Intel 8086

In 1972, Intel launched the 8008, the first 8-bit microprocessor.[note 2] It implemented an instruction set designed by Datapoint Corporation with programmable CRT terminals in mind, which also proved to be fairly general-purpose. The device needed several additional ICs to produce a functional computer, in part due to it being packaged in a small 18-pin "memory package", which ruled out the use of a separate address bus (Intel was primarily a DRAM manufacturer at the time).

Two years later, Intel launched the 8080,[note 3] employing the new 40-pin DIL packages originally developed for calculator ICs to enable a separate address bus. It has an extended instruction set that is source-compatible (not binary compatible) with the 8008[5] and also includes some 16-bit instructions to make programming easier. The 8080 device was eventually replaced by the depletion-load-based 8085 (1977), which sufficed with a single +5 V power supply instead of the three different operating voltages of earlier chips.[note 4] Other well known 8-bit microprocessors that emerged during these years are Motorola 6800 (1974), General Instrument PIC16X (1975), MOS Technology 6502 (1975), Zilog Z80 (1976), and Motorola 6809 (1978).

Intel 8086 CPU die image

The 8086 project started in May 1976 and was originally intended as a temporary substitute for the ambitious and delayed iAPX 432 project. It was an attempt to draw attention from the less-delayed 16-bit and 32-bit processors of other manufacturers — Motorola, Zilog, and National Semiconductor.

Whereas the 8086 was a 16-bit microprocessor, it used the same microarchitecture as Intel's 8-bit microprocessors (8008, 8080, and 8085). This allowed assembly language programs written in 8-bit to seamlessly migrate.[6] New instructions and features — such as signed integers, base+offset addressing, and self-repeating operations — were added. Instructions were added to assist source code compilation of nested functions in the ALGOL-family of languages, including Pascal and PL/M. According to principal architect Stephen P. Morse, this was a result of a more software-centric approach. Other enhancements included microcode instructions for the multiply and divide assembly language instructions. Designers also anticipated coprocessors, such as 8087 and 8089, so the bus structure was designed to be flexible.

The first revision of the instruction set and high level architecture was ready after about three months,[note 5] and as almost no CAD tools were used, four engineers and 12 layout people were simultaneously working on the chip.[note 6] The 8086 took a little more than two years from idea to working product, which was considered rather fast for a complex design in 1976–1978.

The 8086 was sequenced[note 7] using a mixture of random logic[7] and microcode and was implemented using depletion-load nMOS circuitry with approximately 20,000 active transistors (29,000 counting all ROM and PLA sites). It was soon moved to a new refined nMOS manufacturing process called HMOS (for High performance MOS) that Intel originally developed for manufacturing of fast static RAM products.[note 8] This was followed by HMOS-II, HMOS-III versions, and, eventually, a fully static CMOS version for battery powered devices, manufactured using Intel's CHMOS processes.[note 9] The original chip measured 33 mm² and minimum feature size was 3.2 μm.

The architecture was defined by Stephen P. Morse with some help from Bruce Ravenel (the architect of the 8087) in refining the final revisions. Logic designer Jim McKevitt and John Bayliss were the lead engineers of the hardware-level development team[note 10] and Bill Pohlman the manager for the project. The legacy of the 8086 is enduring in the basic instruction set of today's personal computers and servers; the 8086 also lent its last two digits to later extended versions of the design, such as the Intel 286 and the Intel 386, all of which eventually became known as the x86 family. (Another reference is that the PCI Vendor ID for Intel devices is 8086_h.)

All internal registers, as well as internal and external data buses, are 16 bits wide, which firmly established the "16-bit microprocessor" identity of the 8086. A 20-bit external address bus provides a 1 MB physical address space (2²⁰ = 1,048,576 x 1 byte). This address space is addressed by means of internal memory "segmentation". The data bus is multiplexed with the address bus in order to fit all of the control lines into a standard 40-pin dual in-line package. It provides a 16-bit I/O address bus, supporting 64 KB of separate I/O space. The maximum linear address space is limited to 64 KB, simply because internal address/index registers are only 16 bits wide. Programming over 64 KB memory boundaries involves adjusting the segment registers (see below); this difficulty existed until the 80386 architecture introduced wider (32-bit) registers (the memory management hardware in the 80286 did not help in this regard, as its registers are still only 16 bits wide).

Some of the control pins, which carry essential signals for all external operations, have more than one function depending upon whether the device is operated in min or max mode. The former mode is intended for small single-processor systems, while the latter is for medium or large systems using more than one processor (a kind of multiprocessor mode). Maximum mode is required when using an 8087 or 8089 coprocessor. The voltage on pin 33 (MN/MX) determines the mode. Changing the state of pin 33 changes the function of certain other pins, most of which have to do with how the CPU handles the (local) bus.[note 11] The mode is usually hardwired into the circuit and therefore cannot be changed by software. The workings of these modes are described in terms of timing diagrams in Intel datasheets and manuals. In minimum mode, all control signals are generated by the 8086 itself.

The 8086 has eight more or less general 16-bit registers (including the stack pointer but excluding the instruction pointer, flag register and segment registers). Four of them, AX, BX, CX, DX, can also be accessed as twice as many 8-bit registers (see figure) while the other four, SI, DI, BP, SP, are 16-bit only.

Due to a compact encoding inspired by 8-bit processors, most instructions are one-address or two-address operations, which means that the result is stored in one of the operands. At most one of the operands can be in memory, but this memory operand can also be the destination, while the other operand, the source, can be either register or immediate. A single memory location can also often be used as both source and destination which, among other factors, further contributes to a code density comparable to (and often better than) most eight-bit machines at the time.

The degree of generality of most registers is much greater than in the 8080 or 8085. However, 8086 registers were more specialized than in most contemporary minicomputers and are also used implicitly by some instructions. While perfectly sensible for the assembly programmer, this makes register allocation for compilers more complicated compared to more orthogonal 16-bit and 32-bit processors of the time such as the PDP-11, VAX, 68000, 32016, etc. On the other hand, being more regular than the rather minimalistic but ubiquitous 8-bit microprocessors such as the 6502, 6800, 6809, 8085, MCS-48, 8051, and other contemporary accumulator-based machines, it is significantly easier to construct an efficient code generator for the 8086 architecture.

Another factor for this is that the 8086 also introduced some new instructions (not present in the 8080 and 8085) to better support stack-based high-level programming languages such as Pascal and PL/M; some of the more useful instructions are push mem-op, and ret size, supporting the "Pascal calling convention" directly. (Several others, such as push immed and enter, were added in the subsequent 80186, 80286, and 80386 processors.)

A 64 KB (one segment) stack growing towards lower addresses is supported in hardware; 16-bit words are pushed onto the stack, and the top of the stack is pointed to by SS:SP. There are 256 interrupts, which can be invoked by both hardware and software. The interrupts can cascade, using the stack to store the return addresses.

The 8086 has 64 K of 8-bit (or alternatively 32 K of 16-bit word) I/O port space.

The 8086 has a 16-bit flags register. Nine of these condition code flags are active, and indicate the current state of the processor: Carry flag (CF), Parity flag (PF), Auxiliary carry flag (AF), Zero flag (ZF), Sign flag (SF), Trap flag (TF), Interrupt flag (IF), Direction flag (DF), and Overflow flag (OF). Also referred to as the status word, the layout of the flags register is as follows:[8]

There are also four 16-bit segment registers (see figure) that allow the 8086 CPU to access one megabyte of memory in an unusual way. Rather than concatenating the segment register with the address register, as in most processors whose address space exceeds their register size, the 8086 shifts the 16-bit segment only four bits left before adding it to the 16-bit offset (16×segment + offset), therefore producing a 20-bit external (or effective or physical) address from the 32-bit segment:offset pair. As a result, each external address can be referred to by 2¹² = 4096 different segment:offset pairs.

Although considered complicated and cumbersome by many programmers, this scheme also has advantages; a small program (less than 64 KB) can be loaded starting at a fixed offset (such as 0000) in its own segment, avoiding the need for relocation, with at most 15 bytes of alignment waste.

Compilers for the 8086 family commonly support two types of pointer, near and far. Near pointers are 16-bit offsets implicitly associated with the program's code or data segment and so can be used only within parts of a program small enough to fit in one segment. Far pointers are 32-bit segment:offset pairs resolving to 20-bit external addresses. Some compilers also support huge pointers, which are like far pointers except that pointer arithmetic on a huge pointer treats it as a linear 20-bit pointer, while pointer arithmetic on a far pointer wraps around within its 16-bit offset without touching the segment part of the address.

To avoid the need to specify near and far on numerous pointers, data structures, and functions, compilers also support "memory models" which specify default pointer sizes. The tiny (max 64K), small (max 128K), compact (data > 64K), medium (code > 64K), large (code,data > 64K), and huge (individual arrays > 64K) models cover practical combinations of near, far, and huge pointers for code and data. The tiny model means that code and data are shared in a single segment, just as in most 8-bit based processors, and can be used to build .com files for instance. Precompiled libraries often come in several versions compiled for different memory models.

According to Morse et al.,.[9] the designers actually contemplated using an 8-bit shift (instead of 4-bit), in order to create a 16 MB physical address space. However, as this would have forced segments to begin on 256-byte boundaries, and 1 MB was considered very large for a microprocessor around 1976, the idea was dismissed. Also, there were not enough pins available on a low cost 40-pin package for the additional four address bus pins.

In principle, the address space of the x86 series could have been extended in later processors by increasing the shift value, as long as applications obtained their segments from the operating system and did not make assumptions about the equivalence of different segment:offset pairs.[note 12] In practice the use of "huge" pointers and similar mechanisms was widespread and the flat 32-bit addressing made possible with the 32-bit offset registers in the 80386 eventually extended the limited addressing range in a more general way.

The instruction stream is fetched from memory as words and is addressed internally by the processor to the byte level as necessary. An instruction stream queuing mechanism allows up to 6 bytes of the instruction stream to be queued while waiting for decoding and execution. The queue acts as a First-In-First-Out (FIFO) buffer, from which the Execution Unit (EU) extracts instruction bytes as required. Whenever there is space for at least two bytes in the queue, the BIU will attempt a word fetch memory cycle. If the queue is empty (following a branch instruction, for example), the first byte into the queue immediately becomes available to the EU.[10]

Small programs could ignore the segmentation and just use plain 16-bit addressing. This allows 8-bit software to be quite easily ported to the 8086. The authors of most DOS implementations took advantage of this by providing an Application Programming Interface very similar to CP/M as well as including the simple .com executable file format, identical to CP/M. This was important when the 8086 and MS-DOS were new, because it allowed many existing CP/M (and other) applications to be quickly made available, greatly easing acceptance of the new platform.

The following 8086/8088 assembler source code is for a subroutine named _memcpy that copies a block of data bytes of a given size from one location to another. The data block is copied one byte at a time, and the data movement and looping logic utilizes 16-bit operations.

 
 
 
 
 
 
 
 
 
 
 
0000:1000
 
0000:1000
0000:1000 55
0000:1001 89 E5
0000:1003 06
0000:1004 8B 4E 06
0000:1007 E3 11
0000:1009 8B 76 04
0000:100C 8B 7E 02
0000:100F 1E
0000:1010 07
 
0000:1011 8A 04
0000:1013 88 05
0000:1015 46
0000:1016 47
0000:1017 49
0000:1018 75 F7
 
0000:101A 07
0000:101B 5D
0000:101C 29 C0
0000:101E C3
0000:101F

; _memcpy(dst, src, len)
; Copy a block of memory from one location to another.
;
; Entry stack parameters
;      [BP+6] = len, Number of bytes to copy
;      [BP+4] = src, Address of source data block
;      [BP+2] = dst, Address of target data block
;
; Return registers
;      AX = Zero

            org     1000h       ; Start at 0000:1000h

_memcpy     proc
            push    bp          ; Set up the call frame
            mov     bp,sp
            push    es          ; Save ES
            mov     cx,[bp+6]   ; Set CX = len
            jcxz    done        ; If len = 0, return
            mov     si,[bp+4]   ; Set SI = src
            mov     di,[bp+2]   ; Set DI = dst
            push    ds          ; Set ES = DS
            pop     es

loop        mov     al,[si]     ; Load AL from [src]
            mov     [di],al     ; Store AL to [dst]
            inc     si          ; Increment src
            inc     di          ; Increment dst
            dec     cx          ; Decrement len
            jnz     loop        ; Repeat the loop

done        pop     es          ; Restore ES
            pop     bp          ; Restore previous call frame
            sub     ax,ax       ; Set AX = 0
            ret                 ; Return
            end proc

The code above uses the BP (base pointer) register to establish a call frame, an area on the stack that contains all of the parameters and local variables for the execution of the subroutine. This kind of calling convention supports reentrant and recursive code, and has been used by most ALGOL-like languages since the late 1950s.

The above routine is a rather cumbersome way to copy blocks of data. The 8086 provides dedicated instructions for copying strings of bytes. These instructions assume that the source data is stored at DS:SI, the destination data is stored at ES:DI, and that the number of elements to copy is stored in CX. The above routine requires the source and the destination block to be in the same segment, therefore DS is copied to ES. The loop section of the above can be replaced by:

0000:1011 FC
0000:1012 F3
0000:1013 A4

            cld                  ; Copy towards higher addresses
loop        rep                  ; Repeat until CX = 0
            movsb                ; Move the data block

This copies the block of data one byte at a time. The REP instruction causes the following MOVSB to repeat until CX is zero, automatically incrementing SI and DI and decrementing CX as it repeats. Alternatively the MOVSW instruction can be used to copy 16-bit words (double bytes) at a time (in which case CX counts the number of words copied instead of the number of bytes). Most assemblers will properly recognize the REP instruction if used as an in-line prefix to the MOVSB instruction, as in REP MOVSB.

This routine will operate correctly if interrupted, because the program counter will continue to point to the REP instruction until the block copy is completed. The copy will therefore continue from where it left off when the interrupt service routine returns control.

Simplified block diagram over Intel 8088 (a variant of 8086); 1=main & index registers; 2=segment registers and IP; 3=address adder; 4=internal address bus; 5=instruction queue; 6=control unit (very simplified!); 7=bus interface; 8=internal databus; 9=ALU; 10/11/12=external address/data/control bus.

Although partly shadowed by other design choices in this particular chip, the multiplexed address and data buses limit performance slightly; transfers of 16-bit or 8-bit quantities are done in a four-clock memory access cycle, which is faster on 16-bit, although slower on 8-bit quantities, compared to many contemporary 8-bit based CPUs. As instructions vary from one to six bytes, fetch and execution are made concurrent and decoupled into separate units (as it remains in today's x86 processors): The bus interface unit feeds the instruction stream to the execution unit through a 6-byte prefetch queue (a form of loosely coupled pipelining), speeding up operations on registers and immediates, while memory operations became slower (four years later, this performance problem was fixed with the 80186 and 80286). However, the full (instead of partial) 16-bit architecture with a full width ALU meant that 16-bit arithmetic instructions could now be performed with a single ALU cycle (instead of two, via internal carry, as in the 8080 and 8085), speeding up such instructions considerably. Combined with orthogonalizations of operations versus operand types and addressing modes, as well as other enhancements, this made the performance gain over the 8080 or 8085 fairly significant, despite cases where the older chips may be faster (see below).

• EA = time to compute effective address, ranging from 5 to 12 cycles.
• Timings are best case, depending on prefetch status, instruction alignment, and other factors.

As can be seen from these tables, operations on registers and immediates were fast (between 2 and 4 cycles), while memory-operand instructions and jumps were quite slow; jumps took more cycles than on the simple 8080 and 8085, and the 8088 (used in the IBM PC) was additionally hampered by its narrower bus. The reasons why most memory related instructions were slow were threefold:

• Loosely coupled fetch and execution units are efficient for instruction prefetch, but not for jumps and random data access (without special measures).
• No dedicated address calculation adder was afforded; the microcode routines had to use the main ALU for this (although there was a dedicated segment + offset adder).
• The address and data buses were multiplexed, forcing a slightly longer (33~50%) bus cycle than in typical contemporary 8-bit processors.

However, memory access performance was drastically enhanced with Intel's next generation of 8086 family CPUs. The 80186 and 80286 both had dedicated address calculation hardware, saving many cycles, and the 80286 also had separate (non-multiplexed) address and data buses.

The 8086/8088 could be connected to a mathematical coprocessor to add hardware/microcode-based floating-point performance. The Intel 8087 was the standard math coprocessor for the 8086 and 8088, operating on 80-bit numbers. Manufacturers like Cyrix (8087-compatible) and Weitek (not 8087-compatible) eventually came up with high-performance floating-point coprocessors that competed with the 8087.

Compatible—and, in many cases, enhanced—versions were manufactured by Fujitsu,[22] Harris/Intersil, OKI, Siemens, Texas Instruments, NEC, Mitsubishi, and AMD. For example, the NEC V20 and NEC V30 pair were hardware-compatible with the 8088 and 8086 even though NEC made original Intel clones μPD8088D and μPD8086D respectively, but incorporated the instruction set of the 80186 along with some (but not all) of the 80186 speed enhancements, providing a drop-in capability to upgrade both instruction set and processing speed without manufacturers having to modify their designs. Such relatively simple and low-power 8086-compatible processors in CMOS are still used in embedded systems.

The electronics industry of the Soviet Union was able to replicate the 8086 through both industrial espionage and reverse engineering. The resulting chip, K1810VM86, was binary and pin-compatible with the 8086.

i8086 and i8088 were respectively the cores of the Soviet-made PC-compatible EC1831 and EC1832 desktops. (EC1831 is the EC identification of IZOT 1036C and EC1832 is the EC identification of IZOT 1037C, developed and manufactured in Bulgaria. EC stands for Единая Система.) However, the EC1831 computer (IZOT 1036C) had significant hardware differences from the IBM PC prototype. The EC1831 was the first PC-compatible computer with dynamic bus sizing (US Pat. No 4,831,514). Later some of the EC1831 principles were adopted in PS/2 (US Pat. No 5,548,786) and some other machines (UK Patent Application, Publication No. GB-A-2211325, Published June 28, 1989).

Soviet clone K1810VM86

OKI M80C86A QFP-56

NEC μPD8086D-2 (8 MHz) from the year 1984, week 19 JAPAN (clone of Intel D8086-2)