Version
3.5-1846
Branch
hardwareAA
Date
9 years, 7 months ago
Windows x64 Windows x86 Mac OS X Android
Commit
114ba7b74b8a619089c171a83e5658c56016adb9
Change author
Scott Mansell
Change description
VideoSoftware: Optimisations: Promoted all s16s to s32's in TEV

Modern x86 processors are really bad at dealing with 16 bit types. On a
Core 2 cpu, any cache line containing a length changing prefix will have
a decode penalty of 6 cycles.

On Sandy bridge or later, each lenght changing prefix will add a 3
second penalty to decoding (and will be better on average).