Scaffolding for direct benchmarking of crate::filter::unfilter.

[anforowicz]

Can you PTAL? The new benchmarks from this PR should help with work on performance improvements of fn unfilter.

The real motivation for this PR is that it is a prelude to a series of commits at bf2c26b...1c78a3f that results in a significant decoding speed-up (measured on Intel). Special thanks to @veluca93 for help with various SIMD questions that I had. Results that I got from running the benchmarks:

• decode/kodim02.png: -26.6% time
• decode/kodim17.png: -25.7% time
• decode/kodim07.png: -27.1% time
• decode/kodim23.png: -21.8% time
• unfilter/filter=Sub/bpp=3: -71.5% time
• unfilter/filter=Sub/bpp=6: -86.1% time
• unfilter/filter=Avg/bpp=3: -71.8% time
• unfilter/filter=Avg/bpp=6: -70.6% time
• unfilter/filter=Paeth/bpp=3: -51.4% time
• unfilter/filter=Paeth/bpp=6: -62.9% time
• AFAICT other decode/... and unfilter/... benchmarks are unchanged

I note that the speed-up above depends on std::simd which hasn't yet been stabilized, so I thought that it would be best to have a separate PR for just the benchmarks. (i.e. the new benchmarks seem desirable on their own, even if we decide not to pursue the std::simd-based improvements)

/cc @calebzulawski as FYI, since the improvements are possible thanks to std::simd and rust-lang/rust#86656.

[calebzulawski]

That looks great! Feel free to file any issues or let me know if you see any cases where performance is worse than you'd expect.

[anforowicz]

FWIW, I assume that the mips_cross CI failure is unrelated to this PR. Please shout if you think this is not the case.

[calebzulawski]

I don't think it's related (but we ran into similar). I believe mips was demoted to tier 3, so might not have up to date toolchains

[fintelia]

Yeah, the mips failure is unrelated. In the main image crate we switched to powerpc for big endian testing (image-rs/image#1987), so probably just need to do the same thing here.

I should hopefully be able to take a look at the rest of the PR later tonight

[fintelia]

Could you call this from_usize or similar?

[anforowicz]

Done. I assume that you plan to squash the last 2 small commits (cargo fmt + this rename) when merging?

[fintelia]

Really excited to see the progress to enable using SIMD from safe Rust code! Avoiding, or ideally forbidding, unsafe code is one of the big differentiators that we can provide over the C libraries that are generally used for image decoding. But that's often meant that we either have to forgo potential performance improvements or tolerate some unsafe in order to use SIMD