Intel 80386 manual Numeric Coprocessor Data Types, Other Instructions, Stack Instructions

APPLICATION ARCHITECTURE

Table 2-1.

Principal Data Types and Instructions

Type	Size	Instructions
Integer,	8, 16,32	Move, Exchange, Translate,
Ordinal	bits	Test, Compare, Convert,
		Shift, Double Shift,
		Rotate, Not, Negate, And,
		Or, Exclusive Or, Add,
		Subtract, Multiply,
		Divide, Increment,
		Decrement, Convert
		(M Dve with sign/ zero
		extension)
Unpacked	I digit	Adjust for: Add, Subtract,
Decimal		Multiply, Divide
Packed	2 digits	Adjust for: Add, Subtract
Decimal
String (hyte,	0-4G bytes,	M ave, Load, Store,
word, dword) words, dwords Compare, Scan, Repeat
Bit String	1-4G bits	Test, Test and Set, Test
		and Reset, Test and
		Complement, Scan, Inscrt,
		Extract
Near	32 bits	(Samc as Ordinal)
Pointer!
Far Pointer	48 bits	Load

I.A near pointer is a 32-bit offset into a segment defined by one of the segment/descriptor register pairs. A far pointer is a full logical addess, that is, a selector and an offset.

Figure 2-7 shows how exam pies of the basic data types are stored in memory, Multibyte items can be located at any byte address. Depending on the bus design, additional bus cycles may be required to access an operand located at an address that is not a multiple of its size. Therefore, for best performance independent of bus design, most programs align word operands on word bound- aries, dword operands on doubleword bound- aries, and so on.

2.3.2Numeric Coprocessor Data

Types

An 80287 or 80387 Numeric Coprocessor supple- ments the 80386 with the data types and instruc- tions shown in Table 2-2. Most numeric applica- tions store input values and output results in the integer, real, or packed decimal types and reserve

the temporary real type for intermediate values, where its extended range and precision minimize rounding, underflow, and overflow problems in complex computations. In accordance with this model, a numeric coprocessor performs most computations on temporary real values stored in its registers. Loading any type into the register stack automatically converts the type to tempo- rary real. A temporary real value in a register can be converted to any other type by a Store instruction.

Table 2-2. Principal Numeric Coprocessor

Data Types and Instructions

Type	Size	Instructions
Integer	16,32,	Load, Store, Compare,
	64 bits	Add, Subtract, Multiply,
		Divide
Packed	18 digits	Load, Store
Decimal
Real	32,64 bits	Load, Store, Compare,
		Add, Subtract, Multiply,
		Divide
Temporary	80 bits	Add, Subtract, Multiply,
Real		Divide, Square Root, Scale
		Remainder, Integer Part,
		Change, Sign, Absolute
		Value, Extract Exponent
		and Significand, Compare,
		Examine, Test, Exchange
		Tangent, Arctangent, 2'-1,
		y* Log2 (X + 1), y* Log2
		(X), Load Constant (0.0,
		pi, etc.) (80387 adds Sine,
		Cosine, Sine and Cosine,
		Unordered Compare)

Figure 2-8 shows how numeric data types are stored in memory.

2.3.3 Other Instructions

Not all 80386 instructions are associated with data types. The following paragraphs survey the untyped instructions.

2.3.3.1 Stack Instructions

An 80386 stack is a stack of dwords whose base and top are defined by the SS and ESP

2-7