Doesn't render black pixels when CPU is changing palette while PPU is reading it about sameboy

[LIJI32]

Not only this is extremely interesting, it is hard to describe how useful it is for T-cycle accurate measurements of Mode-3. Do you know which DMG CPU is it? Does it pass dmg_sound-2? Is this effect visible in Prehistorik Man's title screen? It does pretty much the same thing. Does this occur when you use different values (e.g. write the same value twice, write two different but non-inverted palettes)? This is so unfortunate I only have a CGB and SGB2 (which is effectively an MGB).

[mattcurrie]

Ok so I've got 2 DMG's here and both exhibit this same behavior:

Red:
Passes dmg_sound-2
Mainboard: DMG-CPU-08

Blue:
Fails test 10 on dmg_sound-2
Mainboard: DMG-CPU-06

Writing the same palette value during mode 3 doesn't cause any of the black pixels. Will try some more tests with different non-inverted palettes.

I've recorded some video of Prehistorik Man and it looks like the black edges are just slightly visible on the edges of the large text on the title screen:

Different alignment for the large text and no black edges on backlit CGB:

[LIJI32]

I meant the actual revision of the CPU chip (It should say something like DMG-CPU B) but if red one passes dmg_sound-2 it means it's a DMG-CPU B (or the equivalent blob variant). My guess is that the value read by the PPU in that moment is the result of a bitwise AND operation between the old and new version. It is possible that other PPU registers are affected by this quirk, I'll have to take these all into account when I implement PPU-CPU conflicts properly.

[mattcurrie]

Ah yep, so I've now opened up the DMGs:

Red:
Passes dmg_sound-2
Mainboard: DMG-CPU-08
CPU: blob (?)

Blue:
Fails test 10 on dmg_sound-2
Mainboard: DMG-CPU-06
CPU: DMG-CPU B

[mattcurrie]

Very nice :) I've updated the test rom to do transitions between all colour combinations. Shows that ORing the values is correct, as you've done.

m3-palette-change-movable-sprite.gb.zip

I noticed an off-by-one error at the start of the line. It's a bit hard to see in the photo due to shadows from the pixels, but there should just be a single column of pixels with the original palette before the column of pixels with the conflict.

[LIJI32]

That's... annoying. This means I have to reverify a lot of T-cycle timings on an actual DMG.