First some code:


My understanding here is that the 16-bit value that's loaded from $8788BA ($83B3) is being pushed on the stack  to be used as the return address. That I follow perfectly, but when the RTS at $8782F9 is hit, instead of returning to $8783B3 like I thought it would, it returns to $8783B4. What the hell? :huh:

I checked the stack and the data at $8788BA and in both places it's $83B3, not $83B4. Just to clarify, I have not modified ANYTHING in this ROM. The thing that's really baffling is that there is an RTS at $8783B3, but if that were being executed the stack pointer should reflect that, which, as you can see above, it doesn't. \

Anyone have any ideas? Or some esoteric knowledge to share that might be of use?

D'oh! I feel really dumb now.  :laugh:

Thanks, Ryusui.

EDIT: Double d'oh! It's Ryusui not Ryusei, isn't it?

A little ambiguous. I'll clarify with the actual behavior for reference.

Here are the three JSR functions:


As you can see, the address pushed onto the stack inside the JSR is actually decremented by one.

Here are the return functions:


The RTS / RTL opcodes increment the value pulled off of the stack. But take a look at RTI, this increment does not happen. Why?


Because the code that's executed when an interrupt triggers does not perform the decrement. This is an important difference, as you can just as easily use rti as a plp + rtl combo if you keep that one little difference in mind.

As for why the SNES does this, I'm really not 100% certain. I believe it is due to the implementation of jsr (addr,x). They likely modified the other two jsr routines so that the same rts / rtl opcodes could be used for all three jsr opcodes. Or maybe it's just to screw with people. It's unfortunately not like the branching opcodes where it's immediately clear why bra $00 branches to the end of that opcode, rather than back to itself. The PC is incremented each time you read in an opcode byte.

---

Now, though Wikipedia will tell you that bsnes is not open source ... if in fact you stop drinking Bruce Perens' Kool-Aid, you'll see that it, in fact, is. I'd strongly recommend downloading it and taking a look at the CPU core when you have questions like this. Not that we don't mind answering, but it'd be a neat way to see how things are really working behind the scenes. I wrote it really cleanly so you should be able to figure out what it's doing without having to study it for a long time or anything.