💾 Archived View for spam.works › mirrors › textfiles › programming › 65xxx.drm captured on 2023-11-14 at 11:47:04.
⬅️ Previous capture (2023-06-16)
-=-=-=-=-=-=-
This is a very long article (36k) of interest to hardware
hackers, assembly language programmers, and machine
architects. It is a description of how I feel the
65xxx family should evolve. Don't count on anything
you read here. Nonetheless, you might find it interesting.
If you're not one of the aforementioned types, you may
want to skip the noise which follows...
The 65C816 Dream Machine
This essay is an attempt to vent my frustrations.
While the 65C816 chip is, without question, better than
the 6502 and 65c02 chips that preceded it, the 65c816
leaves a lot to be desired. Unless you count
microcontrollers like the 8048, F8, or 8051, I've never
encountered a chip as difficult to program in assembly
language as the 65c816. Those stupid M and X bits cause
so much trouble I wonder if they're worth the trouble of
using them. Attempting to use the 65c816 in native mode
while attempting to coexist with other 6502 routines
(requiring emulation mode) such as ProDOS 8 can really
push one's patience. But wait! There's a small chance
things can be improved. The WDM (William D. Mensch)
instruction is reserved by the Western Design Center for
instruction set expansion. While I'm sure Mr. Mensch has
other plans for this opcode, the following treatise
provides my views on how this single opcode should be
used.
The WDM opcode should be used in the next version of
the 65c816 (let's call it the 65c820, just to be amusing)
to change the instruction set. When the 65c820 resets,
it should come up in the 6502 emulation mode, just like
the 65c816 does now. The XCE instruction could be used
to switch to 65c816 mode just like the existing 65c816
part. The WDM opcode, which I'll call NAT (for NATive
mode) will be used to switch the processor to 65c820
native mode. Once in the 65c820 mode, the 65c820 takes
on a completely different character. The only bounds
I've placed on the new instruction set is that if you can
perform an operation with a single instruction on the
65c816, you can perform the same thing on the 65c820 with
a single instruction. All other aspects (including
timing and instruction size) can vary. I've also taken
some liberties with the way certain instructions affect
the flags. For the most part however, 65c816
instructions have an identical counterpart on the 65c820.
Design Issues: There are lot's of reasons for
designing a new instruction set. My criteria are as
follows:
1) The instruction set must mirror the philosophy of the
6500 family. A programmer experienced with the 6502
instruction set must feel comfortable with the 65c820
instruction set.
2) The new instruction set must support high level
language constructs better than the 6502 and 65c816
processors.
3) The new instruction set must be easy to learn and fun
to use.
4) We must remember that fancy instructions are very
difficult to implement in silicon. Hence super fancy
instructions which provide limited functionality must
be left out. For example, the 65c820 doesn't support
floating point instructions (although they could be
added via a coprocessor).
5) The only (commercially popular) computer system that
would ever use the 65c820 is an upgrade of the Apple
IIGS. Hence the instruction set should contain
instructions that enhance the operation of an Apple II
family machine.
6) The original 6502 instruction set was designed with a
small set of basic instructions complemented with a
large set of addressing modes. The 65c816 strayed
from this philosophy, the 65c820 returns to it.
Based on these design issues, I offer the following
machine; the 65c820:
_________________________________________________________
______________________
Programmer's Model:
The 65c820 will contain several additional registers,
above and beyond those available on the 65c816. All
registers are 16 bits. The register bank includes:
A, AX -- Accumulator and accumulator extension
X -- X index register
Y -- Y index register
F -- Stack frame pointer
S -- Stack pointer
D -- Direct page register
P -- Program status word
ABR -- Auxillary bank register
SBR -- Stack bank register
DBR -- Data bank register
PBR -- Program bank register
PC -- Program counter
LBound -- Low bounds register
HBound -- High bounds register
A, X, Y, S, D, & PC are mostly identical to their 65c816
counterparts. AX is the accumulator extension used by the
multiply and divide instructions. F is a special index
register, useful for accessing local variables and
parameters. P differs from the 65c816 version in that it
is 16-bits wide. Accessing the upper byte of this
register is a privileged operation (more on that later
on). DBR and PBR are similar to their 65c816 cousins,
except they are now 16-bits long and allow you to
position the program and data banks on any PAGE boundary
(rather than a bank [64K] boundary). ABR is an auxillary
data bank register. SBR lets you locate the stack
anywhere in the 16Mbyte address space. LBound and HBound
provide some rudimentary memory management functions.
All memory addresses are added to LBound to produce the
true physical address. If the result- ing address is
greater than HBound, an ABORT trap will be issued. This
allows you to load multiple programs into memory and
protect them from being walked on by other programs.
As I alluded to earlier, certain operations are
PRIVILEGED. The 65c820's program status word takes the
following form:
15 14 13 12 11 10 9 8 | 7 6 5 4 3 2 1 0 U/S
I M fpc * * * * | N V * * D dir Z C
The low order 8 bits are identical to the 6502's P
register except the B bit isn't present (it's not
required) and the I bit has been moved to bit 14. The
dir bit controls the direction of various string
instructions (ala 8086). The low order 8 bits are called
the USRPSW (user program status word). The upper 8 bits
are called tye SYSPSW (system program status word) and
can only be accessed while in the system mode. Bit 15
(U/S) is the user/supervisor bit. This bit determines
whether or not you are in the user or system (supervisor)
mode. Bit 14 is the interrupt disable bit. For
protection reasons, a user mode program cannot have
access to the interrupt disable bit (by turning off all
interrupts and not turning them back on, a user mode
program can cause all kinds of havoc). Bit 13 is the
memory management bit. If set, the LBound the HBound
registers determine the location and extent of the
logical address space. If clear, then the logical
address space and physical address space are the same.
The fpc bit determines if a floating point coprocessor is
installed. If not, the floating point expansion
instructions will cause an illegal instruction trap,
otherwise, the FP instructions will be routed to the
floating point coprocessor. The remaining bits in the P
register are reserved for future use.
Opcode Format:
The 65c820's instruction set is broken down into 32
classes. They are
0-MOV, 1-LEA, 2-LEAA, 3-LEAD, 4-LEAS, 5-XCHG, 6-
ADD, 7-ADC 8-SUB, 9-SBC, A-CMP, B-AND, C-OR,
D-XOR, E-ASH, F-LSH 10-ROT, 11-BIT, 12-ADDQ, 13-CMPQ,
14-exp, 15-exp, 16-exp, 17-exp 18-exp, 19-Scc, 1A-Ccc,
1B-Icc, 1C-Brnch,1D-Brnch,1E-exp, 1F-exp
"exp" refers to expansion.
The "typical" instruction format (for opcodes $00..$11)
is
15 14 13 12 11 10 9 8 | 7 6 5 4 3 2 1 0 a a
a a a a s d | r r r o o o o o
where a = addressing mode bits s = size
(0=byte, 1=word) d = direction (0=to addressing
mode loc, 1=from addressing mode loc) r = register
o = opcode (one of the group values above).
There are 64 possible addressing modes (since there are
six "a" bits). The register bits refer to the first
eight of these addressing modes (0..7).
0- A 10- d,F 20- d,X 30- d,Y 1- X
11- a,F 21- a,X 31- a,Y 2- Y 12-
n(d,F) 22- a,FX 32- a,FY 3- S 13- n(a,F)
23- l,X 33- l,Y 4- F 14- n[d,F] 24- (X)
34- (Y) 5- TOS 15- n[a,F] 25- (d,X) 35-
(d),Y 6- Imm 16- (d,F) 26- n(d,X) 36-
n(d),Y 7- d 17- [d,F] 27- [d,X] 37-
[d],Y 8- a 18- (a,F) 28- n[d,X] 38-
n[d],Y 9- l 19- [a,F] 29- n(d,FX) 39-
n(d,F),Y A- d,S 1A- P 2A- n[a,FX] 3A-
n[a,F],Y B- (d,S),Y 1B- D 2B- [a,FX] 3B-
[a,F],Y C- (d) 1C- ABR 2C- (a,X) 3C-
(d),Y+ D- [d] 1D- SBR 2D- n(a,X) 3D-
(d),-Y E- n(d) 1E- DBR 2E- [a,X] 3E-
[d],Y+ F- n[d] 1F- PBR 2F- n[a,X] 3F-
[d],-Y
where:
A, X, Y, S, F, P, D, ABR, SBR, DBR, and PBR are the
corresponding 65c820 registers. Imm refers to an
immediate operand. d refers to an eight-bit value,
usually (but not always) a direct page address. a refers
to a 16-bit absolute address. l refers to a 24-bit long
address n is a displacement of the form one byte, +/-
64 if the H.O. bit is zero. two bytes, H.O. byte
first, +/- 16383 if the H.O. bit is one.
All addressing mode containing F, FX, or FY are relative
to the SBR register. Any "d" address appearing in such an
addressing mode is simply an 8-bit displacement relative
to the frame pointer. FX means add F and X and use the
result as the frame pointer. FY is the same, but using
the Y register.
Y+ and -Y are autoincrement and autodecrement addressing
modes. For autoincrement, the Y register is incremented
after the value contained in Y is used. For auto-
decrement, the Y register is decremented before the value
is used.
Addressing modes of the form n[---]-- compute the
effective address specified by the indirect operation and
then add the specified offset to the effective address to
obtain the true effective address. For example, if Y
contains 5 and location $00 points at $1000 in the DBR,
then 4(0),y refers to location $1009.
The TOS addressing mode refers to the Top Of Stack, more
on this later.
General Instructions:
The general instructions (opcodes $00..$11) all take the
form:
Instr Source, Dest
Where Instr is the instruction mnemonic, Source is the
address of a source operand, and Dest is the address of a
destination operand. At least one of the two operands
must be a "register" addressing mode. The register
addressing modes are the first eight addressing modes
listed above. If the source operand is a register
addressing mode, then the direction bit in the
instruction is zero, otherwise it is one. If the source
addressing mode is the immediate addressing mode, the
flags are set by the result of the operation, but nothing
else is changed. For example, MOVB #0,#n sets the zero
flag since a zero bit is moved, but the zero isn't
actually moved anywhere. Note that a "B" or "W" suffix
is used on the mnemonics to specify the instruction size.
Three important register addressing mode greatly enhance
the capabilities of the 65c820 processor: the TOS, Imm,
and d register addressing modes. Since d is a register
addressing mode, any direct page memory location can be
used as a "register". This greatly enhances the
flexibility of the 65c820. This effectively gives you
256 registers to play around with.
The Imm addressing mode, since it is a register
addressing mode, lets you perform operations between any
register or memory location in the machine (addressable
by a single instruction) with an immediate operand. For
example, CMPB #5,2[3,D],Y is perfectly legal. For a
few instructions, immediate operands don't make much
sense, such instructions will cause an illegal
instruction trap (for example, you cannot load the
effective address of an immediate operand into a
register).
The TOS addressing mode is extremely powerful. If
you've looked ahead at the expansion instructions, you'd
have noticed that there aren't any specific push or pop
instructions (unless you count ENTER, EXIT, SAVE, and
RESTORE). The TOS addressing mode handles all of this
for you. You want to push the accumulator onto the
stack? No problem, MOVW A,TOS will do the job. You
want to pop the X register off of the stack? Use MOVW
TOS,X. You want to add the item on the top of stack to
the item below it on the stack (a VERY common operation
performed by compilers), just use ADDW TOS,TOS. This
instruction will pop two words off of the stack, add
them, and push their sum back onto the stack (leaving two
bytes on the stack rather than the original four). With
the TOS addressing mode, you can push (or pop) any value
anywhere in addressable memory onto the stack with a
single instruction.
Special (but not expansion) Instructions:
There are seven groups of instructions in this category:
ADDQ, CMPQ, Scc, Ccc, Icc, and the branch instructions.
ADDQ (add quick) appears in place of the ubiquitous INC
and DEC instructions. ADDQ lets you add a four-bit signed
value to any addressable item. The register bits, along
with the direction bit, let you specify a signed four-bit
value. This value is added to the specfied address. The
immediate operand MUST be the source operand.
The CMPQ (compare quick) is similar except a compare
operation is performed rather than an addition.
Furthermore, the immediate operand is the destination
operand rather than the source operand.
The Scc (set on condition), Ccc (clear on condition), and
Icc (invert on condition) instructions are used to set
boolean values based on the condition codes. These go
hand in hand with the branch instructions so I'll
describe them all at once.
There are 16 possible conditions, the register and
direction bits are used to specify the condition. These
conditions are
0- RA/A 4- HI 8- GT C- PL 1- CC/LO
5- LT 9- EQ D- VC 2- CS/HS 6- GE
A- NE E- VS 3- LS 7- LE B- MI
F- SR/N
LO (lower) = unsigned less than HS (higher/same) =
unsigned greater than or equal LS (lower/same) = unsigned
less than or equal HI (higher) = unsigned greater than LT
= signed less than GE = signed greater than or equal LE =
signed greater than or equal GT = signed greater than
The Scc, Ccc, and Icc instructions take the form:
Scc{b|w} #Imm, Dest Ccc{b|w} #Imm,
Dest Icc{b|w} #Imm, Dest
If the immediate operand is zero, then Scc will store a
one into the specified location if the condition code is
met, otherwise a one will be stored. Ccc does just the
opposite, it stores a zero if the condition is met, one
otherwise. The Icc instruction will complement the
specified location (logical NOT) if the condition code is
met. If the immediate operand is not zero, the the Scc
in- struction will set the specified bits in the
destination operand if the condition code is met, the Scc
instruction will have no effect if the condition is not
met. The Ccc instruction clears the specified bits in the
destination operand. The Icc instruction inverts the
specified bits. The destination bits are specified with
ones in the immediate operand. For example, SCS #$88,
$00 will set bits three and seven in memory location zero
if the carry flag is set, location $00 will be
unaffected if the carry flag is clear.
The SA/CA/IA (Set always, Clear always, Invert always)
instructions always perform the specified operation. The
SN/CN/IN (set never, clear never, invert never) behave as
though the condition code was not met.
The branch instructions are unusual compared to those
encountered thus far. The instruction is only one byte
long. It takes the form:
7 6 5 4 3 2 1 0 o o o --1C or 1D--
If the opcode is $1C, then the three "o" bits represent
condition codes 0..7 above. Note that the BRA instruction
uses opcode bits %000.
If the opcode is $1D, then the three "o" bits represent
condition codes 8..$F above. There is no BN instruction,
Opcode %111 is the BSR (branch to subroutine)
instruction.
Unlike the 65c816, branches are not limited to +/- 128
bytes. A displacement value, similar to the used by the
general addressing modes allows a one-byte displacement
of +/- 64 bytes or +/- 16383 bytes. More than enough for
most cases.
Math expansion instructions:
The math expansion instructions (opcode $14) use the
three register bits as an opcode expansion yield eight
additional instructions. The instruction format is
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 a a
a a a a s d o o o 1 0 1 0 0 where "aaaaaa"
is a general addressing mode, "s" is the size (B/W), "d"
is the direction (load/store), and "ooo" is the sub-
opcode, decode as follows:
0- MUL 1- DIV 2- MOD 3- REM 4- INDX 5- CHK
6- DIVS 7- FPexp
Sub-opcode 7 is reserved for floating point expansion via
a coprocessor. If the FPC bit in the SYSPSW is not set,
then executing this opcode will cause an illegal
instruction trap. If the FPC bit is set, then an
additional eight bit opcode follows this instruction.
This opcode value plus the physical address provided by
the addressing mode, bounds registers, and applicable
prefix(es) are passed along to the coprocessor.
All of these instructions use the 65c820 accumulator
as the register operand. MULW performs an unsigned 16x16
multiply, leaving the 32-bit result in A, AX. MULB
performs an unsigned 8x8 multiply, leaving the result in
A. DIVW performs an unsigned 32/16 division. The value
in (A,AX) is divided by the specified operand and the
quotient is left in (A,AX). DIVB divides the 16-bit
accumulator by an eight bit value, leaving the result in
A. DIVS{W|B} perform signed divisions. These two
instructions operate on the 16-bit accumulator or 8-bit
accumulator ONLY. The AX register is not used. MOD and
REM compute the modulo and remainder functions (MOD is
unsigned, REM is signed). Their register usage is
identical to DIV/DIVS. There is no need for a signed
multiply instruction since signed and unsigned
multiplication produces the same result, assuming you
ignore the value in AX.
The INDX and CHK instructions are used to perform
array computations. The operand of these two
instructions points at a pair of bytes or words. The
INDX instruction multiplies the accumulator by the first
value and then adds the second value to the accumulator.
The direction bit in the opcode is ignored. The INDX
instruction takes two forms: INDXB and INDXW.
The CHK instruction compares the value in the
accumulator against the first and second values. If the
accumulator lies within these two values (inclusive) then
the overflow flag is cleared. If the accumulator is
outside the range of these two values, then the overflow
flag is set. The direction flag in the opcode is used to
determine whether a signed or unsigned comparison is
used. The CHK instruction takes four forms: CHKSB, CHKSW,
CHKUB, and CHKUW. The "U" and "S" specify unsigned or
signed.
String expansion instructions:
Opcode $15 is used for string operations. The 65c820
processor provides four basic string operations: MOVS
(move), CMPS (compare), XLATS (translate), and FILLS
(fill). The instruction format is as follows:
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 s s
s d d d l l l o o 1 0 1 0 1
where "sss" is the source address, "ddd" is the
destination address, "lll" is the length address, and
"oo" is the opcode. Since all of the addresses are three
bits, they must be register addresses. The source
address is a sixteen bit value taken from one of the
register addressing modes. The sixteen-bit value
obtained at said address is the start of the string
within the data bank (i.e., relative to the DBR
register). The destination address is also a sixteen-bit
register addressing mode value, specifying the start of
the destination address within the auxillary bank (i.e.,
relative to the ABR register). The length value is a
sixteen-bit quantity obtained directly from the register
addressing mode location. Prefix bytes (described later
on) are not allowed in front of a string instruction.
Opcode assignments:
0- MOVS 1- CMPS 2- XLATS 3- FILLS
The direction of the string operation is specified by
the "dir" bit in the USRPSW register. If the bit is
clear, then the source and destination operands are
incremented after each string operation. If the "dir"
bit is clear, then these operands are decremented after
each string operation.
The string instructions take the form:
MNEMONIC src, dest, len
where src, dest, and len are any of A, X, Y, S, F, TOS,
#value, or a direct page address. For the these
operands, the sixteen bit value specified by one of these
addresses is used, relative to the DBR, as the address
(or length) of the specified block. An absolute address
can be specified by an immediate operand. The direct
page address is the address of the 16-bit value within
the direct page, it does not mean that the address of the
block is that address in the direct page. Same with the
TOS, the value on the top of stack contains the address,
the top of stack is not the block itself. The len
operand is always a byte count. Unless an immediate
operand is specified, the operands are always updated to
reflect their new value at the termination of the block
operation.
The MOVS instruction is used to move a string of
bytes from one location to another. A block of "len"
bytes specified by DBR/src is moved to ABR/dest.
The MOVS operation is an example of an instruction
that does not exactly mirror its 65c816 counterpart. It
may take two (or more) instructions to perform the same
operation as the 65c816 MVN and MVP instructions, since
the direction flag may require adjustment before
performing a MOVS instruction. Futhermore, the ABR and
DBR registers may need adjustment before and after the
MOVS instruction to simulate the MVN and MVP
instructions. Finally, the actual count is specified by
length, not count-1 (as on the 65c816), so this may
require some adjustment if you are translating 65c816
code instruction by instruction.
Example:
MVN 0,1
can be simulated by
MOVW #0,DBR MOVW #1,ABR ADDQ #1,
A ;Since MVN assumes A contains count-1 MOVS
X,Y,A MOVW #1,DBR
The CMPS operation compares the two specified
strings. It does a byte by byte comparison until length
bytes are compared or a character in the source string is
not equal to the corresponding character in the
destination string. The condition codes are set to
reflect the ordinality of the two strings (so you can use
any of the branch, Scc, Ccc, or Icc instructions to test
the results). If the z flag is returned set, then the two
strings are equal (through the specified length),
otherwise the source and destination operands are updated
to point at the differing chars and the length operand is
updated to show the number of character processed thus
far (assuming, of course, that these operands weren't
immediate, in which case they would be ignored).
The XLATS instruction is used to translate values in
a string. The source operand points at a table in the
DBR. Each character in the dest string is used as an
index into this table and the value fetched from the
table is stored over the original character in the
destination string.
The FILLS instruction is used to initialize a string
with a fixed value. The source operand is an eight-bit
value. It is stored in successive locations at ABR/dest
for len bytes. If an immediate value is specified, a
sixteen-bit value is encoded into the instruction, but
only the L.O. eight bits are used.
Single byte expansion instructions:
These instructions take the form:
7 6 5 4 3 2 1 0 o o o 1 0 1 1 0
Where "ooo" is decoded as:
0- NOP 1- COP 2- BRK 3- SVC 4- RTS 5- RTL 6- RTI 7- EXIT
SVC is the "supervisor call" instruction. Its
intended use is for making operating system calls. It is
similar in function to the COP instruction.
EXIT is used to deallocate local variables in a
procedure. It undoes the actions of the ENTER
instruction. Basically it performs the following
operations:
MOV F,S MOV TOS, F
The remaining instructions in this group are
identical to their 65c816 counterparts, so they don't
require any futher elaboration.
Single byte w/displacement expansion instructions:
These instructions take the form:
7 6 5 4 3 2 1 0 o o o 1 0 1 1 1
Where "ooo" is decoded as:
0- SAVE n 1- RESTORE n 2- reserved 3- reserved 4- RTS n
5- RTL n 6- ADJSP n 7- ENTER n
The "n" value immediately following these
instructions is a displacement value. If bit seven of
the first byte following the opcode is zero, then the
remaining six bits are used to specify a signed value in
the range +/- 64. If bit seven is one, then the
following 15 bits are used to specify a value in the
range +/-16383. Except possibly for ADJSP, none of these
instructions should ever require more than a single byte
displacement.
SAVE is used to quickly push registers from the set
[A,AX,X,Y,F,D,P] onto the stack. The instruction is
followed by a single byte with bits 0..6 cor- responding
to these registers. Bit seven must always be zero.
RESTORE does just the opposite of SAVE, it pops the
specified registers off of the stack.
RTS n and RTL n perform the specified return from
subroutine operations and then add the specified
displacement to the stack pointer after the return
address has been popped. This provides a convenient
mechanism whereby parameters can be removed from the
stack.
The ADJSP n instruction adds the displacement value
to the stack pointer. This is a shorter version of the
ADD #value,S instruction. A special case was created for
this instruction because it gets used all the time in
languages like "C" or "SDL/65" which allow a variable
number of parameters.
The ENTER n instruction is used to set up an
activation record when a procedure is initially entered.
It performs the following operations:
MOVW F,TOS MOVW S, F ADJSP n
The EXIT instruction can be used to undo the effects of
this instruction.
Prefix expansion instructions:
These instructions take the form:
7 6 5 4 3 2 1 0 o o o 1 1 0 0 0
where "ooo" is decoded as:
0-ABR prefix 1-SBR prefix 2-PBR prefix 3-word index
prefix 4-dword index prefix 5-qword index prefix 6-
XBA/SWA 7-EMU
XBA and EMU aren't true prefix bytes, they're just
single byte instructions that didn't conveniently fit
anywhere else. So I'll describe them first. XBA is
identical to its 65c816 counterpart, it swaps the bytes
in the accumulator. EMU switches from 65c820 native mode
to 65c816 emulation mode. EMU is a privileged
instruction and will cause a privileged instruction trap
if executed from the user mode.
The first three prefix bytes are used to modify the
bank used for data accesses. Addressing modes that
normally access memory through the data bank register
(which are all memory references except direct, long,
TOS, and those involving F) can be "tweaked" to access
memory through the auxillary, stack, or program bank
registers by prefixing the address with the appropriate
prefix. For example,
MOVW #275, ABR:$1000
stores 275 into location $1000 in the auxillary bank
register rather than the data bank register. Indirect
addresses of the form (a,X) and n(a,X) present a minor
problem. Does the prefix specify the bank address of the
absolute operand or the effective address? I've opted
for requiring that the absolute operand reside in the
data bank and the prefix byte determines the effective
address bank.
Any addressing mode utilitizing the frame pointer
register (F) is always relative to the stack bank
register. Prefixes are only allowed for the following
frame-based addressing modes: n(d,F), n(a,F), (d,F),
(a,F), n(d,FX), and n(d,F),Y. The indirect address
always comes out of the stack bank, the prefix applies to
the computed effective address.
Although the ABR:/SBR:/PBR: lexemes immediately
precede the address expression to which they apply (on
the source line), in the object code, the prefix byte
always precedes the instruction to which the prefix
applies. If more than one prefix byte precedes an
instruction, only the last one is used. If a prefix byte
precedes an instruction to which the prefix doesn't make
sense (a branch, for example), then the prefix byte is
ignored. Finally, the prefix byte will be ignored if
there isn't an applicable addressing mode in the current
instruction. E.G.: byt $18 ;ABR prefix
byte MOVW A,X ;ABR prefix has no meaning
here.
Three additional prefix bytes apply to the X and Y
index registers. These are the word index prefix, dword
index prefix, and qword index prefix. These prefix bytes
provide scaled indexed addressing modes for the 65c820.
Without one of these prefixes, the X and Y registers are
always byte offsets. That is, when used as an index
register, the contents of X or Y is added directly to the
effective address being computed. When accessing words,
pointers (double words), or eight byte values (e.g.,
floating point) you have to manually adjust the index
registers by a factor of 2, 4, or 8. The scaled index
addressing prefix bytes let you avoid this problem. The
word prefix multiplies the X or Y register value by two
before using it in the effective address computation.
Likewise, the dword and qword prefixes multiply X or Y by
4 or 8 before using the value. In the source code, these
prefix bytes are specified by the ":W", ":D", and ":Q"
suffixes:
MOVW A,LBL,X:W MOVW
$0,(PTR),Y:D MOVW $2, 2(PTR),Y:D
MOVW F,(TBL,X:W) MOVW A,LBL,Y:Q
If multiple prefixes appear, only the last one is used.
If the prefix doesn't apply to the next instruction, it
is ignored.
Single operand expansion instructions:
The $1E expansion instructions are dedicated to
instructions which require a single operand. The format
for the opcodes is as follows:
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 a a
a a a a s o o o o 1 1 1 1 0
where "aaaaaa" is a general addressing mode, "s" is the
size (B/W), and "oooo" is one of the following opcodes:
0- NOT 8- LAX (load AX register) 1-
NEG 9- SAX (store AX register) 2- ABS
A- XAX (exchange AX register) 3- BOOL (0->0 else->1)
B- LLB (load LBound register) 4- SEX
C- LHB (load HBound register) 5- ZEX
D- SLB (store LBound register) 6- JMP
E- SHB (store HBound register) 7- JSR
F- VAL (validate memory location)
All of these instructions are followed by a single
general address expression. Immediate operands are not
allowed for any of these instructions.
NOT- logically compliments the specified value. NEG-
takes the two's complement of the specified value. ABS-
takes the absolute value of the specified location. BOOL-
If the specified location is not zero, a one is stored
into it.
SEX- (that's sign extension, not what you think). SEXB
checks the high order bit of the specified byte and
copies it into the H.O. byte of the corresponding
address. For example, if X contains $0082 then SEXB X
will store $FF82 into X. If X contains $0002, then SEXB X
will store $0002 into X. SEXW sign extends the
specified location into the AX register.
ZEX- zero extends the specified value. ZEXW simply
stores a zero into AX. ZEXB stores a zero into the H.O.
byte of the specified word.
JMP and JSR are like their 65c816 counterparts except any
valid addressing mode can be used. Note that, unlike
most other instructions, the result is assumed to be in
the current program bank unless a long addressing mode is
specified.
LAX, SAX, and XAX allow you to load, store, and exchange
the contents of the AX register. Note that these three
instructions plus SEX, ZEX, MUL, DIV, and MOD are the
only instructions that deal with the AX register.
LLB, LHB, SLB, and SHB let you load and save the contents
of the bounds registers. These are privileged
instructions which will cause a privilege trap if
executed from the user mode.
VAL- This instruction is used to validate a memory
location. That is, it tests the specified memory
location to see if it lies within the range specified by
the bounds register. The address is a physical address,
not a translated address. The overflow flag is set if a
bounds violation would occur. Note that the M bit in the
SYSPSW need not contain a particular value when using
this instruction. This is a privileged instruction which
will cause a privilege violation if executed in the user
mode.
BIT expansion instructions:
These instructions take the form:
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 a a
a a a a s o o o o 1 1 1 1 1
"aaaaaa" is the destination addressing mode. "s" is the
size (applicable only to MAND, MOR, and MXOR). "oooo" is
the sub-opcode, decoded as:
0- INS dest, start, len 1- EXT dest, start, len 2- FFS
dest, start, len 3- FFC dest, start, len 4- MAND dest,
mask 5- MOR dest, mask 6- MXOR dest, mask
7..F- reserved.
INS is used to insert a value into a bit field. The
value in the accumulator is shifted to the left "start"
bits and the the "len" following bits are stored into
the specified memory location. For example, if memory
location $00 contains $F0 and the accumulator contains
$3, then INS $00,2,4 would leave location $00 containing
$CC. Note that you needn't specify byte or word size as
this is intrinsic from the length.
EXT- extracts a bit field from some location and stores
the right justified value into the accumulator (zeroing
out any unused bits). For example, if memory location
$00 contains $CC and the accumulator contains $FFFF, then
EXT $0,2,4 would leave the accumulator containing 3 and
location $00 containing $CC.
FFS finds the first set bit in the specified location.
The bit position is returned in the accumulator. If
there were no set bits, the accumulator contains "len"+1.
FFC finds the first clear bit in a manner identical to
FFS.
Some notes: These four instructions are followed by a
single byte. The low order four bits contain the start
value, the high order four bits contain the length-1.
"start" + "len" must always be less than or equal to 15.
FFS and FFC use the direction bit in the USRPSW to
determine which way to progress in the bit field when
searching for the set or clear bit.
The MAND, MOR, and MXOR (masked AND, OR, and XOR) will
AND, OR, or XOR the accumulator into the specified memory
location. The difference between these three
instructions and the standard AND, OR, and XOR
instructions is that they are followed by a byte or word
(depending on the instruction size) which contains a mask
for the operation. Wherever a one bit appears in the
mask, the logical operation will take place, wherever a
zero bit appears, the destination's bits will be
unaffected.
_________________________________________________________
__________________
That wraps up my proposed instruction set for the 65c820.
I'll be happy to discuss my design decisions with anyone
who's interested. The next step is to try and convince
someone to actually build this thing! In the mean time,
I might try writing an interpreter and assembler for it.
By the way. Many of you have probably recognized certain
instructions from this processor or that processor
sprinkled throughout. To set the record straight, most
of my ideas have come from my own frustrations with the
65c816, the 8086 family, and the National Semiconductor
32000 family. Despite that fact that a lot of you think
that Intel's parts stink because they're used by IBM,
don't let that prejudice you against many of the design
issues here. The 8086 does have a resonable
archetecture, given the compromises it had to face. It's
certainly better than the 65c816. I've incorporated a
lot of the better ideas (like segment prefixes) into the
design of the 65c820. Once again, don't downplay these
powerful features just because you don't like IBM.