new amiga 500 doom like !

[Dunny]

Originally Posted by utri007

Any change for a possibility to floor and ceiling textures? That could work with unexpanded Amiga 1200.

Floor and ceiling (in my limited experience) require vastly more CPU horsepower to render than the plain vertical strips that walls use. Mind you, I'd not rule out KK managing to get it done.

[KK/Altair]

Originally Posted by Dunny

Floor and ceiling (in my limited experience) require vastly more CPU horsepower to render than the plain vertical strips that walls use. Mind you, I'd not rule out KK managing to get it done.

Floors and ceilings are the same complexity level as a rotozoomer, so it should be doable on A1200. And if performance becomes a problem, we could do it half resolution and it still would be better than flat shading. Still, this engine manages 22-23 FPS (IIRC) on a stock A1200 and gameplay is so smooth I would think three times before sacrificing it for floor/ceiling textures.

[utri007]

Originally Posted by KK/Altair

Floors and ceilings are the same complexity level as a rotozoomer, so it should be doable on A1200. And if performance becomes a problem, we could do it half resolution and it still would be better than flat shading. Still, this engine manages 22-23 FPS (IIRC) on a stock A1200 and gameplay is so smooth I would think three times before sacrificing it for floor/ceiling textures.

OK, I understand, though many early FPSs had a possibility to enable/disable them floor and ceiling textures. So maybe it is not a big job to add?

[KK/Altair]

Originally Posted by utri007

OK, I understand, though many early FPSs had a possibility to enable/disable them floor and ceiling textures. So maybe it is not a big job to add?

I'm 100% focused on the OCS right now, so small jobs related to it still have priority.

[saimon69]

Originally Posted by KK/Altair

I'm 100% focused on the OCS right now, so small jobs related to it still have priority.

Am going a bit OT but i feel that with little modification to the engine you could even do a racing game with a raycaster like that, possibilities are endless

[VladR]

Originally Posted by KK/Altair

Floors and ceilings are the same complexity level as a rotozoomer, so it should be doable on A1200.

Could you please expand on this a little bit ? Do you mean you would get away with just one LUT table for all view angles ? Or what exactly do you mean ?

[KK/Altair]

Originally Posted by VladR

Could you please expand on this a little bit ? Do you mean you would get away with just one LUT table for all view angles ? Or what exactly do you mean ?

Both rotozoomer and floor/ceiling texturing draw image one scanline at a time by stepping over 2D rectangle texture. The only difference is that rotozoomer uses constant steps for entire image, while floors/ceilings vary steps depending on the distance. But the core per-pixel loop remains exactly the same.

[VladR]

Originally Posted by KK/Altair

Both rotozoomer and floor/ceiling texturing draw image one scanline at a time by stepping over 2D rectangle texture. The only difference is that rotozoomer uses constant steps for entire image, while floors/ceilings vary steps depending on the distance. But the core per-pixel loop remains exactly the same.

I see what you meant, thanks.

This would actually be an interesting benchmark on Vampire - compare the Integer versus FloatingPoint Texel LookUp. I doubt Integer would be slower on 040, but around 060 it's getting blurrier, once Integer ops get to around 2-4 cycles, but FP ones are 4/8/more...

[KK/Altair]

Originally Posted by VladR

I see what you meant, thanks.

This would actually be an interesting benchmark on Vampire - compare the Integer versus FloatingPoint Texel LookUp. I doubt Integer would be slower on 040, but around 060 it's getting blurrier, once Integer ops get to around 2-4 cycles, but FP ones are 4/8/more...

There's completely no point for doing it using FP. Fixed point on integers are exactly what it needs and doing it on fixed point it saves conversion from FP to integers anyway (even no shifting is needed if you apply some tricks).

[VladR]

Originally Posted by KK/Altair

There's completely no point for doing it using FP. Fixed point on integers are exactly what it needs and doing it on fixed point it saves conversion from FP to integers anyway (even no shifting is needed if you apply some tricks).

Well, I agree - not on 060, but on Vampire you can have:
- Primary Execution Unit doing FP calculations
- Secondary Execution Unit doing integer / fixed-point (in parallel)

I recently rewrote a scanline traversal (for flatshader) using FP and it completely removed any RAM access, as scanline traversal was using FP registers and the integer unit was handling switching edges and filling scanline without touching RAM (other than the fill).

FP version of traversal (just that particular component) was slower than fixed-point one, but the grand total time (traversal+scanline fill) was lower, because of the better parallelism with second execution unit.

Of course, such code must be interleaved properly, to remove as many pipeline bubbles as humanly possible, but the end result may be very well worth it.


In this use case, if we were only doing the angled floor fill, without any shading or filtering or fog, it wouldn't make sense to go FP as we wouldn't even be using all integer registers in the first place and FP is going to be slower than integer for this scenario.

[KK/Altair]

Well, I agree - not on 060, but on Vampire you can have...

I'm aiming at A500, so my scope barely reaches A1200. But I definitely won't waste a second optimizing for Vampire. But it should still run well enough - I tested the engine with Furia and it pumped out smooth 31 FPS.

[nikosidis]

KK/Altair

Sounds sweet. I just got a Furia Great, cheap upgrade for my A600.

[hammer]

Originally Posted by swinkamor12

Only 12 FPS at 160x100.
No textures on ceiling and floor, so it is still Wolf not real DOOM.
Waste of time, and yet another proof that for 3D Amiga need graphics with chunky pixels.

[ Show youtube player ]

The demonstration has lower door entrances with an overhang geometry and temple-like geometry structures which doesn't exist in Wolf. KK/Altair's Doom clone geometry is superior when compared to Gloom and Wolf.

[d4rk3lf]

@hammer

You're a bit late for that answer, as that guy is banned, because he constantly trashes other works, without anything then million times repeated " Amiga needs chunky pixels".
For all I care, I don't even bother replying to these trolls, because it gives them fuel.
Constructive criticism is one, negativity and trolling is another thing.

@KKAltar.

Come on man.... tease us!

[waldiamiga]

Originally Posted by hammer

[ Show youtube player ]

The demonstration has lower door entrances with an overhang geometry and temple-like geometry structures which doesn't exist in Wolf. KK/Altair's Doom clone geometry is superior when compared to Gloom and Wolf.

There is nothing to worry about, it's an ordinary troll that has been banned or blocked by dozens of users in Poland.
The sad thing is that one compatriot of mine creates great things and the other can only troll.

[KK/Altair]

" Amiga needs chunky pixels".

The funny fact is, that in my engine the whole chunky issue became second or third class problem now. It's done asynchronously on interrupt-driven Blitter, and half of the problem is already bypassed by lowering the resolution (saving one pass) and a crazy coding trick (saving another pass). So Amiga having chunky pixels will probably save just a fraction of a frame now, out of 4-5 the rendering takes now. Right now the key parts are pixel copying and preparing the scanlines from lines. All CPU work.

@KKAltar.

Come on man.... tease us!

Episode 04 ready on YT as unlisted, dropping tomorrow.
Teased enough?

[roondar]

Originally Posted by KK/Altair

The funny fact is, that in my engine the whole chunky issue became second or third class problem now. It's done asynchronously on interrupt-driven Blitter, and half of the problem is already bypassed by lowering the resolution (saving one pass) and a crazy coding trick (saving another pass). So Amiga having chunky pixels will probably save just a fraction of a frame now, out of 4-5 the rendering takes now. Right now the key parts are pixel copying and preparing the scanlines from lines. All CPU work.

This is very interesting to hear. To be honest I've never been quite clear on the overhead of chunky-to-planar for Amiga. On the one hand it's always considered a big problem, on the other we get talented coders like yourself (and others, though they tend to target higher specced Amiga's) pointing out that the overhead is real, but not as cripplingly so as we've been led to believe.

At any rate, I think your description made it even more clear this is quite the accomplishment. The way you do it can't be easy, it sounds like a fairly complicated affair

Episode 04 ready on YT as unlisted, dropping tomorrow.
Teased enough?

Yeah, that'll do

[kriz]

Ohh yeah new episode tomorrow !!

[nikosidis]

Whow, what a day this will be

[Mathesar]

Originally Posted by KK/Altair

Episode 04 ready on YT as unlisted, dropping tomorrow.
Teased enough?

YESSS!!!
Can't wait