Inside Macintosh: PowerPC Numerics / Part 3 - Numerics in PowerPC Assembly Language

Chapter 14 - Assembly-Language Numeric Operations

Comparison Operations

The assembler provides two floating-point comparison instructions:

fcmpo	Ordered comparison
fcmpu	Unordered comparison

The only difference is that the ordered comparison instruction generates an invalid exception if one of the input registers contains a NaN.

The comparison instructions have three operands. They are of the form

instr DST, SRC1, SRC2

DST

A field in the Condition Register (0 through 7) into which the result of the comparison is placed.

SRC1, SRC2

Two floating-point registers.

Comparison instructions are interpreted as

DST \xC6 SRC1 compare SRC2 

The comparison instructions compare the contents of two floating-point registers and place the results of the comparison in a Condition Register field as well as in bits 16 through 19 (field 4) of the FPSCR. The results in the Condition Register and FPSCR are interpreted as follows:

Result	Meaning
0001	Unordered
0010	SRC1 = SRC2
0100	SRC1 > SRC2
1000	SRC1 < SRC2

Use a conditional branch instruction after the comparison instruction to use the results of the comparison, as shown in the following example:

fcmpo    2,f0,f11 # compare f0 to f11 and put result in CR2
blt      2,addr1  # go to addr1 if bit 0 (<) of CR2 is 1
bgt      2,addr2  # go to addr2 if bit 1 (>) of CR2 is 1
beq      2,addr3  # go to addr3 if bit 2 (=) of CR2 is 1
bun      2,addr4  # go to addr4 if bit 3 (unordered) of CR2 is 1