Page 1 of 1
Assemler - efficient alternative to self-modifying code?
Posted: Fri Dec 19, 2025 2:04 pm
by pifu
Hi,
I need to assemble a double-indirect jump. I solved it with self-modifying code:
Code: Select all
;SWITCH:
;A:X (lo:hi) = adr, OFS:OFS+1 (lo:hi) = ofs
;jump to address stored in MEM[adr+ofs]:MEM[adr+ofs+1]
SWITCH CLC
ADC OFS ;A <- lo(adr + ofs)
STA @jmp + 1 ;save it as JMP() arg, lo
TXA
ADC OFS + 1 ;A <- hi(adr + ofs)
STA @jmp + 2 ;save it as JMP() arg, hi
@jmp JMP ($FFFF) ;SMC: JMP (adr + ofs)
but that's not "ROMable", so I sketched up this one:
Code: Select all
;SWITCH:
;A:X (lo:hi) = adr, OFS:OFS+1 (lo:hi) = ofs
;jump to address stored in MEM[adr+ofs]:MEM[adr+ofs+1]
SWITCH CLC
ADC OFS ;A <- lo(adr + ofs)
STA TMP ;TMP <- lo(adr + ofs)
TXA
ADC OFS + 1 ;A <- hi(adr + ofs)
STA TMP + 1 ;TMP + 1 <- hi(adr + ofs)
LDY # 1
LDA (TMP),Y
TAX ;X <- MEM[adr+ofs+1]
DEY
LDA (TMP),Y ;A <- MEM[adr+ofs]
STA TMP ;TMP <- MEM[adr+ofs]
STX TMP + 1 ;TMP + 1 <- MEM[adr+ofs+1]
JMP (TMP)
May someone write something better, please?
Re: Assemler - efficient alternative to self-modifying code?
Posted: Fri Dec 19, 2025 3:17 pm
by drogon
May someone write something better, please?
Better? Who knows Might just be different...
One way to make something that's normally ROMmable is to copy it to RAM - if you can spare a few bytes in ZP then it would work just fine - also using ZP addressing you might save a byte or 2 in ROM...
Another way:
If you have a table of addresses, then you can simply push them into the stack and RTS. Note that the table of addresses would be the address-1. From
http://6502.org/tutorials/6502opcodes.html#RTS:
Code: Select all
LDX #1 ; Index to jump to
JSR EXEC
JMP SOMEWHERE
LOBYTE
.BYTE <ROUTINE0-1,<ROUTINE1-1
.BYTE <ROUTINE2-1,<ROUTINE3-1
HIBYTE
.BYTE >ROUTINE0-1,>ROUTINE1-1
.BYTE >ROUTINE2-1,>ROUTINE3-1
EXEC
LDA HIBYTE,X
PHA
LDA LOBYTE,X
PHA
RTS
This would assume a table size of <= 256 entries.
Another way might be possible if you are using a 65C02 and goes along the lines of:
Code: Select all
asl ; Index in A, we double it
tax
jmp (table,x)
table: .word target0
.word target1
.word target2
... etc.
the limitation here is that the table has to be <= 128 entries although checking for > 128 is possible with another line of code or 2. I do this in my 65c816 bytecode interpreter with the advantage there that the index is 16-bits wide so I can accommodate all 256 bytecodes without a secondary check.
-Gordon
Re: Assemler - efficient alternative to self-modifying code?
Posted: Fri Dec 19, 2025 4:04 pm
by pifu
One way to make something that's normally ROMmable is to copy it to RAM - if you can spare a few bytes in ZP then it would work just fine
You right! So I just need to move the JMP() instruction to 3 bytes in ZP and jump there from my first routine... Easy
Many thanks!
Re: Assemler - efficient alternative to self-modifying code?
Posted: Fri Dec 19, 2025 5:03 pm
by teamtempest
Expanding slightly on drogon's answer, you could also use RTS to make the jump this way:
Code: Select all
;SWITCH:
;A:X (lo:hi) = adr, OFS:OFS+1 (lo:hi) = ofs
;jump to address stored in MEM[adr+ofs]:MEM[adr+ofs+1] (must actually be address-1 of target)
SWITCH CLC
ADC OFS ;A <- lo(adr + ofs)
STA TMP ;TMP <- lo(adr + ofs)
TXA
ADC OFS + 1 ;A <- hi(adr + ofs)
STA TMP + 1 ;TMP + 1 <- hi(adr + ofs)
LDY # 1
LDA (TMP),Y
PHA ;STACK <- MEM[adr+ofs+1]
DEY
LDA (TMP),Y ;A <- MEM[adr+ofs]
PHA ;STACK <- MEM[adr+ofs]
RTS
You do still need a couple of bytes of zero page, but you need those anyway. There's no self-modifying code and no need to copy anything from ROM.
Re: Assemler - efficient alternative to self-modifying code?
Posted: Sat Dec 20, 2025 11:30 am
by jgharston
If you have a table of addresses, then you can simply push them into the stack and RTS. Note that the table of addresses would be the address-1.
Or push the absolute address and then jump to it with PHP:RTI