Spent yesterday documenting an hidden register in the Sega Dreamcast that allows for multi-target rendering on a single frame. KOS, the homebrew SDK for the Dreamcast, doesn't expose this functionality and the register was previously a magic number offset during init in another generally undocumented base register pointer. Bit 31 at 0x005F8160 when set to 1 enables the TA_LIST_CONT register in the PVR, which tells it to continue rendering onto a second object list in the tile accelerator. This means in order to do this, you must bin an entirely second object list for these different target if the target is going to render something different from the previous target. You pass the Tile Accelerator TSP setup a parameter with a pointer location to the base of your object lists. Object lists must be back to back in memory in an array, so when the PVR finishes one list, it'll move onto the next. You can do as many of these lists as you have room for in memory. Since KOS doesn't expose this normally, you can't do this kind of rendering using any of the KOS calls, so you have to bit bang the hardware directly through communication ports. That's what I was working on yesterday, patching KOS to expose this so anybody can use it in any homebrew application. Example of multipass rendering to create an HDR-like bloom:
Doing reading about how the PVR itself works from various documents. There was a demo long long ago, before KOS was even a thing, which showed off multipass rendering but it was uncommented and wouldn't run with modern libraries. Rewriting the demo so itd compile again and commenting the code lead to exploring a bunch of magic numbers used in the source code, and digging through them all and the PVR manual showed how it all worked, so I could modify KOS to support the feature.
Wow. Cool beans. Spoiler Well, that’s that then. I’m gonna go make my own Sonic ROM hack, with blackjack and hookers! In fact, forget the Sonic ROM hack! Ah, screw the whole thing. Spoiler This also reminds me of this post here: https://sonicresearch.org/community/index.php?threads/a-new-68k-assembler.4179/
