Vol. 1 8-19

PROGRAMMING WITH THE X87 FPU

Reverse versions of the subtract (FSUBR) and divide (FDIVR) instructions enable efficient coding. For example, the 
following options are available with the FSUB and FSUBR instructions for operating on values in a specified x87 FPU 
data register ST(i) and the ST(0) register:
FSUB:

ST(0) ← ST(0) − ST(i)

ST(i) ← ST(i) − ST(0)

FSUBR:

ST(0) ← ST(i) − ST(0)

ST(i) ← ST(0) − ST(i)

These instructions eliminate the need to exchange values between the ST(0) register and another x87 FPU register 
to perform a subtraction or division.
The pop versions of the add, subtract, multiply, and divide instructions offer the option of popping the x87 FPU 
register stack following the arithmetic operation. These instructions operate on values in the ST(i) and ST(0) regis-
ters, store the result in the ST(i) register, and pop the ST(0) register.
The FPREM instruction computes the remainder from the division of two operands in the manner used by the Intel 
8087 and Intel 287 math coprocessors; the FPREM1 instruction computes the remainder in the manner specified in 
IEEE Standard 754.
The FSQRT instruction computes the square root of the source operand.
The FRNDINT instruction returns a floating-point value that is the integral value closest to the source value in the 
direction of the rounding mode specified in the RC field of the x87 FPU control word.
The FABS, FCHS, and FXTRACT instructions perform convenient arithmetic operations. The FABS instruction 
produces the absolute value of the source operand. The FCHS instruction changes the sign of the source operand. 
The FXTRACT instruction separates the source operand into its exponent and fraction and stores each value in a 
register in floating-point format.

8.3.6 

Comparison and Classification Instructions

The following instructions compare or classify floating-point values:
FCOM/FCOMP/FCOMPPCompare floating point and set x87 FPU

condition code flags.

FUCOM/FUCOMP/FUCOMPPUnordered compare floating point and set 

x87 FPU condition code flags.

FICOM/FICOMPCompare integer and set x87 FPU 

condition code flags.

FCOMI/FCOMIPCompare floating point and set EFLAGS 

status flags.

FUCOMI/FUCOMIPUnordered compare floating point and 

set EFLAGS status flags.

FTST

Test (compare floating point with 0.0).

FXAMExamine.

Comparison of floating-point values differ from comparison of integers because floating-point values have four 
(rather than three) mutually exclusive relationships: less than, equal, greater than, and unordered.
The unordered relationship is true when at least one of the two values being compared is a NaN or in an unsup-
ported format. This additional relationship is required because, by definition, NaNs are not numbers, so they 
cannot have less than, equal, or greater than relationships with other floating-point values.
The FCOM, FCOMP, and FCOMPP instructions compare the value in register ST(0) with a floating-point source 
operand and set the condition code flags (C0, C2, and C3) in the x87 FPU status word according to the results (see 
Table 8-6).