Add NES emulation runtime benchmark

[koute]

This PR adds a new real-world runtime benchmark to the suite: Pinky, my NES emulator.

It uses a simple test ROM I've found here and emulates its execution for a given number of frames (which can be freely adjusted to make the benchmark shorter/longer).

[Kobzol]

Great! I have wanted to include some kind of interpreter in the suite, an emulator should do nicely.

I wanted to avoid the git dependency, but then I noticed that we already have a few of them in compile benchmarks, so it's probably not such a big deal.

I wonder if in the benchmark loop, the framebuffer has to be written out? Couldn't we just benchmark the emulation of instructions itself?

[koute]

I wonder if in the benchmark loop, the framebuffer has to be written out? Couldn't we just benchmark the emulation of instructions itself?

I added it because technically that is the output of the emulator, but practically speaking it probably doesn't make much difference so I can remove it if you want. (It's just a simple loop which goes over its internal palettized framebuffer, plugs each entry into a lookup table, and writes out the corresponding ABGR color.)

There's also an audio callback that I've omitted, but I'm not sure whether the compiler is smart enough to delete the final parts of the audio processing pipeline if that's empty. Maybe we should black_box both the video and audio outputs just to be sure?

[Kobzol]

I tried it without the framebuffer and the instruction count stays pretty much the same (~3.4 billion instructions). So I would just remove the framebuffer completely, and return nes from the benchmarking closure instead of the buffer. The returned value is black_boxed automatically by the benchmarking infrastructure.

Other than that one nit, looks good to me! I'll also let @nnethercote take a look.

[Kobzol]

Oh, one more thing, it would be nice if you included some short comment either to the top of the main.rs file, or above the benchmark function, to shortly describe what is the benchmark doing.

We should have a list of runtime benchmark descriptions in some README.md (same as for the compilation time benchmarks), but that's not prepared yet.

[nnethercote]

I think this is a great choice for a benchmark.

[koute]

Sorry for the delay!

I simplified the benchmark and added more comments as requested.

[Kobzol]

Thanks, looks good to me!

I'll let Nick react to his remarks, but I think that this is good to go now.

[nnethercote]

Lovely, thank you!