Skip to content
No description, website, or topics provided.
Java C R Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.settings do not automatically reformat by default Dec 6, 2013
lib make fastr run with latest truffle api, added (vanilla) truffle api s… Sep 13, 2013
native installation instructions Jan 31, 2014
src/r recover executable nodes after a dynamic call through three dots Feb 3, 2014
test/r recover executable nodes after a dynamic call through three dots Feb 3, 2014
.checkstyle checkstyle Mar 18, 2013
.checkstyle_checks.xml Big refactoring of builtins Mar 9, 2013
.classpath make fastr run with latest truffle api, added (vanilla) truffle api s… Sep 13, 2013
.gitignore
.project checkstyle Mar 18, 2013
LICENSE.txt License Oct 2, 2013
README.md link to bitbucket/oracle fastr Jun 17, 2015
build.xml eclipse classpath and ant build patched for new truffle jar. Jun 28, 2013
nr.sh
pom.xml HotSpot specific hack to ensure java.library.path has been reloaded Oct 17, 2013
r.sh r.sh patched for the new truffle api Jun 28, 2013

README.md

FastR

FastR implements the R Language. Currently, FastR can run the R implementation of the Language Shootout Benchmarks and the Benchmark 25 suite.

This version of FastR has been published at VEE'15: "A fast abstract syntax tree interpreter for R". The project now continues here.

Quick Start

  1. download the latest code: wget https://github.com/allr/fastr/archive/master.zip
  2. unzip it: unzip master.zip
  3. build: cd fastr-master ; ant
  4. run the console: ./r.sh
  5. run the binarytrees benchmark for size 5: ./r.sh --args 5 -f test/r/shootout/binarytrees/binarytrees.r

Minimal Requirements

To run the binarytrees benchmark as shown above, FastR requires Java. All Shootout benchmarks can be run this way, but some of the mandelbrot only on Unix, as they spawn the cat process.

Full Installation

To run the benchmarks from the Benchmark 25 suite, and for best performance of all benchmarks, build native glue code which links FastR to the GNU-R Math Library, system Math library, and openBLAS. The build scripts are tested on Ubuntu 13.10. Any platform supported by GNU-R and Java could be supported by FastR.

  1. install Oracle JDK8 (for best performance); if you must use JDK7, customize native/netlib-java/build.sh
  2. set JAVA_HOME and PATH accordingly
  3. follow the steps in Quick Start
  4. install Ubuntu packages r-base, r-mathlib, libopenblas-base
  5. build glue code for system libraries and GNU-R: cd native ; ./build.sh
  6. build glue code for native BLAS and LAPACK: cd netlib-java ; ./build.sh
  7. check the glue code can be loaded: cd ../.. ; ./nr.sh should give output
    Using LAPACK: org.netlib.lapack.NativeLAPACK
    Using BLAS: org.netlib.blas.NativeBLAS
    Using GNUR: yes
    Using System libraries (C/M): yes
    Using MKL: not available
  8. run the matfunc-1 benchmark: ./nr.sh -f test/r/benchmark25/perfres/b25-matfunc-1.r

To ensure that the openBLAS library is used, run the matcal-4 benchmark with the system profiler: perf record ./nr.sh -f test/r/benchmark25/perfres/b25-matcal-4.r.

Check with perf report that DGEMM from openBLAS is used, e.g. dgemm_kernel_SANDYBRIDGE from libopenblas.so.0. Also expect to see the random number generator, e.g. qnorm5 from libRmath.so.1.0.0.

Running Tests

ant tests

The outputs will appear in junit directory.

Eclipse

The code contains settings and project configuration for Eclipse Juno.

Authors:

Tomas Kalibera, Petr Maj, Floreal Morandat and Jan Vitek

You can’t perform that action at this time.