Use arithmetic right shift for sign extension in MMX and SSE2 paths
authorFanael Linithien <fanael4@gmail.com>
Mon, 7 Dec 2020 13:46:29 +0000 (14:46 +0100)
committerJoost VandeVondele <Joost.VandeVondele@gmail.com>
Sat, 12 Dec 2020 08:20:15 +0000 (09:20 +0100)
commitc7f0a768cb9d5972861baae0f215d69f9e86a626
tree39796fca52fae5bce40f041e1835fa794a180a9e
parent8630d03dd4e1ec7e492ddf6c40d9d9e4cdba2a58
Use arithmetic right shift for sign extension in MMX and SSE2 paths

This appears to be slightly faster than using a comparison against zero
to compute the high bits, on both old (like Pentium III) and new (like
Zen 2) hardware.

closes https://github.com/official-stockfish/Stockfish/pull/3254

No functional change.
src/nnue/layers/affine_transform.h