t*s
发帖数: 10 | 1 MMX provide some special instructions to fast small
integer operation. The design is mostly following
VLIW (very large instruction width). One shortcut
Intel take is use the 64 floating point register,
therefore it can execute four short integer ops
in one cpu cycle (the maximum speedup).
You should be able to find these information from
INTEL, there are some papers regarding the speedup
of MMX tech in term of MMX application. Not very
impressive, average 30% fast, some even suffer. |
|
