New Blitter cp2

[sandruzzo]

I all,

I dont' know if this technique was investingated before, and maybe left down. I only was looking to find out an new way to do so. This way is not meant to be fast, maybe it'll be, who knows.

As we know cp2 is a matter of rotation, and we can do it with blitter very easily. In my humble opinion the real "problem" is porelly design hw, that are only capable to works in x direction: Why they did design hw able to work good even in y direction?

Anyway, fisrt we have to set our planes pointer to interleaved way, which is some sort of y chuncky buffer, and then we have to map into them our chuncky pixel into the same direction: how to do that?

With blitter, we can map 4,(4bits planes) pixels at ones, or 2, (8bits planes) pixel, just by using line drawing! Since blitter is able to map a patter into line, we have to load BLTADAT with our chuncky pixels, ad draw vertical line 16 pixels height. That's all!

I dont' know if this way can be fast, maybe by using blitter + cpu + copper, someone could get the fastest possible cp2, or maybe we can do it with only cpu. Like I said, this technique it's meant only yo explore new way to do c2p

Just my 2 cents

[britelite]

Originally Posted by sandruzzo

Like I said, this technique it's meant only yo explore new way to do c2p

So, when can we see it in action?

[sandruzzo]

Originally Posted by britelite

So, when can we see it in action?

So far it's on test, and not optimized. Since Blitters' line way, its possible to do 2x1 very easly!

[chb]

Originally Posted by sandruzzo

With blitter, we can map 4,(4bits planes) pixels at ones, or 2, (8bits planes) pixel, just by using line drawing! Since blitter is able to map a patter into line, we have to load BLTADAT with our chuncky pixels, ad draw vertical line 16 pixels height. That's all!

That will very likely be awfully slow compared to the classic blitter c2p. But no harm in trying, though you may want to use BLTBDAT for the pattern instead, even if BLTADAT might give some funky results.

[sandruzzo]

Originally Posted by chb

That will very likely be awfully slow compared to the classic blitter c2p. But no harm in trying, though you may want to use BLTBDAT for the pattern instead, even if BLTADAT might give some funky results.

My fault, was BLTBDAT. Like I said, just to try an exotic way to do cp2!

[hooverphonique]

IIRC, somebody has mentioned doing chunky->planar conversion using blitter line mode here on EAB before..

[britelite]

Originally Posted by sandruzzo

Since Blitters' line way, its possible to do 2x1 very easly!

Indeed, it's just a shame that easy doesn't always equal fast

[sandruzzo]

Originally Posted by britelite

Indeed, it's just a shame that easy doesn't always equal fast

Maybe we can do it very fast by using only cpu... 2x1 would be almost equal ad 1x1 or better, since blitter line mapping

[britelite]

Originally Posted by sandruzzo

Maybe we can do it very fast by using only cpu... 2x1 would be almost equal ad 1x1 or better, since blitter line mapping

Especially when using the CPU you don't want to be setting individual bits at a time, but rather use large chunks (like the cpu c2p's are already doing).

[sandruzzo]

Originally Posted by britelite

Especially when using the CPU you don't want to be setting individual bits at a time, but rather use large chunks (like the cpu c2p's are already doing).

Maybe coupling this line mapping with Ray casting could speed up all!

[britelite]

Originally Posted by sandruzzo

Maybe coupling this line mapping with Ray casting could speed up all!

Sorry, but it won't

[chb]

Originally Posted by britelite

Indeed, it's just a shame that easy doesn't always equal fast

I'd even bet that if you'd substitute 0x8000 in BPLADAT by 0xC000, 2x1 would come for free. But it's very likely still slow as hell.

[sandruzzo]

Originally Posted by chb

I'd even bet that if you'd substitute 0x8000 in BPLADAT by 0xC000, 2x1 would come for free. But it's very likely still slow as hell.

It take 8 cycles for pixel, so to map 4 complete 4bits pixel we need 16*8 = 128 cycles

[britelite]

Originally Posted by sandruzzo

It take 8 cycles for pixel, so to map 4 complete 4bits pixel we need 16*8 = 128 cycles

Which is way too slow. For a 160x100 pixels screen this would already amount to 512000 cycles, which is over 3 frames, and this is just the conversion. So unfortunately it's way slower than currently used methods.

[sandruzzo]

Just uploaded in the zone! my demo. Feel free to use it, and if someone need help, I'm here!

[britelite]

Originally Posted by sandruzzo

Just uploaded in the zone! my demo.

How about also making an executable for us lazy ones?

[sandruzzo]

Originally Posted by britelite

How about also making an executable for us lazy ones?

ok.

[sandruzzo]

Originally Posted by britelite

How about also making an executable for us lazy ones?

Done. Exe plus all files. If you need more, let me know

[britelite]

Originally Posted by sandruzzo

Done. Exe plus all files. If you need more, let me know

Alright, only tried it in WinUAE, and the executable crashes on a standard A500-configuration. It did work with an expanded conf though, so I was able to run it. I must admit I was hoping to see at least something move

I hope this was a good exercise for you, as you probably now see why this isn't a very efficient way of doing a c2p conversion. You should probably leave this approach alone now, and for example try making a more traditional blitter c2p next.

[sandruzzo]

Originally Posted by britelite

Alright, only tried it in WinUAE, and the executable crashes on a standard A500-configuration. It did work with an expanded conf though, so I was able to run it. I must admit I was hoping to see at least something move

I hope this was a good exercise for you, as you probably now see why this isn't a very efficient way of doing a c2p conversion. You should probably leave this approach alone now, and for example try making a more traditional blitter c2p next.

Like I said, I was curious to try it, and explore new way to do stuff. I like to do that, like I did with sprites' parallax on Rygar, which is the fastest way to do on Amiga ocs/ecs.