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. *** Randy Hyde