The X87 Floating Point Unit (FPU) for the IA32 Instruction Set


The 8 floating point registers %st(0) , ... , %st(7) form a stack, with
the most recent entry in %st(0).  Entries "older" than 8 disappear.

The floating point processor has several condition codes:

PE: Precision exception
UE: Underflow
OE: Overflow
ZE: Divide-by-zero
DE: Result unnormalized
IE: Invalid operation
SF: Stack fault
ES: Error / no error
C0 - C3: comparison of two floating point numbers

Floating point instructions used in main.s and mult2.s:

faddp             Replace %st(0) by %st(0) + %st(1) and both addends are gone.
flds Src          Push Mem[Src] onto the stack.
flsz              Push 0 onto the stack.
fmuls Src         Replace %st(0) by  %st(0) * M[Src].
fnstsw Src        Save the floating point flags in Src.
fstpl Src         Pop %st(1) to M[Src] (64 bits)
fstps Src         Pop %st(0) to M[Src] (32 bits)
fucompp           Compare %st(0) and %st(1) to set condition codes,
                      and don't report error if one is NaN.
fxch %st(i)       Swap %st(0) and %st(i).

Other sample floating point instructions:

fadds Src         Replace %st(0) by %st(0) + M[Src].

There are also floating point sin, cos, and sqrt instructions.

Other IA32 instructions used in those files:

leave             Abbreviation for "movl %ebp, %esp" and "pop %ebp"
                      which sets stack pointer to beginning of frame
                      and restores the saved %edp from the top of the stack

sahf              Store flags.
seta  Src         Src = ~CF & ~ZF.
sete  Src         Src = ZF.

testb Src,Dest    Set condition codes for Src & Dest (low-order byte)

Source:
Assembly Language Programming In Gnu/Linux For Ia32 Architectures
by Rajat Moona

X86-64:  In the X86-64 extension to IA32: 

   -- There are still 8 registers in the floating point stack.

   -- There is an additional set of 16 128-bit registers that can be used for
      SIMD instructions.
      http://en.wikipedia.org/wiki/Streaming_SIMD_Extensions