applepy/cycle_notes.txt

There are two ways we could represent cycle information:

1. just have an array of cycles for each opcode + the adjustment for
   page boundary crossing (which seems opcode-specific, oddly)

2. add cycles in individual methods like those accessing memory and the 
   operations themselves, to model *why* something takes the cycles it does.

I prefer 2 on the grounds of it being more instructive but it assumes that
the way we do things is closely aligned to the way the 6502 is doing them
internally. Even if we end up having to do 1, I'd love to understand and
document some of the "why".

What follows is an attempt to "find the patterns" in the cycle times (as
given on http://www.6502.org/tutorials/6502opcodes.html )

NOTE: there appears to be an error in AND and ORA zero page timings on that
webpage given above. I've now corrected this below.

There are 10 classes of instructions when it comes to cycle times:


Class I
(followed by ADC, AND, BIT, CMP, CPX, CPY, EOR, LDA, LDX, LDY, ORA, SBC, STA,
STX, STY)

immediate       2
zero page       3
zero page, x    4
zero page, y    4
absolute        4
absolute, x     4 (+1 if page crossed or writing)
absolute, y     4 (+1 if page crossed or writing)
indirect, x     6
indirect, y     5 (+1 if page crossed or writing)

Note 1: the zero page indexed and x-index indirect don't have the page cross
addition because they wrap.

Note 2: writes to indexed non-zero-page memory (e.g. STA) have the +1 even
if not page crossing.


Class II
(followed by ASL, DEC, INC, LSR, ROL, ROR)

implied         2
zero page       5
zero page, x    6
absolute        6
absolute, x     7

Note 3: these take 2 cycles longer than Class I because they involve a
read-modify-write. Because the absolute, x is a write to an indexed
non-zero-page memory location, there is an additional +1 even if not page
crossing (see Note 2)


Class IIIa
(followed by CLC, CLD, CLI, CLV, DEX, DEY, INX, INY, NOP, SEC, SED, SEI, TAX,
TAY, TSX, TXA, TXS, TYA)

implied         2


Class IIIb
(followed by PHA, PHP)

implied         3


Class IIIc
(followed by PLA, PLP)

implied         4


Class IIId
(followed by RTI, RTS)

implied         6


Class IIIe
(followed by BRK)

implied         7


Class IV
(followed by BCC, BCS, BEQ, BMI, BNE, BPL, BVC, BVS)

branch not taken    2
branch taken        3 (+1 if page crossed)


Class V
(followed by JMP)

absolute        3
indirect        5


Class VI
(followed by JSR)

absolute        6



This seems a possible implementation (not yet considering page boundaries):

1. all instructions start with 2
2. absolute (including absolute indexed) adds 2
3. absolute indexed adds an additional 1 *if* instruction is of RMW type
4. zero page (including zero page indexed) adds 1
5. zero page indexed adds an addition 1
6. indirect (JMP) adds 4
7. indirect x adds 4
8. indirect y adds 3 plus an additional 1 *if* instruction is of RMW type
9. ASL, LSR, ROL, ROR add 2 cycles if not implied
10. JMP subtracts 1 cycle
11. JSR adds 2
12. RTS adds 4
13. branches add 1 if taken
14. DEC and INC add 2 cycles
15. PHA and PHP add 1 cycle
16. PLA and PLP add 2 cycles
17. BRK adds 5 cycles
18. RTI adds 4 cycles

RMW instructions are the absolute,x of ASL, DEC, INC, LSR ROL, ROR and STA
as well as indirect,y and absolute,y of STA


It may be possible to simplify even further given the particular functions
some of these share in command (where the cycle change could be placed
instead)