Update PI DMA alignment as per CEN64

[loganmc10]

Taken from: https://github.com/n64dev/cen64/blob/master/pi/controller.c

Tested using: https://github.com/PeterLemon/N64/tree/master/CPUTest/DMAAlignment-PI-cart

Before:

After:

There are still 2 tests that fail here that pass on CEN64: n64dev/cen64#200. However, I couldn't get those tests to pass without breaking the graphics in AI Shougi 3, so I left it at the point where AI Shougi 3 would work. AI Shougi 3 graphics are messed up in CEN64, so I assume there is still more work to be done on their implementation.

[Narann]

Thanks for the hard work. The code is clean, thanks for this!

I can merge this because I suspect no-one will have the required knowledge to validate this and the scope of this flag seems to be limited.

EDIT: I saw some force pushes, tell me when you think the PR is ready for merge. :)

[loganmc10]

Yeah it's ready now, I don't plan on making any further changes

[loganmc10]

Actually hold on for a little bit. @rasky did some hardware tests to figure out how to get the test ROM and AI Shougi 3 working

[Narann]

Haha! :) No worry!

[loganmc10]

Ok I'm happy with this now. The test ROM passes at the same level as CEN64, AI Shougi works. I also tested other games I know that have specific PI alignment requirements (like command & conquer and taz express). Everything seems to be working normally.

This work was based on hardware tests done by @rasky:

As well as tests previously documented here: https://n64brew.dev/wiki/Peripheral_Interface#Unaligned_DMA_transfer

[Narann]

Thanks for the hard work!

I don't know how the compatibility situation is but can you just convert the // by /* */ ?

[loganmc10]

ok done

[Narann]

Thanks! :)

[m4xw]

This caused a regression for 64DD for me.
Reverting pi_controller solves this. No game would boot (stuck on black screen) or it would just enter the IPL.
Didnt check yet which change in particular caused this, just a headsup.

[loganmc10]

No problem, I'll fix this and submit a new pr

[m4xw]

Thanks, if you read discord, my suspicion rn is the pi->regs[PI_CART_ADDR_REG] touching it in some way that doesnt trigger the

      if ((pi->regs[PI_CART_ADDR_REG] == MM_DD_C2S_BUFFER) ||
           (pi->regs[PI_CART_ADDR_REG] == MM_DD_DS_BUFFER)) {

condition for the if(dd) stuff at end of pi event.
Didnt spend time today investigating further tho.

[m4xw]

Forgot to add, as far i can tell from the talks with luigi, we can consider the end of pi dma event stuff for DD a hack. So feel free to rework as you see fit.