Use Ffreep Macro To Pop One Register From The Fpu Stack; Floating-Point Compare Instructions - AMD Athlon Processor x86 Optimization Manual

AMD Athlon™ Processor x86 Code Optimization

Use FFREEP Macro to Pop One Register from the FPU Stack

Floating-Point Compare Instructions

98
In FPU intensive code, frequently accessed data is often
pre-loaded at the bottom of the FPU stack before processing
floating-point data. After completion of processing, it is
desirable to remove the pre-loaded data from the FPU stack as
quickly as possible. The classical way to clean up the FPU stack
is to use either of the following instructions:
FSTP ST(0)
FCOMPP
On the AMD Athlon processor, a faster alternative is to use the
FFREEP instruction below. Note that the FFREEP instruction,
although insufficiently documented in the past, is supported by
all 32-bit x86 processors. The opcode bytes for FFREEP ST(i)
are listed in Table 22 on page 212.
FFREEP ST(0)
FFREEP ST (i) wo rks like FFREE ST(i) except tha t it
increments the FPU top-of-stack after doing the FFREE work.
In other words, FFREEP ST(i) marks ST(i) as empty, then
increments the x87 stack pointer. On the AMD Athlon
processor, the FFREEP instruction converts to an internal NOP,
which can go down any pipe with no dependencies.
Many assemblers do not support the FFREEP instruction. In
these cases, a simple text macro can be created to facilitate use
of the FFREEP ST(0).
FFREEP_ST0 TEXTEQU
For branches that are dependent on floating-point comparisons,
use the following instructions:
FCOMI
FCOMIP
FUCOMI
FUCOMIP
;removes one register from stack
;removes two registers from stack
;removes one register from stack
<DB 0DFh, 0C0h>
Use FFREEP Macro to Pop One Register from the FPU
22007E/0—November 1999