Next: , Previous: Top, Up: Top

1 Compiling the IS benchmark

To analyze the IS benchmark, the first thing we must do is obtain a copy of it. The easiest way to do this is to download the Berkeley UPC source code distribution, and then copy the NPB2.4 directory from the upc-tests directory, as in

     $ tar xvzf berkeley_upc-2.8.0.tar.gz
     $ cd berkeley_upc-2.8.0
     $ cp -av upc-tests/NPB2.4 ~

This should result in a new directory named NPB2.4 being placed in your home directory. Note that your system may require slightly different commands as above, especially if you are using a different version of Berkeley UPC.

Once you have a copy of the NPB2.4 benchmarks, the next thing to do is to compile them. Normally, this benchmark suite is compiled using a command such as

     $ make IS NP=32 CLASS=B

To generate performance data files for use with PPW, you must compile the benchmark with the ppwupcc command instead of the regular upcc command. The ppwupcc command is set up as a very thin wrapper to the normal upcc command, and will pass through any flags it does not understand. The idea is to substitute ppwupcc into your makefiles or build scripts, and ppwupcc will take care of the rest of bookkeeping required by PPW, such as linking against any libraries it requires and keeping snapshots of source code files for later viewing.

For the IS benchmark, if you grabbed a copy of the NPB2.4 source code from Berkeley UPC, then you simply change your compilation command to

     $ make UPCC=ppwupcc IS NP=32 CLASS=B

If you obtained the NPB2.4 source tree from somewhere else, you might have to hack up the make.def file in the config directory.