Benchmark for BoundLocaleDataProvider

[sffc]

Issue: #3260

Merge of #7662 and #7621 with a benchmark.

$ ICU4X_REGION_POSTCARD_PATH=../../_debug/regionnames.postcard cargo bench --all-features region

...

locale_names/region/postcard/all/1by1
                        time:   [6.8448 ms 6.9295 ms 7.0177 ms]
Found 6 outliers among 100 measurements (6.00%)
  6 (6.00%) high mild

locale_names/region/postcard/all/boundlocale
                        time:   [3.6443 ms 3.6895 ms 3.7372 ms]
Found 34 outliers among 100 measurements (34.00%)
  15 (15.00%) low severe
  1 (1.00%) low mild
  3 (3.00%) high mild
  15 (15.00%) high severe

I built regionnames.postcard like this: cargo run -p icu4x-datagen --all-features -- -m LocaleNamesRegionLongV1 CaseMapV1 PluralsCardinalV1 TimezoneNamesLocationsOverrideV1 --format blob -o _debug/regionnames.postcard

I included the target marker as well as a small handful of others to simulate walking through the marker trie.

[sffc]

I added a bench on the old ZeroMap struct:

locale_names/region/baked/all/1by1
                        time:   [5.9787 ms 6.0411 ms 6.1111 ms]
Found 5 outliers among 100 measurements (5.00%)
  5 (5.00%) high mild

locale_names/region/baked/all/old_map_struct
                        time:   [2.3787 ms 2.3947 ms 2.4130 ms]
Found 18 outliers among 100 measurements (18.00%)
  1 (1.00%) low mild
  1 (1.00%) high mild
  16 (16.00%) high severe

locale_names/region/postcard/all/1by1
                        time:   [6.9537 ms 7.0637 ms 7.1818 ms]
Found 14 outliers among 100 measurements (14.00%)
  13 (13.00%) high mild
  1 (1.00%) high severe

locale_names/region/postcard/all/boundlocale
                        time:   [3.3650 ms 3.3756 ms 3.3897 ms]
Found 11 outliers among 100 measurements (11.00%)
  2 (2.00%) high mild
  9 (9.00%) high severe

[robertbastian]

Can you add a benchmark for loading like 10 names? I don't think loading 300+ names at once is the most realistic use case.

[sffc]

Can you add a benchmark for loading like 10 names? I don't think loading 300+ names at once is the most realistic use case.

It is a realistic use case, because pickers have a lot of regions/languages, but I added one to load 10% of the names.

locale_names/region/baked/10x10pct/1by1
                        time:   [72.051 µs 72.475 µs 72.879 µs]
Found 17 outliers among 100 measurements (17.00%)
  5 (5.00%) low severe
  4 (4.00%) high mild
  8 (8.00%) high severe

locale_names/region/baked/10x10pct/old_map_struct
                        time:   [24.617 µs 24.832 µs 25.026 µs]

locale_names/region/postcard/10x10pct/1by1
                        time:   [68.274 µs 69.017 µs 69.688 µs]
Found 23 outliers among 100 measurements (23.00%)
  1 (1.00%) high mild
  22 (22.00%) high severe

locale_names/region/postcard/10x10pct/boundlocale
                        time:   [34.649 µs 34.948 µs 35.321 µs]
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high mild

[sffc]

I want to point out that the benchmark performance is more dependent on the shape of the data build than the exact workload. I built a postcard file containing display names data for all regions and all locales. If I were to use a postcard file containing display names for only one locale, then the diff is going to be smaller.