KGEN: Fortran Kernel Generator

A Python tool that extracts partial codes out of a large Fortran application and converts them into a standalone/verifiable/executable kernel

Send questions and comments to KGEN Dev. Team ([email protected]).

Documentation

https://ncar.github.io/kgendocs

Discussion-group

https://groups.google.com/forum/#!forum/kgen-discuss

Changes from KGEN ver. 0.7.2

[ User Interface ]

• "--invocation" option is changed from mandatory to optional
• "--repr-etime", "--repr-papi", and "--repr-code" options are added for representative extensions
• "--state-clean", "--state-build", and "--state-run" options are discarded.

[ Major Improvements ]

KGen measures three types of characteristics from original application and generates kernel and input data in a way to reprouce the types of characteristics in generated kernel.

Overview

• KGEN extracts an arbitrary region of Fortran source code as a stand-alone executable software
• In addition, it generates input & output data for executing and verifying the generated kernel
• All KGEN-generated kernels include correctness check and timing measurement

Dependencies

• Linux OS
• Python (>=2.7 and < 3.0)
• C Pre-Processor(cpp)
• Make build tool(make)
• System call trace tool(strace)
• Stream Editor (sed)

Obtaining the Source Code

The latest KGEN can be obtained from the Git repository.

git clone https://github.com/NCAR/KGen.git

Building & Installation

Current KGEN does not require to build or to install.

Instructions & Use

1. Download from KGen Github repository.

   >> git clone https://github.com/NCAR/KGen.git

2. Try a kernel generation example in "examples/simple" directory

   >> cd examples/simple; # move to an example directory

   >> vi src/Makefile; # Modify FC if required

   >> make; # extract a kernel

   >> cd kernel; # move to a kernel directory

   >> make; # build and run a kernel