Added option to select GPU to CUDA benchmark.

[mattkjames7]

I had a bit of fun with this nifty little utility, I had to make a small modification to allow me to benchmark multiple GPUs on the same system. It's a really minor change, just added an optional --gpu argument e.g.

./mixbench-cuda --gpu 2

It still defaults to GPU 0 if no argument is passed.

[ekondis]

Thank you for this. Regarding the --gpu argument, I'm just a little hesitant from the fact that this diverges from the way the device is specified in the OpenCL & SYCL implementations. In the latter, the device index is passed directly without a named argument. I understand that using a named argument is a better practice but in this case I would prefer without it.

[mattkjames7]

So you would prefer a call something more like this?

./mixbench-cuda 2

I'll make an edit when I get a chance!

[ekondis]

So you would prefer a call something more like this?

./mixbench-cuda 2

I'll make an edit when I get a chance!

Yes, exactly. To maintain a somewhat similar invocation pattern between all benchmarks.