notapaper_recode

You need CUDA and boost::random to compile this...

After compilation:

./maprecode -h

The output is Semicolon separated: *Size of Dict in KB *Throughput MB/s CPU sequential *Throughput MB/s CPU openmp parallelized *Throughput MB/s GPU transfer only *Throughput MB/s GPU streaming without initial map-transfer and without processing *Throughput MB/s GPU transfer,streaming and processing <- the Total

See http://blog.notapaper.de/article4.html