LonestarGPU Benchmark Suite v2

The LonestarGPU (LSG) suite contains CUDA implementations of several
irregular algorithms that exhibit amorphous data parallelism.

Requirements:
        CUDA 5.5
        Fermi or later GPU hardware
        CUB v1.1.1

The instructions below assume LSG has been installed in $LSGDIR.

Each app directory under $LSGDIR/apps/$APP contains a README that
explains what $APP does, how to run it, details of any variant
implementations and other useful info.


INSTALLATION

You will need to download and install CUB from here:

http://nvlabs.github.io/cub/

Place a symlink to the top-level CUB directory in $LSGDIR. Assuming
the top-level CUB directory is $CUBDIR:

$ cd $LSGDIR
$ ln -s $CUBDIR


BUILDING

Assuming you're in $LSGDIR:

$ make inputs # downloads the inputs required for LSG

$ make # compiles all applications and variants

Optionally, `make optional-variants' under each app directory compiles
even more additional variants which may require additional
downloads. See the README for each app for details.


RUNNING

Each directory under $LSGDIR/apps contains a simple `run' script that
runs the application with all recommended inputs.


MULTIPLE GPUS

The LSG benchmarks will always use device 0. If your system has
multiple GPUs, use the CUDA_VISIBLE_DEVICES environment variable to
control which GPU the benchmark uses.

$ CUDA_VISIBLE_DEVICES=1 ./run