Speed up tail processing when hashing aligned C strings
authorJohn Naylor <john.naylor@postgresql.org>
Sun, 31 Mar 2024 05:19:16 +0000 (12:19 +0700)
committerJohn Naylor <john.naylor@postgresql.org>
Sun, 31 Mar 2024 05:19:16 +0000 (12:19 +0700)
commit07f0f6abfc7f6c55cede528d9689dedecefc734a
tree6e6fb87400b6350c0328509384041c78b0ad3f42
parentb1484a3f1910bfd0e254afe40085dfc3351bda8c
Speed up tail processing when hashing aligned C strings

After encountering the NUL terminator, the word-at-a-time loop exits
and we must hash the remaining bytes. Previously we calculated the
terminator's position and re-loaded the remaining bytes from the input
string. We already have all the data we need in a register, so let's
just mask off the bytes we need and hash them immediately. The mask can
be cheaply computed without knowing the terminator's position. We still
need that position for the length calculation, but the CPU can now
do that in parallel with other work, shortening the dependency chain.

Ants Aasma and John Naylor

Discussion: https://postgr.es/m/CANwKhkP7pCiW_5fAswLhs71-JKGEz1c1%2BPC0a_w1fwY4iGMqUA%40mail.gmail.com
src/include/common/hashfn_unstable.h