Справочник Пользователя для AMD 250

25112

Rev. 3.06

September 2005

Use of a two-byte near-return can improve performance. The single-byte near-return (opcode C3h) of 
the RET instruction should be used carefully. Specifically, avoid the following two situations:

•

Any kind of branch (either conditional or unconditional) that has the single-byte near-return RET 
instruction as its target. See “Examples.” 

•

A conditional branch that occurs in the code directly before the single-byte near-return RET 
instruction. See “Examples.” 

This optimization applies to:

•

32-bit software

•

64-bit software

The processor is unable to apply a branch prediction to the single-byte near-return form (opcode C3h) 
of the RET instruction.

The easiest way to assure the utilization of the branch prediction mechanism is to use a two-byte RET 
instruction. A two-byte RET has a REP instruction inserted before the RET, which produces the 
functional equivalent of the single-byte near-return RET instruction, but is not affected by the 
prediction limitations outlined above. To use a two-byte RET, define a text macro named 

REPRET

 and 

use it instead of the RET instruction to force the intended object code.

REPRET TEXTEQU <DB 0F3h, 0C3h>

Avoid branches in which the target of the branch is a single-byte near-return:

   jmp label   ; Jump to a single-byte near-return RET instruction.

   ...

label:

   ret         ; RET is potentially mispredicted.

Avoid branches that immediately precede a single-byte near-return:

jz  label   ; Conditional branch is not taken.

ret         ; RET is a fall-through instruction,

            ;  potentially mispredicted.