Java/C Comparative Benchmarks

Description

Benchmarks include:

• Bit twiddling
• Numerical computing
• In-memory data structure creation and walking
• Concurrent computing
• Calling out from Java/Python/Jython to native libraries
• Utilization of graphics processing units (GPU) from Java and C

Performance is measured in two ways. In some tests it is measured as the time to complete a task. In some tests it is measured as the rate for doing work -- such as number of floating operations per second. Some results are reported both ways. Most of the test programs report the results to the console as well as to a local results file.

The subdirectories (each with one or more tests) are :

• GPU utilization. Uses a translation to Java of Apple's GPU/OpenCL FFT's).
  Also uses Olivier Chafik's OpenCL for Java, JavaCL.

  • FFT-J-OpenCL-BridJ-Apple/

• Non GPU, simple FFTs

  • FFT-J/
  • FFT-C/
  • FFT-Scala/
  • FFT-Testing-Common/
  • FFTW-C/

• Bit-twiddling

  • JLRSGenTest/
  • CLRSGenTest/
  • SLRSGenTest/

• Memory structure creation with random traversal using red-black trees

  • RbTree-C-Test/
  • RbTree-J-Test/
  • RbTree-S-Test/
  • RbTree-Py-Test/

• Matrix multiply

  • MatMul-C
  • MatMul-J
  • MatMul-S
  • MatMul-Py

• Concurrent computing

  • Parallelism-Benchmarks-C/
  • Parallelism-Benchmarks-J/

• Method invocation and callout

  • MethodCalls-C
  • MethodCalls-Jy (using Java JNI callout)
  • MethodCalls-Py (using standard CPython extension mechanism)
  • MethodCalls-S
  • MethodCalls-J (uses Olivier Chafik's BridJ and JNAerator)

Platforms

The tests were developed and tested to run on Mac OS X and Linux.