfs a C++11 compiler is needed. We recommend one of the following:
  • Clang 3.1
  • GCC 4.7
• On Linux, the AMD APP SDK must be installed
• Boost program_options
  • Mac OS X: brew install boost --with-c++11
  • Ubuntu: sudo apt-get install libboost-dev libboost-program-options-dev

• Mac OS X: use provided Xcode project or type make.
• Linux: type make.

• Complete rule implementation
• Using multiple OpenCL devices
• Partitioning large graphs to fit in device memory