Could Dreamcast et al handle this/that game/effect? *DC tech retrospective *spawn

[Cloofoofoo]

What you describe is still impressive and makes me just as curious. Whole development houses avoided using it even though it was in the documentation because it was too difficult. Yet only handful of people managed the impossible port by taking advantage of it. What else is in documentations that devs barely used or not used at all? And how did a handful of people in their spare time manage such incredible feats when devs avoided such hardware features? There is a similar case also with PS2 where devs barely took advantage of one of the VUs

Well the Dreamcast wasn't worth investment in most companies eyes seems to be the most clear cut answer really, in a sec I'll mention a similarity it would have with the PS2 had the DC same work/budget/ time.

For ninja docs I noticed that mentioned ocram/ocindex / ocache modes but guess . It states don't touch these when using the ninja API. Iam guessing naomilib isn't any better. For kamui I am guessing they didnt think it was worth the effort going all super low level cutting your CPU cache in half and using the other half as a really fast scratch pad for matrix/ vector math ( useful for tnl, physics) and having to manage all manually. So c/c++ with some ( at most) assembly is their choice.

So what did these guys do? They decided to use the scratch pad method and do up to 128 verts ( if I remember) in that 8kb to do the tnl math and just jam to the GPU as quick as possible . Falco seems to have taken a bit further and realized you can do the same and accelerate the physics too. Mind you this guys are technically at a disadvantage because they don't have the use sh4 pipeline simulator that gives you exact readout of how the code runs and where it cache misses( professional devs had this!) . They had to make their own homebrew version that isn't 100% accurate. This is the similarity I was speaking of with the PS2 . Using its vector co-processor to assist in fast math while using its cache as fast memory( in this case dc only had its fpu) and shove it to the GPU as quickly as they can.

You asked where are the professionals and why weren't they using this? It seems besides the lack of want to push/fund the dc it was also lack of time. I was reading in their gta3/dc3 chat that they found some homebrew dc code from a former DC dev who was pushing a lot of polygons ( something like 3.4 mpps at 60 fps). Upon inspection it seems he was doing a lot of the same tricks they were doing on gta3/vc to accelerate the matrix math ( But this demo was from 2001) but written fully in assembly.

Why does this matter? Because this was written by software engineer who was responsible for the graphics portion of the Dreamcast cancelled game soul reaver 2. Seems given enough time and genuine interest to push/fund the machine they would have gotten there past 2001.