Floating point registers

The CPU itself has no means for floating point arithmetic operations. In 1980, Intel announced the Intel 8087 - the floating point coprocessor for the 8086 line. 8087 remained as a separate installable device until 1989, when Intel came up with the 80486 (i486) processor, which had an integrated 8087 circuit. However, when talking about floating point registers and floating point instructions, we still refer to 8087 as a floating-point unit (FPU) or, sometimes, still as a floating-point coprocessor (however, the latter is becoming more and more rare).

8087 has eight registers, 80 bits each, arranged in a stack fashion, meaning that operands are pushed onto this stack from the memory and results are popped from the topmost register to the memory. These registers are named ST0 to ST7 (ST--stack) and the most used one, that is, the ST0 register, may be referred to as simply ST.

The floating-point coprocessor supports several data types:

• 80-bit extended-precision real
• 64-bit double-precision real
• 32-bit single-precision real
• 18-digit decimal integer
• 64-bit binary integer
• 32-bit binary integer
• 16-bit binary integer

The floating-point coprocessor will be discussed in more detail in Chapter 3, Intel Instruction Set Architecture (ISA).