I understand, all of instruction sets have their shortcommings, which are especially annoying if you code in assembly and can't figure out a clean and efficient way to implement something. I'm not experienced with PIC assembly so I can't tell how often this happens on PIC. However, on 6502, it sometimes happened to me that I badly needed a third Z index register or something to do some function, but instead I had to constantly save/restore X/Y to different values, which is a bit annoying and wastes a lot of instructions.


This sound painful to code for in assembly !! Why didn't they add the hardware to detect such cases and delay the next instruction ? Especially since this is needed anyway even for non-memory instructions.

You're supposed to use the following instruction slot to do something useful, so you don't want to delay it. In the case where you can't figure out anything to fill the slot with you insert a NOP. At least I believe that's how it's done on MIPS (PIC32) too, as it's quite common with pipelined architectures. What I have more problems with understanding is how MIPS can get away with not having a carry flag..

-Tor

Carry flag isn't needed much in a 32 bit machine and it's possible to calculate it. I wish they had an instruction to calculate the carry and place it in a register though.

_________________
http://www.finitron.ca

You only delay the instruction if it uses the same register that just got loaded. Not only this makes the pipeline transparent to the programmer, but it also minimizes ROM usage for your program, as if you can't fit anything useful, you don't have to insert a NOP.


I'm no expert, I think MIPS exposes delay slots after branches, but I'm pretty sure instructions dependencies are handled correctly. I might be wrong though.


For shift/rotates just check the value before the shift/rotate operation.

For additions, if and only if the result is smaller than any of the operands (unsigned comparison) then a carry has happened.

I calculate carry in my emulators but it's an expensive operation.. but if there are only 32-bit additions to worry about then the impact isn't that big presumably. Anyway we're heading over into anycpu.org territory.

-Tor

In my PIC32 emulators for the 8080, 6800 and 65C02 I use lookup tables as I have loads of emply flash program memory. Saves a lot of time computing overflow and half-carry

_________________
Andrew Jacobs
6502 & PIC Stuff - http://www.obelisk.me.uk/
Cross-Platform 6502/65C02/65816 Macro Assembler - http://www.obelisk.me.uk/dev65/
Open Source Projects - https://github.com/andrew-jacobs