Skip to content

Instantly share code, notes, and snippets.

@seanjensengrey

seanjensengrey/octal_x86.txt

Last active July 9, 2025 04:26
Show Gist options
  • Save seanjensengrey/f971c20d05d4d0efc0781f2f3c0353da to your computer and use it in GitHub Desktop.
Save seanjensengrey/f971c20d05d4d0efc0781f2f3c0353da to your computer and use it in GitHub Desktop.
Download ZIP

Revisions

  1. seanjensengrey revised this gist May 23, 2024. 1 changed file with 1 addition and 1 deletion.
    2 changes: 1 addition & 1 deletion octal_x86.txt
    View file
    Original file line number Diff line number Diff line change
    @@ -1,4 +1,4 @@
    # source:http://reocities.com/SiliconValley/heights/7052/opcode.txt
    # source:http://geocities.com/SiliconValley/heights/7052/opcode.txt
    From: [email protected] (Mark Hopkins)
    Newsgroups: alt.lang.asm
    Subject: A Summary of the 80486 Opcodes and Instructions
  2. seanjensengrey created this gist Apr 1, 2018.
    1,144 changes: 1,144 additions & 0 deletions octal_x86.txt
    View file
    Original file line number Diff line number Diff line change
    @@ -0,0 +1,1144 @@
    # source:http://reocities.com/SiliconValley/heights/7052/opcode.txt
    From: [email protected] (Mark Hopkins)
    Newsgroups: alt.lang.asm
    Subject: A Summary of the 80486 Opcodes and Instructions

    (1) The 80x86 is an Octal Machine
    This is a follow-up and revision of an article posted in alt.lang.asm on
    7-5-92 concerning the 80x86 instruction encoding.

    The only proper way to understand 80x86 coding is to realize that ALL 80x86
    OPCODES ARE CODED IN OCTAL. A byte has 3 octal digits, ranging from 000 to
    377. In fact, each octal group (000-077, 100-177, etc.) tends to encode a
    specific variety of operation. All of these are features inherited from the
    8080/8085/Z80.
    For some reason absolutely everybody misses all of this, even the Intel
    people who wrote the reference on the 8086 (and even the 8080). The opcode
    scheme outlined briefly below is expanded starting in the 80386, but
    consistently with the overall scheme here.

    As an example to see how this works, the mov instructions in octal are:

    210 xrm mov Eb, Rb
    211 xrm mov Ew, Rw
    212 xrm mov Rb, Eb
    213 xrm mov Rw, Ew
    214 xsm mov Ew, SR
    216 xsm mov SR, Ew

    The meanings of the octal digits (x, m, r, s) and their correspondence to the
    operands (Eb, Ew, Rb, Rw, SR) are the following:

    The digit r (0-7) encodes the register operand as follows:
    REGISTER (r): 0 1 2 3 4 5 6 7
    Rb = Byte-sized register AL CL DL BL AH CH DL BH
    Rw = Word-sized register AX CX DX BX SP BP SI DI

    The segment register digit s (0-7) encodes the segment register as follows:
    SEGMENT REGISTER (s): 0 1 2 3 4 5 6 7
    SR = Segment register ES CS SS DS <Reserved>

    The digits x (0-3), and m (0-7) encode the address mode according to
    the following scheme. One or more bytes (labeled: Disp) may immediately
    follow xrm as described below.

    TABLE 1: 16-BIT ADDRESSING MODE (x, m):
    Eb = Address of byte-sized object in memory or register
    Ew = Address of word-sized object in memory or register
    Dw = Unsigned word
    Dc = Signed byte ("character"), range: -128 to +127 (decimal).
    Db = Unsigned byte

    x m Disp Eb Ew
    ------------------
    3 r Rb Rw
    0 6 Dw DS:[Dw]
    0 m Base:[0] (except for xm = 06).
    1 m Dc Base:[Dc]
    2 m Dw Base:[Dw]

    x 0 Disp DS:[BX + SI + Disp]
    x 1 Disp DS:[BX + DI + Disp]
    x 2 Disp SS:[BP + SI + Disp]
    x 3 Disp SS:[BP + DI + Disp]
    x 4 Disp DS:[SI + Disp]
    x 5 Disp SS:[DI + Disp]
    x 6 Disp DS:[BP + Disp] (except for xm = 06)
    x 7 Disp DS:[BX + Disp]

    This expands into the following table:

    TABLE 1a: 16-BIT ADDRESSING MODE (x, m) for the expansion impaired. :)
    xm Eb/Ew xm Eb/Ew xm Eb/Ew xm Eb/Ew
    00 DS:[BX + SI] 10 Dc DS:[BX + SI + Dc] 20 Dw DS:[BX + SI + Dw] 30 AL/AX
    01 DS:[BX + DI] 11 Dc DS:[BX + BI + Dc] 21 Dw DS:[BX + DI + Dw] 31 CL/CX
    02 SS:[BX + SI] 12 Dc SS:[BP + SI + Dc] 22 Dw SS:[BP + SI + Dw] 32 DL/DX
    03 SS:[BX + DI] 13 Dc SS:[BP + DI + Dc] 23 Dw SS:[BP + DI + Dw] 33 BL/BX
    04 DS:[SI] 14 Dc DS:[SI + Dc] 24 Dw DS:[SI + Dw] 34 AH/SP
    05 DS:[DI] 15 Dc DS:[DI + Dc] 25 Dw DS:[DI + Dw] 35 CH/BP
    06 Dw DS:[Dw] 16 Dc SS:[BP + Dc] 26 Dw SS:[BP + Dw] 36 DH/SI
    07 DS:[BX] 17 Dc DS:[BX + Dc] 27 Dw DS:[BX + Dw] 37 BH/DI

    Operands where x is 0, 1, or 2 are all pointers. If the instruction is a WORD
    instruction (211, 213, 214, 216 are), then this pointer addresses a
    word-sized object. The format of the object at the indicated address will
    always be low-order byte first, and high-order byte second. Otherwise the
    instruction is a BYTE instruction (210, 212) and the pointer addresses
    byte-sized object at the indicated address.

    The default segments (DS:, SS:) can be overridden with a segment prefix. In
    all cases it's understood that everything has the default segment DS, except
    for the two stack/frame pointers (BP and SP) whose default segment is SS.
    That will be explained below.

    Modes where x = 1, or 2 will require displacement bytes (Dc or Dw) to follow
    the opcode as explained above.

    When x = 3, WORD sized instructions address the word registers (AX, CX, ...)
    and the BYTE size instructions the byte registers (AL, CL, ...).

    EXAMPLE 1: The instruction opcode: 210 135 375
    Here, xm = 15, and r = 3, so the operands are:

    mov Eb, Rb
    =>
    mov byte ptr DS:[DI + Dc], BL

    The displacement, Dc, is 375 (or fd in hexadecimal), which is the signed byte
    -3. So the instruction reads:

    mov byte ptr DS:[DI - 3], BL

    or just:
    mov [DI - 3], BL

    In C-like notation, the meaning of this operation would be:

    ((byte *)DS) [DI - 3] = BL;

    EXAMPLE 2: The instruction opcode: 216 332
    Here, xm = 32, and s = 3, so the operands are:

    mov SR, Ew
    =>
    mov DS, DX

    A move to CS is not possible (because the far jump instruction already does
    that) so that the opcode sequence:

    216 x2m

    is free to be used for encoding something else.

    EXAMPLE 3: As an illustration of why it's better to think in octal, just look
    at the opcodes for the binary arithmetic instructions:

    0P0 xrm Op Eb, Rb
    0P1 xrm Op Ew, Rw
    0P2 xrm Op Rb, Eb
    0P3 xrm Op Rw, Ew
    0P4 Db Op AL, Db
    0P5 Dw Op AX, Dw

    They all have the same form, with a single digit encoding the operator as
    follows:
    P Op P Op
    0 add 1 or
    2 adc 3 sbb
    4 and 5 sub
    6 xor 7 cmp

    That's a good fraction of your reference table right there.

    EXAMPLE 4: The same mapping is used in the immediate to memory/register form
    of these operations:

    200 xPm Db Op Eb, Db
    201 xPm Dw Op Ew, Dw
    203 xPm Dc Op Ew, Dc

    (2) An Outline of 80x86 Instructions and Encoding
    The authors of 8080 and 8086 references (including Intel's own references)
    are apparently not aware of the octal nature of their own machines, and the
    result is an almost grotesque complication and bungling up in the presentation
    of something that is actually fairly simple. Thus, people claim that it's
    almost impossible to know 8086 binary by heart, whereas in fact I know most of
    it by memory. I'll straighten out the mess for you here.

    As alluded to above, instructions are encoded as follows:

    op xrm Const

    where * op is a 1 or 2 byte opcode,
    * xrm (if present) constitutes 3 octal digits whose normal uses are:
    r = Register operand, xm = Memory or Register operand.
    It may be followed immediately by a "displacement" byte or word,
    depending solely on the digits x and m.
    * Const (if present) denotes a byte or word value whose presence and
    format depends solely on what op (and sometimes xrm) is.

    In some cases, the opcode itself may be separated out into octal digits, e.g.

    0s6 = push (Segment Register #s).

    The one major exception to the coding scheme are all the conditional code
    operations. Since there are 16 distinct conditional codes, they are
    represented as a hexadecimal digit. The conditional jump in octal ranges
    from 160 to 177, which is 7x in hexadecimal, where x is a hex digit encoding
    the jump's condition. I'll represent them by the format: 160+CC.

    The register and address encoding was described above. The '386 expands on
    this a little with the addition of two segment registers:

    SEGMENT REGISTER (s): 0 1 2 3 4 5 6 7
    SR = Segment register ES CS SS DS FS GS <Reserved>

    In TABLE 1, note that the addresses encoded on modes 0m, 1m, 2m are the same
    regardless of whether you're referring to Eb or Ew. What distinguishes them
    is the size of the object being pointed to and this can be explicitly
    indicated in traditional '86 assemblers like the following examples:

    byte ptr [BP]
    word ptr [BX + DI]

    As explained before, all addresses, except those involving BP refer to the
    data segment, DS. All the BP's refer to the stack segment, SS. This
    is about to be explained.

    (3) Segmentation and Registers
    The 80x86 was designed with more or less specific uses for its registers.
    In fact, the names are supposed to reflect their main uses:

    AX (AH:AL) = Accumulator
    BX (BH:BL) = Base Register
    CX (CH:CL) = Counting Register
    DX (DH:DL) = Data Register

    CS = Code Segment -- where constants and programs lie.
    DS = Data Segment -- where static variables lie.
    SS = Stack Segment -- where auto variables and function parameters lie.
    SP, BP = Stack and Frame Pointers, used to segment out the
    local variables and function parameters.
    ES = Extra Segment -- used in combination with the index registers for
    string operations as follows:
    DS:[SI] -- points to the Source of the string operation.
    ES:[DI] -- points to the Destinction of the string operation.

    The typical setup for the stack is as follows:

    High Addresses FUNCTION DEFINITION: FUNCTION CALL:
    ... mov BP, SP push Parameters
    Parameters push BP call Function
    Return Address sub SP, Locals
    BP -> Old BP ... function ...
    Local Variables mov SP, BP
    SP -> ... pop BP
    Low Addresses ret Parameters

    this dictates a certain protocol in calling functions with parameters and
    returning from them, as shown above. In fact, this is so much so that the
    opening and closing sequences above have all been defined as single operations
    starting with the 80286 so that the function definition above can be rewritten
    as:
    FUNCTION DEFINITION:
    enter Locals, 0
    ... function ...
    leave
    ret Parameters

    (4) Word and Address Size on the 80386 and Above
    Starting with the 80386, operations can be done with not just 16-bit words
    but also 32 bit words. Generally the same operation is defined for both sets
    and context is used to determine which is which in the following two ways:

    * Which mode the machine is running in
    Protected mode -- both word sizes and address sizes are 32-bits
    Real & Virtual modes -- 16-bits.
    * The presence of certain prefixes to override either the default
    word size, address size or both on an instruction-by-instruction
    basis.

    (a) Word Size
    When the word size for the current operation is 32-bits, everything listed
    above as "word" is interpreted as 32-bits, including registers. The register
    numbering corresponding to this word size is:

    REGISTER (r): 0 1 2 3 4 5 6 7
    Rb = Byte-sized register AL CL DL BL AH CH DL BH
    Rd = Dword-sized register EAX ECX EDX EBX ESP EBP ESI EDI

    (b) Address Size
    When the address size is switched to 32-bits, the address scheme listed in
    TABLE 1 is altered in its entirety.

    TABLE 2: 32-BIT ADDRESSING MODE (x, m): Encoding of scaled index SI:
    x m Disp Eb Ew si SI
    -------------------- ---------------
    0 6 Dw DS:[Dw] s0 EAX * 2^s
    0 4 sir [Rd + SI + 0] s1 ECX * 2^s
    1 4 sir Dc [Rd + SI + Dc] s2 EDX * 2^s
    2 4 sir Dw [Rd + SI + Dw] s3 EBX * 2^s
    0 r [Rd + 0] (except r = 4, 6) 04 0
    1 r Dc [Rd + Dc] (except r = 4) s5 EBP * 2^s
    2 r Dw [Rd + Dw] (except r = 4) s6 ESI * 2^s
    3 r Rb Rw s7 EDI * 2^s

    The encodings si = 14, 24 and 34 remain undefined.

    This alteration is INDEPENDENT of the word size setting. That means that
    even the "Dw"'s, "Rw"'s in the chart above will vary in interpretation as
    16-bit or 32-bit objects depending on the word size setting. That leads to
    4 possible combinations, not just 2.

    EXAMPLE 5: The opcode sequence 211 135 375
    This is the operation
    mov Ew, Rw
    where xm = 15, r = 3 and Disp = -3. The 4 combinations are:

    Addr-Size Word-Size Operation
    16 16 mov word ptr [DI - 3], BX
    16 32 mov dword ptr [DI - 3], EBX
    32 16 mov word ptr [EBP - 3], BX
    32 32 mov dword ptr [EBP - 3], EBX

    EXAMPLE 6: The opcode sequence 211 134 302 375 with 32-bit addressing.
    This is the move instruction where xm = 14 and r = 3.

    mov Ew, [E]BX ([E]BX since r = 3)

    It uses the indexed register addressing. The address, Ew, may be derived
    as follows:
    x m sir Disp Ew Comments
    1 4 sir Dc [EDX + SI + Dc]
    1 4 si2 375 [EDX + SI - 3] (Rd = EDX for r = 2)
    1 4 302 375 [EDX + 8*EAX - 3] (SI = 8*EAX for si = 30)

    Therefore, this instruction represents one of the following:

    Word-Size Operation: 211 134 302 375
    16 mov word ptr [EDX + 8*EAX - 3], BX
    32 mov dword ptr [EDX + 8*EAX - 3], EBX

    (5) The Opcode Summary
    The chart below summarises all the machine instructions. The following
    abbreviations are used:

    Registers: Immediate Data Constant:
    Rb (byte sized) Db (byte sized)
    Rw (word sized) Dw (word sized)
    Rd (dword sized) Dc (signed byte)

    Register/Memory Address: Relative Code Address:
    Eb (byte sized) Cb (byte sized)
    Ew (word sized) Cw (word sized)

    Memory Address: Code Address:
    Es (16 bit selector) Af (32/48 bit absolute far code address)
    En (near 16/32 bit pointer)
    Ef (far i32/48 bit pointer)
    Ep (pointer to 6-byte object)
    Ea (generic address)

    Processor Extensions:
    * = 80186 extension
    $ = 80286 extension
    # = 80386 extension
    @ = 80486 extension

    The switch between 16 and 32 bit word size affects all operands labeled
    Rw, Ew, Dw, Cw, En and even Af and Ef. The latter two objects refer to
    far code addresses which are 4 bytes when the word size is 16 bits, and
    6 bytes else.

    The only such operands not actually affected by the word-size switch are
    those whose size a consequence of the operation's meaning. These include
    the following: RET, BOUND, ARPL, SMSW, LMSW, LAR and LSL.

    The switch between 16 and 32 bit address size affects all the operands
    labeled Eb, Ew, Es, En, Ef, Ep, and Ea. Each of these is interpreted
    according to the xm digts in the opcode according to either the 16-bit
    addres table described near the start of the article or the 32-bit address
    table just described above.

    NOTE: In the following presentation everything is in octal.

    ARITHMETIC & LOGIC
    ------------------
    Comments:
    * All of these operations affect all 6 arithmetic flags, except NOT (which
    affects no flags), and INC and DEC (which don't affect CF).
    * IMUL and MUL only affect CF and OF predictably.
    * IDIV and DIV affect no flags predictably.
    * AND, OR, XOR, and TEST all set CF and OF to 0 and alter AF unpredictably.
    * CMP and TEST have no affect on any operands. They're used for setting
    flags. CMP is used for doing relational operators (< > <= >= == !=), and
    TEST for doing bit-testing.
    * CMP and TEST can have their operands listed in either order.
    P Op Description
    0 ADD L, E L += E
    2 ADC L, E L += E + CF
    5 SUB L, E L -= E
    3 SBB L, E L -= E + CF
    7 CMP L, E (void)(L - E)
    1 OR L, E L |= E
    4 AND L, E L &= E
    6 XOR L, E L ^= E
    0P0 xrm Op Eb, Rb
    0P1 xrm Op Ew, Rw
    0P2 xrm Op Rb, Eb
    0P3 xrm Op Rw, Ew
    0P4 Db Op AL, Db
    0P5 Dw Op AX, Dw
    200 xPm Db Op Eb, Db
    201 xPm Dw Op Ew, Dw
    203 xPm Dc Op Ew, Dc

    NOT L L = ~L
    366 x2m not Eb
    367 x2m not Ew
    NEG L L = -L
    366 x3m neg Eb
    367 x3m neg Ew

    INC L L++
    10r inc Rw
    376 x0m inc Eb
    377 x0m inc Ew
    DEC L L--
    11r dec Rw
    376 x1m dec Eb
    377 x1m dec Ew

    TEST L, E (void)(L&E)
    204 xrm test Rb, Eb
    205 xrm test Rw, Ew
    250 Db test AL, Db
    251 Dw test AX, Dw
    366 x0m Db test Eb, Db
    367 x0m Dw test Ew, Dw

    IMUL L, E, D L = (signed)E*D
    IMUL L, E L = (signed)L*E
    # 017 257 xrm Dw imul Rw, Ew
    * 151 xrm Dw imul Rw, Ew, Dw
    * 153 xrm Db imul Rw, Ew, Dc

    In the following operations:
    Operand Size ACC' ACC
    1 AX AL
    2 DX:AX AX
    4 EDX:EAX EAX
    P Op Description
    4 MUL E ACC' = (unsigned) ACC*E
    5 IMUL E ACC' = (signed) ACC*E
    6 DIV E ACC' = (unsigned) ACC%E : ACC/E
    7 IDIV E ACC' = (signed) ACC%E : ACC/E
    366 xPm Op Eb
    367 xPm Op Ew

    SHIFTS & ROTATIONS
    ------------------
    Comments:
    * Where applicable, N is masked off by 0x1f.
    * For Rxx and Sxx, OF is predictably affected only when N is 1.
    * SHLD and SHRD affect all 6 arithmetic flags, but OF and AF unpredictably.
    * RxL: OF = (CF != high order bit of L) before shift
    * RxR: OF = (high order bit of L != next high order bit of L) before shift
    * SxL: OF = (CF != sign bit of L) after shift
    * SxR: OF = (sign bit of L) after shift
    P Op Description
    0 ROL CF <- [<-<-<-] <- high order bit Rotate
    1 ROR low order bit -> [->->->] -> CF
    2 RCL CF <- [<-<-<-] <- CF Rotate Through CF
    3 RCR CF -> [->->->] -> CF
    4 SHL CF <- [<-<-<-] <- 0 Shift (unsigned)
    5 SHR 0 -> [->->->] -> CF
    4 SAL CF <- [<-<-<-] <- 0 Shift (signed)
    7 SAR sign bit -> [->->->] -> CF
    * 300 xPm Db Op Eb, Db
    * 301 xPm Db Op Ew, Db
    320 xPm Op Eb, 1
    321 xPm Op Ew, 1
    322 xPm Op Eb, CL
    323 xPm Op Ew, CL

    SHLD L, E, N CF:L = L:E << N
    SHRD L, E, N L:CF = E:L >> N
    # 017 244 Db shld Ew, Rw, Db
    # 017 245 shld Ew, Rw, CL
    # 017 254 Db shrd Ew, Rw, Db
    # 017 255 shrd Ew, Rw, CL

    TYPE CONVERSIONS
    ----------------
    [] Decimal Conversions
    Comments:
    * DAA and DAS are used for adjusting the results of addition and subtraction
    respectively back to packed BCD format. They will alter all 6 of the
    arithmetic flags, OF unpredictably.
    * AAA, AAS, AAD, and AAM are used for adjusting the results of the four
    basic arithmetic operations back to unpacked BCD format or ASCII format.
    However, AAD is used *before* a divide operation. They too affect all
    6 of the arithmetic flags, but only AF and CF predictably (for AAA and
    AAS) or SF, ZF and PF (for AAD and AAM).
    * In the following, A0 stands for the lower 4 bits of AL and A1 the upper
    4 bits of AL.
    * The binary codes for AAM and AAD each consist of an opcode followed by
    a constant 10 (012 in octal). It has been said that this "10" is
    actually a hidden parameter to a more general AAD and AAM operator,
    which can actually be used for any base other than 10. Some processors
    will not allow AAD to be generalized in this way, however. The reason it
    was left out in the open like this was supposedly because the original
    8086 design literally ran out of space to pack in the opcode.
    DAA if (A0 > 9) AF = 1; if (AF) AL += (0x10 - 10);
    if (A1 > 9) CF = 1; if (CF) AL += (0x10 - 10)*0x10;
    DAS if (A0 > 9) AF = 1; if (AF) AL -= (0x10 - 10);
    if (A1 > 9) CF = 1; if (CF) AL += (0x10 - 10)*0x10;
    AAA if (A0 > 9) AF = 1; CF = AF; if (CF) A0 += (0x10 - 10), AH++;
    AAS if (A0 > 9) AF = 1; CF = AF; if (CF) A0 -= (0x10 - 10), AH--;
    AAM AX = AL/10 : AL%10
    AAD AX = (10*AH + AL)%0x10
    047 daa
    057 das
    067 aaa
    077 aas
    324 012 aam
    325 012 aad

    [] Sign Conversions
    Comments:
    * In converting from a shorter to longer operand size, sign conversion
    involves either taking the leading (sign) bit and replicating it leftward
    (conversion to signed), or placing zero's on the left (for conversion
    to unsigned).
    MOVSX L, E L = (signed)E
    MOVZX L, E L = (unsigned)E
    # 017 266 xrm movsx Rw, Eb
    # 017 267 xrm movsx Rw, Ew
    # 017 266 xrm movzx Ew, Rb
    # 017 277 xrm movzx Ew, Rw

    CBW AX = (signed)AL
    CWDE EAX = (signed)AX
    CWD DX:AX = (signed)AX
    CDQ EDX:EAX = (signed)EAX
    230 cbw / (#) cwde
    231 cwd / (#) cdq

    [] Byte Ordering
    * Used to convert between "little Endian" (Intel byte ordering) and "big
    Endian" (Motorola byte ordering). Typical use: networking applications.
    BSWAP L L[0]:L[1]:L[2]:L[3] = L[3]:L[2]:L[1]:L[0]
    @ 017 31r bswap Rd

    [] Table Lookup
    XLATB AL = [BX + AL]
    327 xlatb

    SEMAPHORES & SYNCHRONIZATION
    ----------------------------
    Comments:
    * All these operations affect all 6 arithmetic flags. BT, BTS, BTR, BTC
    affect only CF predictably; and BSF and BSR affect only ZF predictably.
    * ACC is either AL, AX or EAX in CMPXCHG, depending on the operand size.
    * WAIT is used in the '486 to force a pending unmasked interrupt from the
    internal floating point processing unit.
    * LOCK is a prefix used in multi-CPU contexts to assure exclusive access to
    memory for the following two-step read & modify operations:
    (INC, DEC, NEG, NOT) Mem (ADD, ADC, SUB, SBB) Mem, Src
    (BT, BTS, BTR, BTC) Mem, Src (AND, XOR, OR) Mem, Src
    XCHG Reg, Mem XCHG Mem, Reg
    But XCHG automatically does its own LOCK so does not need to be prefixed.
    P Op Description
    4 BT L, N CF = L.N;
    5 BTS L, N CF = L.N; L.N = 1;
    6 BTR L, N CF = L.N; L.N = 0;
    7 BTC L, N CF = L.N; L.N = !L.N;
    # 017 2P3 xrm Op Ew, Rw
    # 017 272 xPm Db Op Ew, Db

    BSF L, E ZF = !E; if (ZF) L = First 1-bit position in E; else L = ???
    BSR L, E ZF = !E; if (ZF) L = Last 1-bit position in E; else L = ???
    # 017 274 xrm bsf Rw, Ew
    # 017 275 xrm bsr Rw, Ew

    CMPXCHG L, E ZF = (ACC == L); if (ZF) L = E; else ACC = L;
    @ 017 246 xrm cmpxchg Eb, Rb
    @ 017 247 xrm cmpxchg Ew, Rw
    XADD L, L' <L, L'> = <L + L', L>
    @ 017 300 xrm xadd Eb, Rb
    @ 017 301 xrm xadd Ew, Rw

    NOP Delay 1 cycle.
    WAIT Wait for coprocessor unit.
    LOCK Hardware memory bus semaphore.
    HLT Wait for a reset or interrupt.
    220 nop
    233 wait
    360 lock
    364 hlt

    INT N push [E]FLAGS, CS, [E]IP; TF = 0;
    if (the Nth entry in the IDT is a Interrupt Gate) IF = 0;
    jmp to the far address listed under the Nth entry in the IDT
    INTO if (OF) INT 4
    IRET if (NT) return to task listed under TSS.BackLink;
    else pop [E]IP, CS, [E]FLAGS;
    314 int 3
    315 Db int Db
    316 into
    317 iret

    FLAGS
    -----
    Comments:
    * No flags are affected except the explicit moves to the FLAGS register:
    POPF[D] and SAHF, but SAHF only sets the arithmetic flags (except OF).
    POPF pop FLAGS
    POPFD pop EFLAGS
    PUSHF push FLAGS
    PUSHFD push EFLAGS
    SAHF FLAGS |= (AH & 0xd5)
    LAHF AH = FLAGS;
    234 pushf / (#) pushfd
    235 popf / (#) popfd
    236 sahf
    237 lahf
    CMC CF = !CF
    CLC CF = 0
    STC CF = 1
    CLI IF = 0 (Interrupts off)
    STI IF = 1 (Interrupts on)
    CLD DF = 0 (Set string ops to increment)
    STD DF = 1 (Set string ops to decrement)
    365 cmc
    370 clc
    371 stc
    372 cli
    373 sti
    374 cld
    375 std

    CONDITIONAL OPERATIONS
    ----------------------
    (NOTE: The values listed for CC are in octal).

    CC Condition(s) Definition Descriptions
    07 A NBE !CF && !ZF x > y x > 0 (unsigned)
    03 AE NB !CF x >= y x >= 0 (unsigned)
    02 B NAE CF x < y x < 0 (unsigned)
    06 BE NA CF || ZF x <= y x <= 0 (unsigned)
    17 G NLE SF == OF && !ZF x > y x > 0 (signed)
    15 GE NL SF == OF x >= y x >= 0 (signed)
    14 L NGE SF != OF x < y x < 0 (signed)
    16 LE NG SF != OF || ZF x <= y x <= 0 (signed)
    04 E Z ZF x == y x == 0
    05 NE NZ !ZF x != y x != 0
    00 O OF Overflow (signed overflow)
    01 NO !OF No overflow (signed overflow)
    02 C CF Carry (unsigned overflow)
    03 NC !CF No carry (unsigned overflow)
    10 S SF (Negative) sign
    11 NS !SF No (negative) sign
    12 P PE PF Parity [even]
    13 NP PO !PF No parity (parity odd)
    CC cc Cond.

    Jcc Rel if (Cond) EIP += Rel;
    SETcc L L = (Cond)? 1: 0;
    # 017 200+CC Cw jcc Cw '=&h80+cc
    # 017 220+CC x0m setcc Rb '=&h90+cc
    160+CC jcc Cb '=&h70+cc

    STACK OPERATIONS
    ----------------
    Comments:
    * PUSHA[D] uses the value SP had before the operation started.
    * POPA[D] doesn't actually affect [E]SP, which is why it's bracketed out.
    * POP CS is not allowed because it's already subsumed by the RET (far)
    operation. Instead, 017 is used as a 2-byte operation prefix.
    * POP SS inhibits interrupts in order to allow [E]SP to be altered in the
    following operation -- for what should be obvious reasons.
    PUSH E SP -= sizeof E; SS:[SP] = E;
    PUSHA push AX, CX, DX, BX, SP, BP, SI, DI
    PUSHAD push EAX, ECX, EDX, EBX, ESP, EBP, ESI, EDI
    # 017 240 push FS
    # 017 250 push GS
    0s6 push SR (s = 0-3)
    12r push Rw
    * 140 pusha / (#) pushad
    150 Dw push Dw
    152 Dc push Dc
    377 x6m push Ew
    POP L L = SS:[SP]; SP += sizeof L;
    POPA pop DI, SI, BP, (SP), BX, DX, CX, AX
    POPAD pop EDI, ESI, EBP, (ESP), EBX, EDX, ECX, EAX
    # 017 241 pop FS
    # 017 251 pop GS
    0s7 pop SR (s = 0, 2-3)
    13r pop Rw
    * 141 popa / (#) popad
    217 x0m pop Ew

    TRANSFER OPERATIONS
    -------------------
    Comments:
    * XCHG can have its operands listed in either order.
    * MOV CS, ... is not allowed, since this is already subsumed by JMPs.
    * LCS ... is not allowed either for the same reason.
    * XCHG AX, AX is one and the same as NOP.
    XCHG L, E <L, E> = <E, L>
    206 xrm xchg Rb, Eb
    207 xrm xchg Rw, Ew
    22r xchg AX, Rw (r != 0)
    MOV L, E L = E;
    210 xrm mov Eb, Rb
    211 xrm mov Ew, Rw
    212 xrm mov Rb, Eb
    213 xrm mov Rw, Ew
    214 xsm mov Es, SR (s = 0-3, (#) 4-5)
    216 xsm mov SR, Es (s = 0,2-3, (#) 4-5)
    240 Dw mov AL, [Dw]
    241 Dw mov AX, [Dw]
    242 Dw mov [Dw], AL
    243 Dw mov [Dw], AX
    26r Db mov Rb, Db
    27r Dw mov Rw, Dw
    306 x0m Db mov Eb, Db
    307 x0m Dw mov Ew, Dw
    LEA L, An L = &An;
    215 xrm lea Rw, En (x != 3)
    LSeg L, Af Seg:L = &Af;
    # 017 262 xrm lss Rw, Ef (x != 3)
    # 017 264 xrm lfs Rw, Ef (x != 3)
    # 017 265 xrm lgs Rw, Ef (x != 3)
    304 xrm les Rw, Ef (x != 3)
    305 xrm lds Rw, Ef (x != 3)

    ADDRESSING
    ----------
    Comments:
    * The current mode of the machine determines its default mode (16 or 32
    bits).
    * RAND: and ADDR: (not an standard name, since Intel has none) are
    prefixes that alter the default for the next instruction only.
    * RAND: changes the word size between 16 and 32 bits.
    * ADDR: changes the address size between 16 and 32 bits.
    * seg: cannot override the implied ES:[DI] operand in any string op,
    but can override the DS in the implied DS:[SI] operands there.
    seg: Segment override prefix
    ADDR: Address size toggle
    RAND: Operand size toggle
    305 xrm lds Rw, Ef (x != 3)
    046 ES:
    056 CS:
    066 SS:
    076 DS:
    # 144 FS:
    # 145 GS:
    # 146 RAND:
    # 147 ADDR:

    PORT I/O
    --------
    Comments:
    * In protected mode the user of these operations must pass the I/O
    Privilege Level (IOPL) else they are blocked by an interrupt.
    This allows the Operating System to spool I/O devices in a
    multitasking system (since the OS handles interrupts) to avoid having
    processes all trying to use the same device at once.
    IN ACC, Port ACC = IO[Port]
    344 Db in AL, Db
    345 Db in AX, Db
    354 in AL, DX
    355 in AX, DX
    OUT Port, ACC IO[Port] = ACC
    346 Db out Db, AL
    347 Db out Db, AX
    356 out DX, AL
    357 out DX, AX

    STRING OPERATIONS
    -----------------
    Comments:
    * In all these operations below, Src denotes DS;[ESI] and Dest ES:[EDI].
    * Dest cannot be overridden by a segment prefix, only Src.
    * The pointes (ESI, EDI) are bumped up (DF = 0) or down (DF = 1) after
    the operation by sizeof Operand.
    * ACC is either AL, AX or EAX depending on the operand size.
    * The flags altered are exactly those altered by the corresponding
    MOV, IN, OUT, or CMP operation (namely: only SCAS and CMPS alter the
    flags and in the same way as CMP) and these are therefore the only ones
    that can be prefixed by REP[N]E/REP[N]Z.
    * REP with all string ops, but REP LODS doesn't do anything sensible.
    INS in Dest, DX
    OUTS out DX, Src
    MOVS mov Dest, Src
    CMPS cmp Dest, Src
    STOS mov Dest, ACC
    LODS mov ACC, Src
    SCAS cmp ACC, Dest
    * 154 insb
    * 155 insw / (#) insd
    * 156 outsb
    * 157 outsw / (#) outsd
    244 movsb
    245 movsw / (#) movsd
    246 cmpsb
    247 cmpsw / (#) cmpsd
    252 stosb
    253 stosw / (#) stosd
    254 lodsb
    255 lodsw / (#) lodsd
    256 scasb
    257 scasw / (#) scasd
    REP Op while (CX-- > 0) Op
    REPE /REPZ Op while (CX-- > 0 && ZF) Op
    REPNE/REPNZ Op while (CX-- > 0 && !ZF) Op
    362 repne / repnz / rep
    363 repe / repz

    CONTROL FLOW
    ------------
    Comments:
    * The distinction between near and far jumps/calls/returns is built right
    into the 8086 language, which pretty much forces you to explicitly
    declare a routine as "near" or "far" and be consistent about it. The
    intended usage runs pretty much like C's static vs. global functions,
    with each C file being analogous to an 8086 segment.
    * The 8086 was specifically designed to be a Pascal (and PL/I) machine,
    though. Intel wrongly assumed that one of these languages would become
    like C is now. So the ENTER and LEAVE operators were added (and BOUND
    to do array bounds-checking). The segmentation structure was intended
    to support these types of languages.
    JCXZ Rel if (!CX) IP += Rel;
    JECXZ Rel if (!ECX) IP += Rel;
    LOOPcc Rel if (!--CX && cc) IP += Rel;
    340 Cb loopnz Cb / loopne Cb
    341 Cb loopz Cb / loope Cb
    342 Cb loop Cb
    343 Cb jcxz Cb / (#) jecxz Cb
    JMP Rel IP += Rel;
    JMP FAR Af CS:IP = Af;
    CALL Rel push IP; IP += Rel;
    CALL FAR Af push CS, IP; IP = Af;
    232 Af call Af
    350 Cw call Cw
    351 Cw jmp Cw
    352 Af jmp far Af
    353 Cb jmp Cb
    377 x2m call En
    377 x3m call far Ef
    377 x4m jmp En
    377 x5m jmp far Ef
    RET Params pop IP; SP += Params (default: Params = 0)
    RET FAR Params pop IP, CS; SP += Params (default: Params = 0)
    302 Dw ret Dw
    303 ret
    312 Dw ret far Dw
    313 ret far
    ENTER Locs, N push EBP;
    (sub EBP, 4; push [EBP]) N-1 times, if N > 0
    mov EBP, ESP
    (add EBP, 4*(N-1); push EBP), if N > 0
    sub ESP, Locs
    LEAVE mov ESP, EBP; pop EBP
    * 310 Dw Db enter Dw, Db
    * 311 leave

    SYSTEM CONTROL & MEMORY PROTECTION
    ----------------------------------
    BOUND A, AA if (A not in range AA[0]..AA[1]) INT 5
    ARPL L, E ZF = (L.RPL < E.RPL);
    if (ZF) L.RPL = E.RPL;
    * 142 xrm bound Rw, Ed
    $ 143 xrm arpl Es, Rw

    SLDT Sel Sel = LDTR
    STR Sel Sel = TR
    LLDT Sel LDTR = Sel
    LTR Sel TR = Sel
    VERR Sel ZF = (Sel is accessible and has read-access)
    VERW Sel ZF = (Sel is accessible and has write-access)
    LAR L, Sel ZF = (Sel is accessible);
    if (ZF) L = the access rights of Sel's descriptor.
    LSL L, Sel ZF = (Sel is accessible);
    if (ZF) L = the segment limit of Sel's descriptor.
    $ 017 000 x0m sldt Ew
    $ 017 000 x1m str Ew
    $ 017 000 x2m lldt Ew
    $ 017 000 x3m ltr Ew
    $ 017 000 x4m verr Ew
    $ 017 000 x5m verw Ew
    $ 017 002 xrm lar Rw, Ew
    $ 017 003 xrm lsl Rw, Ew

    SGDT Desc Desc = GDTR
    SIDT Desc Desc = IDTR
    LGDT Desc GDTR = Desc
    LIDT Desc IDTR = Desc
    $ 017 001 x0m sgdt Ep
    $ 017 001 x1m sidt Ep
    $ 017 001 x2m lgdt Ep
    $ 017 001 x3m lidt Ep

    SMSW L L = MSW ... note that MSW is CR0 bits 0-15.
    LMSW E MSW = E
    CLTS MSW.3 = 0 ... clears the Task Switched flag.
    $ 017 001 x4m smsw Ew
    $ 017 001 x6m lmsw Ew
    $ 017 006 clts

    INVD Invalidate internal cache.
    WBINVD Invalidate internal cache, after writing it back.
    INVLPD Ea Invalidate Ea's page.
    @ 017 010 invd
    @ 017 011 wbinvd
    @ 017 020 x7m invlpg Ea

    MOV Reg, SysReg
    MOV SysReg, Reg
    # 017 040 3nr mov Rd, CRn (n = 0-3)
    # 017 041 3nr mov Rd, DRn (n = 0-3, 6-7)
    # 017 042 3nr mov CRn, Rd (n = 0, 2-3)
    # 017 043 3nr mov DRn, Rd (n = 0-3, 6-7)
    # 017 044 3nr mov Rd, TRn (n = 6-7)
    # 017 046 3nr mov TRn, Rd (n = 6-7)

    CO-PROCESSOR ESCAPE SEQUENCE
    ----------------------------
    Comments:
    * This escape sequence is intended to be used with an external co-processor
    with the most common application being the 80x87 floating point unit.
    * Starting in the 80486, the floating point unit was made internal to the
    processor.

    ESC TL, Ea Escape, operation TL, address mode Ea.
    33T xLm esc TL Ea

    (6) Floating Point Operations
    The Floating Point unit consists of 8 internal registers arranged in a
    circular stack, and the Control Word (CW), Status Word (SW) and Tag Word (TW)
    registers. The floating point stack registers all store data in Real80
    format (described below).
    Operations are carried out on data in the following formats (low-order bits
    on right):

    INTEGER: 16/32/64 bits (Int16, Int32, Int64)
    BCD: (BCD80)
    S 0000000 D D D D D D D D D D D D D D D D D D
    S = 1-bit sign (1 = negative, 0 = positive)
    D = 4-bit digit (encodes digits 0-9).
    FLOATING POINT: 32/64/80 bits (Real32, Real64, Real80)
    S Exponent Mantissa
    S = 1-bit sign (1 = negative, 0 = positive)
    Exponent = 8/11/15 bit biased exponent
    Mantissa = 23/52/64 bit decimal fraction.
    The values of floating point numbers in each format are as follows:
    Real32: (-1)^S (1 + Mantissa)/2^23 x 2^(Exponent - 127)
    Real64: (-1)^S (1 + Mantissa)/2^52 x 2^(Exponent - 1023)
    Real80: (-1)^S Mantissa/2^63 x 2^(Exponent - 16383)

    The floatng point formats do not cover all the logical combination of binary
    0's and 1's, and the remaining combinations are defined for special purposes:

    Sign Exponent Mantissa Meaning
    S 0 0 0 ... 0 0 0 0 ... 0 0
    S 0 0 0 ... 0 ... 1 ... DENORMAL (Infinitesimal)
    S 1 1 1 ... 1 0 0 0 ... 0 INFINITY
    S 1 1 1 ... 1 0 ... 1 ... Signalling NaN (Not a Number)
    S 1 1 1 ... 1 1 ... Quiet NaN

    This is all IEEE standard format. Quiet NaN's are set by the FP Unit to
    indicate invalid operations.

    Notation:
    ST(n) -- the nth item below the stack top.
    ST ----- ST(0), the stack top.
    Int*, BCD*, Real* -- described above.

    All Int*, BCD*, and Real* operands are stored in memory and are encoded in
    the 80x86's current addressing mode (16 or 32 bit). All opcodes are
    listed in the format:

    T L xm for 8086 escape code 33T xLm

    Since only memory addresses are used in the operations, that frees up all
    the combinations xm where x = 3. These are generally used to encode the
    operations that do not involve memory addresses. In the following
    presentation where "xm" is listed generally, it is understood that x is not 3.

    The operations FENI, FDISI are specific to the 8887; FSETPM to the 80287
    and FUCOM*, FPREM1, and the trig. operations FSIN, FCOS, FSINCOS are all
    present only in the 80387 and after.

    DATA TRANSFER
    -------------
    Comments:
    * The followng table is used:
    P 0 2 3
    F-OP fld fst fstp
    I-OP fild fist fistp
    FLD Arg ST = (Real80)Arg
    FST Arg Arg = (typeof Arg)ST
    FSTP Arg Arg = (typeof Arg)ST; pop();
    FXCH Arg Arg <--> ST, with appropriate type conversions.
    1 P xm F-OP Real32
    3 P xm I-OP Int32
    5 P xm F-OP Real64
    7 P xm I-OP Int16
    3 5 xm fld Real80
    3 7 xm fstp Real80
    7 4 xm fbld BCD80
    7 5 xm fild Int64
    7 6 xm fbstp BCD80
    7 7 xm fistp Int64
    1 0 3m fld ST(m)
    1 1 3m fxch ST(m)
    5 2 3m fst ST(m)
    5 3 3m fstp ST(m)

    COMPARISON
    ----------
    Comments:
    * The followng table is used:
    P 2 3
    F-OP fcom fcomp
    I-OP ficom ficomp
    FCOM Arg cmp ST, Arg
    FCOMP Arg cmp ST, Arg; pop();
    0 P xm F-OP Real32
    2 P xm I-OP Int32
    4 P xm F-OP Real64
    6 P xm I-OP Int16
    0 P 3m F-OP ST(m)
    FCOMPP cmp ST, ST(1); pop(); pop();
    6 3 31 fcompp
    FTST cmp ST, 0.0
    1 4 34 ftst
    FXAM examine ST
    1 4 35 fxam
    FUCOM Arg unordered compare ST, Arg
    FUCOMP Arg unordered compare ST, Arg; pop();
    FUCOMPP Arg unordered compare ST, ST(1); pop(); pop();
    5 4 3m fucom ST(m)
    5 5 3m fucomp ST(m)
    2 5 31 fucompp

    ARITHMETIC OPERATIONS
    ---------------------
    Comments:
    * The followng table is used:
    P 0 1 4 5 6 7
    F-OP fadd fmul fsub fsubr fdiv fdivr
    I-OP fiadd fimul fisub fisubr fidiv fidivr
    P-OP faddp fmulp fsubp fsubrp fdivp fdivrp
    * Dest is ST and Src the listed operand except where noted below.
    FADD Arg Dest += Src
    FSUB Arg Dest += Src
    FSUBR Arg Dest = Src - Dest
    FMUL Arg Dest *= Src
    FDIV Arg Dest /= Src
    FDIVR Arg Dest = Src/Dest
    0 P xm F-OP Real32
    2 P xm I-OP Int32
    4 P xm F-OP Real64
    6 P xm I-OP Int16
    0 P 3m F-OP ST(m)
    4 P 3m F-OP ST(m) (Dest = ST(m), Src = ST)
    6 P 3m P-OP ST(m) (Dest = ST(m), Src = ST)

    CONSTANTS
    ---------
    FLD1 ST = 1.0
    FLDL2T ST = log_2(10)
    FLDL2E ST = log_2(e)
    FLDPI ST = pi
    FLDLG2 ST = log_10(2)
    FLDLN2 ST = ln(2)
    FLDZ ST = 0.0
    1 5 30 fld1
    1 5 31 fldl2t
    1 5 32 fldl2e
    1 5 33 fldpi
    1 5 34 fldlg2
    1 5 35 fldln2
    1 5 36 fldz

    BUILT-IN FUNCTIONS
    ------------------
    Comments:
    * The stack replacements entail pop()'s.
    FCHS ST = -ST
    FABS ST = |ST|
    F2XM1 ST = 2^ST - 1
    FYL2X Replace the stack: ST(1), ST -> ST(1)*log_2(ST)
    FPTAN Replace the stack: ST -> tan(ST), 1.0
    FPATAN Replace the stack: ST(1), ST -> atan(ST(1)/ST)
    FXTRACT Replace the stack: ST -> exponent(ST), mantissa(ST)
    FPREM1 ST = remainder(ST/ST(1)), IEEE consistent
    FPREM ST = remainder(ST/ST(1))
    FYL2XPI Replace the stack: ST(1), ST -> ST(1)*log_2(ST + 1)
    FSQRT ST = sqrt(ST)
    FSINCOS Replace the stack: ST -> sin(ST), cos(ST)
    FRNDINT ST = round(ST)
    FSCALE ST *= 2^(int)ST(1)
    FSIN ST = sin(ST)
    FCOS ST = cos(ST)
    1 4 30 fchs
    1 4 31 fabs
    1 6 30 f2xm1
    1 6 31 fyl2x
    1 6 32 fptan
    1 6 33 fpatan
    1 6 34 fxtract
    1 6 35 fprem1
    1 7 30 fprem
    1 7 31 fyl2xpi
    1 7 32 fsqrt
    1 7 33 fsincos
    1 7 34 frndint
    1 7 35 fscale
    1 7 36 fsin
    1 7 37 fcos

    CONTROL
    -------
    Comments:
    * The save and load operations for the environment and state are used
    primarily for multitasking applications where 2 or more processes are
    using the FP unit concurrently.
    FNOP Delay 1 cycle.
    FLDENV Arg Load FP environment from [Arg]
    FLDCW Arg CW = Arg
    FSTENV Arg Save FP environment to [Arg]
    FSTCW Arg Arg = CW
    FDECSTP TOP = (TOP - 1) mod 8
    FINCSTP TOP = (TOP + 1) mod 8
    FENI Enable interrupts (8087 only)
    FDISI Disable interrupts (8087 only)
    FCLEX Clear out FP exception flags
    FINIT Initialize FP registers
    FSETPM Enter Protected Mode (80287 only)
    FFREE ST(m) Mark register m as unused.
    FRSTOR Arg Restore FP state from [Arg]
    FSAVE Arg Save FP state to [Arg]
    FSTSW Arg Arg = SW
    1 2 30 fnop
    1 4 xm fldenv Ea
    1 5 xm fldcw Ea
    1 6 xm fstenv Ea
    1 7 xm fstcw Ea
    1 6 36 fdecstp
    1 6 37 fincstp
    3 4 30 feni
    3 4 31 fdisi
    3 4 32 fclex
    3 4 33 finit
    3 4 34 fsetpm
    5 0 3m ffree ST(m)
    5 4 xm frstor Ea
    5 6 xm fsave Ea
    5 7 xm fstsw Ea
    7 4 30 fstsw AX