Add criterion benchmarks

[alexheretic]

I was curious to quantify the performance of SmolStr inline strings vs naive Strings. I wrote some criterion benchmarks for lengths 12, 50, 1000 to see inline vs heap perf differences.

I would expect inline SmolStrs to generally perform better than Strings, and heap SmolStrs to perform a little worse. However, the numbers seem generally worse almost everywhere except SmolStr::clone. Can anyone comment on this and help me understand why this would be?

Update: It seems the microbenches can be an easy place for the allocator to optimise, so I'll take a look elsewhere at improving a bench comparing allocating with non-allocating. Here we can just have some benches for this crate. I've stripped the benches to only being for this crate here.

Results

format_smolstr! len=12  time:   [27.925 ns 27.977 ns 27.990 ns]

format_smolstr! len=50  time:   [66.376 ns 66.394 ns 66.399 ns]

format_smolstr! len=1000
                        time:   [95.812 ns 95.867 ns 95.880 ns]

SmolStr::from len=12    time:   [14.482 ns 14.483 ns 14.486 ns]

SmolStr::from len=50    time:   [9.7307 ns 9.7373 ns 9.7390 ns]

SmolStr::from len=1000  time:   [16.582 ns 16.582 ns 16.582 ns]

SmolStr::clone len=12   time:   [4.1175 ns 4.1258 ns 4.1278 ns]

SmolStr::clone len=50   time:   [2.8261 ns 2.8313 ns 2.8523 ns]

SmolStr::clone len=1000 time:   [2.8571 ns 2.8640 ns 2.8916 ns]

SmolStr::eq len=12      time:   [2.3383 ns 2.3391 ns 2.3423 ns]

SmolStr::eq len=50      time:   [2.5612 ns 2.5624 ns 2.5668 ns]

SmolStr::eq len=1000    time:   [9.1096 ns 9.1241 ns 9.1820 ns]

to_lowercase_smolstr len=12
                        time:   [66.165 ns 66.192 ns 66.300 ns]

to_lowercase_smolstr len=50
                        time:   [347.27 ns 348.20 ns 348.43 ns]

to_lowercase_smolstr len=1000
                        time:   [5.6151 µs 5.6191 µs 5.6201 µs]

to_ascii_lowercase_smolstr len=12
                        time:   [13.539 ns 13.587 ns 13.599 ns]

to_ascii_lowercase_smolstr len=50
                        time:   [99.236 ns 99.382 ns 99.966 ns]

to_ascii_lowercase_smolstr len=1000
                        time:   [1.0966 µs 1.0988 µs 1.0993 µs]

replace_smolstr len=12  time:   [23.562 ns 23.815 ns 23.878 ns]

replace_smolstr len=50  time:   [81.027 ns 81.274 ns 81.336 ns]

replace_smolstr len=1000
                        time:   [640.14 ns 640.54 ns 642.17 ns]

[alexheretic]

I'll probably strip the std-string benches out of here so these are just for this crate fns.

[alexheretic]

FYI I moved my comparison with std to https://github.com/alexheretic/rust-string-bench.

[Veykril]

Thanks!