These are some performance comparisons casting of types build with type_hierarchy (TH) vs classical RTTI based
dynamic_cast using ni::match.
l0, l1, l2 referes to objects of types on the respective inheritence level, i.e. root, 1st inherited level,
and 2nd inherited level, repsecitvely. Times have been measured on a Macbook Pro 13" i5 2 core (Apple clang 9.0)
and a Windows machine using MSVC 2017.
| cast | RTTI (clang) | TH (clang) | TH speed-up (clang) | RTTI (msvc) | TH (msvc) | speed-up (msvc) |
|---|---|---|---|---|---|---|
| l0 -> l1 succ. | 13 ns | 1.2 ns | 11 x faster | 24 ns | 4.8 ns | 5 x faster |
| l0 -> l1 fail | 19 ns | 1.2 ns | 16 x faster | 43 ns | 4.8 ns | 9 x faster |
| l1 -> l2 succ. | 14 ns | 1.2 ns | 12 x faster | 24 ns | 4.8 ns | 5 x faster |
| l1 -> l2 fail | 19 ns | 1.2 ns | 16 x faster | 52 ns | 4.8 ns | 11 x faster |
| l0 -> l2 succ. | 15 ns | 1.2 ns | 12 x faster | 35 ns | 4.8 ns | 7 x faster |
| l0 -> l2 fail | 21 ns | 1.2 ns | 17 x faster | 52 ns | 4.8 ns | 11 x faster |
| l2 -> l1 match | 1.2 ns | 1.2 ns | 1 x faster | 4.8 ns | 4.8 ns | 1 x faster |
| l2 -> l1 fail | 16 ns | 1.2 ns | 13 x faster | 52 ns | 4.8 ns | 11 x faster |