SWEET 16: A Pseudo 16 Bit Microprocessor by Steve Wozniak Description: ------------ While writing APPLE BASIC for a 6502 microprocessor, I repeatedly encountered a variant of MURPHY'S LAW. Briefly stated, any routine operating on 16-bit data will require at least twice the code that it should. Programs making extensive use of 16-bit pointers (such as compilers, editors, and assemblers) are included in this category. In my case, even the addition of a few double-byte instructions to the 6502 would have only slightly alleviated the problem. What I really needed was a 6502/RCA 1800 hybrid - an abundance of 16-bit registers and excellent pointer capability. My solution was to implement a non-existant (meta) 16-bit processor in software, interpreter style, which I call SWEET 16. SWEET 16 is based on sixteen 16-bit registers (R0-15), which are actually 32 memory locations. R0 doubles as the SWEET 16 accumulator (ACC), R15 as the program counter (PC), and R14 as the status register. R13 holds compare instruction results and R12 is the subroutine return stack pointer if SWEET 16 subroutines are used. All other SWEET 16 registers are at the user's unrestricted disposal. SWEET 16 instructions fall into register and non-register categories. The register ops specify one of the sixteen registers to be used as either a data element or a pointer to data in memory, depending on the specific instruction. For example INR R5 uses R5 as data and ST @R7 uses R7 as a pointer to data in memory. Except for the SET instruction, register ops take one byte of code each. The non-register ops are primarily 6502 style branches with the second byte specifying a +/-127 byte displacement relative to the address of the following instruction. Providing that the prior register op result meets a specified branch condition, the displacement is added to the SWEET 16 PC, effecting a branch. SWEET 16 is intended as a 6502 enhancement package, not a stand alone processor. A 6502 program switches to SWEET 16 mode with a subroutine call and subsequent code is interpreted as SWEET 16 instructions. The nonregister op RTN returns the user program to 6502 mode after restoring the internal register contents (A, X, Y, P, and S). The following example illustrates how to use SWEET 16. 300 B9 00 02 LDA IN,Y ;get a char 303 C9 CD CMP #"M" ;"M" for move 305 D0 09 BNE NOMOVE ;No. Skip move 307 20 89 F6 JSR SW16 ;Yes, call SWEET 16 30A 41 MLOOP LD @R1 ;R1 holds source 30B 52 ST @R2 ;R2 holds dest. addr. 30C F3 DCR R3 ;Decr. length 30D 07 FB BNZ MLOOP ;Loop until done 30F 00 RTN ;Return to 6502 mode. 310 C9 C5 NOMOVE CMP #"E" ;"E" char? 312 D0 13 BEQ EXIT ;Yes, exit 314 C8 INY ;No, cont. NOTE: Registers A, X, Y, P, and S are not disturbed by SWEET 16. Instruction Descriptions: ------------------------- The SWEET 16 opcode listing is short and uncomplicated. Excepting relative branch displacements, hand assembly is trivial. All register opcodes are formed by combining two Hex digits, one for the opcode and one to specify a register. For example, opcodes 15 and 45 both specify register R5 while codes 23, 27, and 29 are all ST ops. Most register ops are assigned in complementary pairs to facilitate remembering them. Therefore, LD ans ST are opcodes 2N and 3N respectively, while LD @ and ST @ are codes 4N and 5N. Opcodes 0 to C (Hex) are assigned to the thirteen non-register ops. Except for RTN (opcode 0), BK (0A), and RS (0B), the non register ops are 6502 style branches. The second byte of a branch instruction contains a +/-127 byte displacement value (in two's complement form) relative to the address of the instruction immediately following the branch. If a specified branch condition is met by the prior register op result, the displacement is added to the PC effecting a branch. Except for the BR (Branch always) and BS (Branch to a Subroutine), the branch opcodes are assigned in complementary pairs, rendering them easily remembered for hand coding. For example, Branch if Plus and Branch if Minus are opcodes 4 and 5 while Branch if Zero and Branch if NonZero are opcodes 6 and 7. SWEET 16 Opcode Summary: ------------------------ Register OPS- 1n SET Rn Constant (Set) 2n LD Rn (Load) 3n ST Rn (Store) 4n LD @Rn (Load Indirect) 5n ST @Rn (Store Indirect) 6n LDD @Rn (Load Double Indirect) 7n STD @Rn (Store Double Indirect) 8n POP @Rn (Pop Indirect) 9n STP @Rn (Store POP Indirect) An ADD Rn (Add) Bn SUB Rn (Sub) Cn POPD @Rn (Pop Double Indirect) Dn CPR Rn (Compare) En INR Rn (Increment) Fn DCR Rn (Decrement) Non-register OPS- 00 RTN (Return to 6502 mode) 01 BR ea (Branch always) 02 BNC ea (Branch if No Carry) 03 BC ea (Branch if Carry) 04 BP ea (Branch if Plus) 05 BM ea (Branch if Minus) 06 BZ ea (Branch if Zero) 07 BNZ ea (Branch if NonZero) 08 BM1 ea (Branch if Minus 1) 09 BNM1 ea (Branch if Not Minus 1) 0A BK (Break) 0B RS (Return from Subroutine) 0C BS ea (Branch to Subroutine) 0D (Unassigned) 0E (Unassigned) 0F (Unassigned) Register Instructions: ---------------------- SET: SET Rn,Constant [ 1n Low High ] The 2-byte constant is loaded into Rn (n=0 to F, Hex) and branch conditions set accordingly. The carry is cleared. EXAMPLE: 15 34 A0 SET R5 $A034 ;R5 now contains $A034 LOAD: LD Rn [ 2n ] The ACC (R0) is loaded from Rn and branch conditions set according to the data transferred. The carry is cleared and contents of Rn are not disturbed. EXAMPLE: 15 34 A0 SET R5 $A034 25 LD R5 ;ACC now contains $A034 STORE: ST Rn [ 3n ] The ACC is stored into Rn and branch conditions set according to the data transferred. The carry is cleared and the ACC contents are not disturbed. EXAMPLE: 25 LD R5 ;Copy the contents 36 ST R6 ;of R5 to R6 LOAD INDIRECT: LD @Rn [ 4n ] The low-order ACC byte is loaded from the memory location whose address resides in Rn and the high-order ACC byte is cleared. Branch conditions reflect the final ACC contents which will always be positive and never minus 1. The carry is cleared. After the transfer, Rn is incremented by 1. EXAMPLE 15 34 A0 SET R5 $A034 45 LD @R5 ;ACC is loaded from memory ;location $A034 ;R5 is incr to $A035 STORE INDIRECT: ST @Rn [ 5n ] The low-order ACC byte is stored into the memory location whose address resides in Rn. Branch conditions reflect the 2-byte ACC contents. The carry is cleared. After the transfer Rn is incremented by 1. EXAMPLE: 15 34 A0 SET R5 $A034 ;Load pointers R5, R6 with 16 22 90 SET R6 $9022 ;$A034 and $9022 45 LD @R5 ;Move byte from $A034 to $9022 56 ST @R6 ;Both ptrs are incremented LOAD DOUBLE-BYTE INDIRECT: LDD @Rn [ 6n ] The low order ACC byte is loaded from memory location whose address resides in Rn, and Rn is then incremented by 1. The high order ACC byte is loaded from the memory location whose address resides in the incremented Rn, and Rn is again incremented by 1. Branch conditions reflect the final ACC contents. The carry is cleared. EXAMPLE: 15 34 A0 SET R5 $A034 ;The low-order ACC byte is loaded 65 LDD @R6 ;from $A034, high-order from ;$A035, R5 is incr to $A036 STORE DOUBLE-BYTE INDIRECT: STD @Rn [ 7n ] The low-order ACC byte is stored into memory location whose address resides in Rn, and Rn is the incremented by 1. The high-order ACC byte is stored into the memory location whose address resides in the incremented Rn, and Rn is again incremented by 1. Branch conditions reflect the ACC contents which are not disturbed. The carry is cleared. EXAMPLE: 15 34 A0 SET R5 $A034 ;Load pointers R5, R6 16 22 90 SET R6 $9022 ;with $A034 and $9022 65 LDD @R5 ;Move double byte from 76 STD @R6 ;$A034-35 to $9022-23. ;Both pointers incremented by 2. POP INDIRECT: POP @Rn [ 8n ] The low-order ACC byte is loaded from the memory location whose address resides in Rn after Rn is decremented by 1, and the high order ACC byte is cleared. Branch conditions reflect the final 2-byte ACC contents which will always be positive and never minus one. The carry is cleared. Because Rn is decremented prior to loading the ACC, single byte stacks may be implemented with the ST @Rn and POP @Rn ops (Rn is the stack pointer). EXAMPLE: 15 34 A0 SET R5 $A034 ;Init stack pointer 10 04 00 SET R0 4 ;Load 4 into ACC 55 ST @R5 ;Push 4 onto stack 10 05 00 SET R0 5 ;Load 5 into ACC 55 ST @R5 ;Push 5 onto stack 10 06 00 SET R0 6 ;Load 6 into ACC 55 ST @R5 ;Push 6 onto stack 85 POP @R5 ;Pop 6 off stack into ACC 85 POP @R5 ;Pop 5 off stack 85 POP @R5 ;Pop 4 off stack STORE POP INDIRECT: STP @Rn [ 9n ] The low-order ACC byte is stored into the memory location whose address resides in Rn after Rn is decremented by 1. Branch conditions will reflect the 2-byte ACC contents which are not modified. STP @Rn and POP @Rn are used together to move data blocks beginning at the greatest address and working down. Additionally, single-byte stacks may be implemented with the STP @Rn ops. EXAMPLE: 14 34 A0 SET R4 $A034 ;Init pointers 15 22 90 SET R5 $9022 84 POP @R4 ;Move byte from 95 STP @R5 ;$A033 to $9021 84 POP @R4 ;Move byte from 95 STP @R5 ;$A032 to $9020 ADD: ADD Rn [ An ] The contents of Rn are added to the contents of ACC (R0), and the low-order 16 bits of the sum restored in ACC. the 17th sum bit becomes the carry and the other branch conditions reflect the final ACC contents. EXAMPLE: 10 34 76 SET R0 $7634 ;Init R0 (ACC) and R1 11 27 42 SET R1 $4227 A1 ADD R1 ;Add R1 (sum=B85B, C clear) A0 ADD R0 ;Double ACC (R0) to $70B6 ;with carry set. SUBTRACT: SUB Rn [ Bn ] The contents of Rn are subtracted from the ACC contents by performing a two's complement addition: ACC = ACC + Rn + 1 The low order 16 bits of the subtraction are restored in the ACC, the 17th sum bit becomes the carry and other branch conditions reflect the final ACC contents. If the 16-bit unsigned ACC contents are greater than or equal to the 16-bit unsigned Rn contents, then the carry is set, otherwise it is cleared. Rn is not disturbed. EXAMPLE: 10 34 76 SET R0 $7634 ;Init R0 (ACC) 11 27 42 SET R1 $4227 ;and R1 B1 SUB R1 ;subtract R1 ;(diff=$340D with c set) B0 SUB R0 ;clears ACC. (R0) POP DOUBLE-BYTE INDIRECT: POPD @Rn [ Cn ] Rn is decremented by 1 and the high-order ACC byte is loaded from the memory location whose address now resides in Rn. Rn is again decremented by 1 and the low-order ACC byte is loaded from the corresponding memory location. Branch conditions reflect the final ACC contents. The carry is cleared. Because Rn is decremented prior to loading each of the ACC halves, double-byte stacks may be implemented with the STD @Rn and POPD @Rn ops (Rn is the stack pointer). EXAMPLE: 15 34 A0 SET R5 $A034 ;Init stack pointer 10 12 AA SET R0 $AA12 ;Load $AA12 into ACC 75 STD @R5 ;Push $AA12 onto stack 10 34 BB SET R0 $BB34 ;Load $BB34 into ACC 75 STD @R5 ;Push $BB34 onto stack C5 POPD @R5 ;Pop $BB34 off stack C5 POPD @R5 ;Pop $AA12 off stack COMPARE: CPR Rn [ Dn ] The ACC (R0) contents are compared to Rn by performing the 16 bit binary subtraction ACC-Rn and storing the low order 16 difference bits in R13 for subsequent branch tests. If the 16 bit unsigned ACC contents are greater than or equal to the 16 bit unsigned Rn contents, then the carry is set, otherwise it is cleared. No other registers, including ACC and Rn, are disturbed. EXAMPLE: 15 34 A0 SET R5 $A034 ;Pointer to memory 16 BF A0 SET R6 $A0BF ;Limit address B0 LOOP1 SUB R0 ;Zero data 75 STD @R5 ;clear 2 locations ;increment R5 by 2 25 LD R5 ;Compare pointer R5 D6 CPR R6 ;to limit R6 02 FA BNC LOOP1 ;loop if C clear INCREMENT: INR Rn [ En ] The contents of Rn are incremented by 1. The carry is cleared and other branch conditions reflect the incremented value. EXAMPLE: 15 34 A0 SET R5 $A034 ;(Pointer) B0 SUB R0 ;Zero to R0 55 ST @R5 ;Clr Location $A034 E5 INR R5 ;Incr R5 to $A036 55 ST @R5 ;Clrs location $A036 ;(not $A035) DECREMENT: DCR Rn [ Fn ] The contents of Rn are decremented by 1. The carry is cleared and other branch conditions reflect the decremented value. EXAMPLE: (Clear 9 bytes beginning at location A034) 15 34 A0 SET R5 $A034 ;Init pointer 14 09 00 SET R4 9 ;Init counter B0 SUB R0 ;Zero ACC 55 LOOP2 ST @R5 ;Clear a mem byte F4 DCR R4 ;Decrement count 07 FC BNZ LOOP2 ;Loop until Zero Non-Register Instructions: -------------------------- RETURN TO 6502 MODE: RTN 00 Control is returned to the 6502 and program execution continues at the location immediately following the RTN instruction. the 6502 registers and status conditions are restored to their original contents (prior to entering SWEET 16 mode). BRANCH ALWAYS: BR ea [ 01 d ] An effective address (ea) is calculated by adding the signed displacement byte (d) to the PC. The PC contains the address of the instruction immediately following the BR, or the address of the BR op plus 2. The displacement is a signed two's complement value from -128 to +127. Branch conditions are not changed. NOTE: The effective address calculation is identical to that for 6502 relative branches. The Hex add & Subtract features of the APPLE ][ monitor may be used to calculate displacements. d = $80 ea = PC + 2 - 128 d = $81 ea = PC + 2 - 127 d = $FF ea = PC + 2 - 1 d = $00 ea = PC + 2 + 0 d = $01 ea = PC + 2 + 1 d = $7E ea = PC + 2 + 126 d = $7F ea = PC + 2 + 127 EXAMPLE: $300: 01 50 BR $352 BRANCH IF NO CARRY: BNC ea [ 02 d ] A branch to the effective address is taken only is the carry is clear, otherwise execution resumes as normal with the next instruction. Branch conditions are not changed. BRANCH IF CARRY SET: BC ea [ 03 d ] A branch is effected only if the carry is set. Branch conditions are not changed. BRANCH IF PLUS: BP ea [ 04 d ] A branch is effected only if the prior 'result' (or most recently transferred dat) was positive. Branch conditions are not changed. EXAMPLE: (Clear mem from A034 to A03F) 15 34 A0 SET R5 $A034 ;Init pointer 14 3F A0 SET R4 $A03F ;Init limit B0 LOOP3 SUB R0 55 ST @R5 ;Clear mem byte ;Increment R5 24 LD R4 ;Compare limit D5 CPR R5 ;to pointer 04 FA BP LOOP3 ;Loop until done BRANCH IF MINUS: BM ea [ 05 d ] A branch is effected only if prior 'result' was minus (negative, MSB = 1). Branch conditions are not changed. BRANCH IF ZERO: BZ ea [ 06 d ] A Branch is effected only if the prior 'result' was zero. Branch conditions are not changed. BRANCH IF NONZERO BNZ ea [ 07 d ] A branch is effected only if the priot 'result' was non-zero Branch conditions are not changed. BRANCH IF MINUS ONE BM1 ea [ 08 d ] A branch is effected only if the prior 'result' was minus one ($FFFF Hex). Branch conditions are not changed. BRANCH IF NOT MINUS ONE BNM1 ea [ 09 d ] A branch effected only if the prior 'result' was not minus 1. Branch conditions are not changed. BREAK: BK [ 0A ] A 6502 BRK (break) instruction is executed. SWEET 16 may be re-entered non destructively at SW16d after correcting the stack pointer to its value prior to executing the BRK. RETURN FROM SWEET 16 SUBROUTINE: RS [ 0B ] RS terminates execution of a SWEET 16 subroutine and returns to the SWEET 16 calling program which resumes execution (in SWEET 16 mode). R12, which is the SWEET 16 subroutine return stack pointer, is decremented twice. Branch conditions are not changed. BRANCH TO SWEET 16 SUBROUTINE: BS ea [ 0c d ] A branch to the effective address (PC + 2 + d) is taken and execution is resumed in SWEET 16 mode. The current PC is pushed onto a SWEET 16 subroutine return address stack whose pointer is R12, and R12 is incremented by 2. The carry is cleared and branch conditions set to indicate the current ACC contents. EXAMPLE: (Calling a 'memory move' subroutine to move A034-A03B to 3000-3007) 15 34 A0 SET R5 $A034 ;Init pointer 1 14 3B A0 SET R4 $A03B ;Init limit 1 16 00 30 SET R6 $3000 ;Init pointer 2 0C 15 BS MOVE ;Call move subroutine 45 MOVE LD @R5 ;Move one 56 ST @R6 ;byte 24 LD R4 D5 CPR R5 ;Test if done 04 FA BP MOVE 0B RS ;Return Theory of Operation: -------------------- SWEET 16 execution mode begins with a subroutine call to SW16. All 6502 registers are saved at this time, to be restored when a SWEET 16 RTN instruction returns control to the 6502. If you can tolerate indefinate 6502 register contents upon exit, approximately 30 usec may be saved by entering at SW16 + 3. Because this might cause an inadvertant switch from Hex to Decimal mode, it is advisable to enter at SW16 the first time through. After saving the 6502 registers, SWEET 16 initializes its PC (R15) with the subroutine return address off the 6502 stack. SWEET 16's PC points to the location preceding the next instruction to be executed. Following the subroutine call are 1-,2-, and 3-byte SWEET 16 instructions, stored in ascending memory like 6502 instructions. the main loop at SW16B repeatedly calls the 'execute instruction' routine to execute it. Subroutine SW16C increments the PC (R15) and fetches the next opcode, which is either a register op of the form OP REG with OP between 1 and 15 or a non-register op of the form 0 OP with OP between 0 and 13. Assuming a register op, the register specification is doubled to account for the 3 byte SWEET 16 registers and placed in the X-reg for indexing. Then the instruction type is determined. Register ops place the doubled register specification in the high order byte of R14 indicating the 'prior result register' to subsequent branch instructions. Non-register ops treat the register specifcation (right-hand half-byte) as their opcode, increment the SWEET 16 PC to point at the displacement byte of branch instructions, load the A-reg with the 'prior result register' index for branch condition testing, and clear the Y-reg. When is an RTS really a JSR? ---------------------------- Each instruction type has a corresponding subroutine. The subroutine entry points are stored in a table which is directly indexed into by the opcode. By assigning all the entries to a common page, only a single byte to address need be stored per routine. The 6502 indirect jump might have been used as follows to transfer control to the appropriate subroutine. LDA #ADRH ;High-order byte. STA IND+1 LDA OPTBL,X ;Low-order byte. STA IND JMP (IND) To save code, the subroutine entry address (minus 1) is pushed onto the stack, high-order byte first. A 6502 RTS (return from subroutine) is used to pop the address off the stack and into the 6502 PC (after incrementing by 1). The net result is that the desired subroutine is reached by executing a subroutine return instruction! Opcode Subroutines: ------------------- The register op routines make use of the 6502 'zero page indexed by X' and 'indexed by X direct' addressing modes to access the specified registers and indirect data. The 'result' of most register ops is left in the specified register and can be sensed by subsequent branch instructions, since the register specification is saved in the high- order byte of R14. This specification is changed to indicate R0 (ACC) for ADD and SUB instructions and R13 for the CPR (compare) instruction. Normally the high-order R14 byte holds the 'prior result register' index times 2 to account for the 2-byte SWEET 16 registers and the LSB is zero. If ADD, SUB, or CPR instructions generate carries, then this index is incremented, setting the LSB. The SET instruction increments the PC twice, picking up data bytes in the specified register. In accordance with 6502 convention, the low-order data byte precedes the high-order byte. Most SWEET 16 non-register ops are relative branches. The corresponding subroutines determine whether or not the 'prior result' meets the specified branch condition and if so, update the SWEET 16 PC by adding the displacement value (-128 to +127 bytes). The RTN op restores the 6502 register contents, pops the subroutine return stack and jumps indirect through the SWEET 16 PC. This transfers control to the 6502 at the instruction immediately following the RTN instruction. The BK op actually executes a 6502 break instruction (BRK), transferring control to the interrupt handler. Any number of subroutine levels may be implemented within SWEET 16 code via the BS (Branch to Subroutine) and RS (Return from Subroutine) instructions. The user must initialize and otherwise not disturb R12 if the SWEET 16 subroutine capability is used since it is utilized as the automatic return stack pointer. Memory Allocation: ------------------ The only storage that must be allocated for SWEET 16 variables are 32 consecutive locations in page zero for the SWEET 16 registers, four locations to save the 6502 register contents, and a few levels of the 6502 subroutine return address stack. if you don't need to preserve the 6502 register contents, delete the SAVE and RESTORE subroutines and the corresponding subroutine calls. This will free the four page zero locations ASAV, XSAV, YSAV, and PSAV. User Modifications: ------------------- You may wish to add some of your own instructions to this implementation of SWEET 16. If you use the unassigned opcodes $0E and $0F, remember that SWEET 16 treats these as 2-byte instructions. You may wish to handle the break instruction as a SWEET 16 call, saving two bytes of code each time you transfer into SWEET 16 mode. Or you may wish to use the SWEET 16 BK (break) op as a 'CHAROUT' call in the interrupt handler. You can perform absolute jumps within SWEET 16 by loading the ACC (R0) with the address you wish to jump to (minus 1) and executing a ST R15 instruction.