This patch replaces the use of std::uint16_t with std::uint32_t. This increases performance of emplace/insert by up to 15%, see attached benchmark results.
The exact reason has not be determined but this is likely due to std::uint16_t tending to require more CPU cycles for math operations. It may also be related to less efficient optimization when using std::uint16_t types in the with vectorization.
2022-04-avoid-uin16.txt
This patch replaces the use of
std::uint16_twithstd::uint32_t. This increases performance of emplace/insert by up to 15%, see attached benchmark results. The exact reason has not be determined but this is likely due tostd::uint16_ttending to require more CPU cycles for math operations. It may also be related to less efficient optimization when usingstd::uint16_ttypes in the with vectorization. 2022-04-avoid-uin16.txt