This is the source code for the prototype implementation of the technique presented in Eliminating Abstraction Overhead of Java Stream Pipelines using Ahead-of-Time Program Optimization.
The project uses maven for dependency management. It is compiled with
mvn compileUnit tests can be run with
mvn test
Most of the tests should run successfully, but it is expected that some 16-17 tests will fail.
The misc/jmh.sh script will optimise the micro benchmarks specified in dk.casa.streamliner.asm.TransformASM , and use JMH to benchmark the programs.
Clone and build the benchmarked projects with the RQ2/clone_and_build.py script.
Run the analysis and optimisation on the projects with the command:
env MAVEN_OPTS="-Xmx20g -Xss512m --add-opens java.base/java.lang.module=ALL-UNNAMED" mvn compile exec:java -Dexec.mainClass=dk.casa.streamliner.asm.RQ2.Experiment
Which pre-analysis to use is chosen with a command-line argument. For instance, to use the SPARK-powered analysis, add -Dexec.args="spark" to the command. The "wala" analysis is similarly available.
Additional options can be specified for SPARK by passing them as a second argument:
-Dexec.args="spark cs-demand:true,on-fly-cg:true". A list of available options can be found here.
Be aware that these analyses contain bugs and issues that cause many type queries to fail. Some stats on query results will be shown when these analyses are used.
See dk.casa.streamliner.asm.OptimizeSkeleton for a guide on how to apply the tool in more contexts.