Gapbs Versions Save

• In-place in-memory graph builder to reduce peak memory usage, use -m (thanks @connormas)
• PR - faster with Gauss-Seidel method and better scheduling directives (thanks @connormas)
• BFS - faster for high-diameter graphs with better scheduling directives
• TC - faster with encouraging use of gcc's parallel sort for relabeling
• Bugfix for loading in .mtx files (thanks Ramesh Peri, Intel)
• Switch to GitHub Actions for CI

• Improved SSSP performance on high-diameter graphs with bucket fusion (thanks @yunmingzhang17)
• Improved BC & CC performance with streamlined/tuned synchronization
• GAP Benchmark graphs added to SuiteSparse (thanks @DrTimothyAldenDavis)
• Bugfixes for graph loading .mtx and .gr (thanks @DrTimothyAldenDavis)
• Bugfix for BC on high-diameter graphs (thanks Tze Meng Low, CMU)
• Bugfixes to printing and testing parallelism (thanks @Kacpro @idanyani)

New CC implementation and increased portability:

• For timing, now uses C++11's std::chrono instead of gettimeofday (thanks @michaelsutton)
• CC changed to new blazing fast Afforest algorithm (thanks to @michaelsutton for algorithm and implementation)
• Prior CC implementation (Shiloach-Vishkin) retained, but no longer default CC implementation

More robustness and performance enhancements:

• Support for running suite with older version of gunzip
• Prints warning if number of nodes will overflow type used for NodeID
• Improved CC performance and bugfix (thanks @michaelsutton)
• Installation automation with spack (thanks @ledif)
• Added CONTRIBUTING file, will welcome high performance kernel implementations, especially for non-benchmark kernels

After nearly 2 years in beta, we are officially releasing gapbs 1.0! Since the last release we have added:

• Automation for building benchmark graphs and executing benchmark
• Revised benchmark specification
• Support for floating point weights for SSSP
• Small bugfixes (thanks to @ArnauPrat & @dunkeki)

Small refinements and cleanups:

• Add vertex iterator to simplify serial code (verifiers and stats)
• More robust controller for direction-optimizing BFS
• Miscellaneous style fixes and warnings removed

A few weeks late, but more compatibility and robustness:

• Added support for Metis input graph format
• Added support for Matrix Market input graph format
• Added support for sunCC and verified on SPARC/Solaris
• Simplified SSSP implementation

• Added verifiers for kernels (use -v)
• Added testing for whole codebase (make test)
• Added continuous integration via travis
• Added reference to benchmark specification

Small improvements including:

• Better style compliance
• More comments
• Improvements to TC, BC, PR

Preview release for ISCA 2015