3D Shader by Werbad Issues in DX9+

It's probably not you. I just tried the GTA building examples that I made years ago using Werbad's shader, which definitely DID work, but are now all broken. Clickteam must have changed something. Whatever they did, it seems to have broken the "Quad deform" shader that I made as well. In both cases, Fusion seems to display only one of the two triangles needed to draw a quad (curiously, it looks okay in the frame editor, but not at runtime).

Shaders don't work in standard mode - you need to use at least DX9 for them to have any effect, and DX11 requires you to use specially converted shader files.

Hmm yeah, I've noticed this a while ago...
Thought it would have something to do with the PS_3_0 which has a broken implementation, but I guess it's something else...
Would be really cool if we get this fixed somehow

Ah, one thing Sketchy / MuddyMole works for me...

MuddyMole I've compared your Quad.mfa with a copy of MMF2 HWA from 2012 and CF 2.5 builds 291 and 293, no difference. Not surprising as AFAIK we didn't change anything it the DX9 engine.

Maybe something was modified in DirectX itself in the meantime.

Here's a couple of examples of 3D Shader that seems broken, perhaps it can help identifying the issue...

Was this always like this?
That's how it looks on MMF2...
I'm sure Quad shader looks a lot better...

Please login to see this picture.

But it's concerning that a shader with nothing really special broke.
Makes me think if some of the stuff I've tried and didn't worked was because of the same issue.

Edit: the math on this other examples seems better, but still got a few issues on MMF2...
Please login to see this link.
Please login to see this picture.

That's strange...

I've just had another look though.
The latest version (which I decided to call "QuadPlus" for some reason) doesn't work - Please login to see this link.

The two earlier versions - "Quad" (w/o lighting) and "QuadDeform" (w lighting) - ARE both still working.

So, that would be an option for SHINGEN to use instead of Werbad's shader (you use it differently, but it can produce exactly the same results)
Please login to see this link.

I don't recall Werbad's shader being particularly complex though, so maybe it just needs tweaking somehow...

Quote from NaitorStudios

Was this always like this?
That's how it looks on MMF2...
I'm sure Quad shader looks a lot better...

Please login to see this picture.

But it's concerning that a shader with nothing really special broke.
Makes me think if some of the stuff I've tried and didn't worked was because of the same issue.

Edit: the math on this other examples seems better, but still got a few issues on MMF2...
Please login to see this link.
Please login to see this picture.

Nope. It's broken. The wall objects are being positioned and resized correctly, but that's not what the shader was for - the shader was to make the corners look right.

Have a look at this example that I built as an .exe back when Werbad's shader still worked:
Please login to see this link.

It still works fine for me, whereas the .mfa doesn't.

Really weird considering MMF2 isn't working correctly as well while the old built EXE behaves differently...

Okay, soooo...
I just tested Werbad's shader on its own, and it actually works fine.
Which means that the problem is with the events - and I think I've found it!
The issue is with passing integers as parameters that should be floats.

Whenever there's expression to set a shader parameter, like this:
( ( Y Top( "Roof" ) - Min(Y Top( "Floor" ), Y Top( "Roof" )) ) + ( _Height( "Roof" ) * 50 ) ) / ( Max(Y Bottom( "Floor" ), Y Bottom( "Roof" )) - Min(Y Top( "Floor" ), Y Top( "Roof" )) )

...if you force Fusion to use floats (eg. replace "50" with "50.0"), it will work.

This never used to be an issue, but apparently now is.

eg.
Please login to see this link.

Oh, this is good to know, but kinda of weird how something changed in the events programming causing this...

Edit: I still get a few missing triangles tho, like it was my MMF2 on the example above
But that buildings one works great with the .0
QuadPlus still broken tho

There is a difference with Quad shader between dx9 and dx11 where dx9 will crop pixels out of range and dx11 will bend the pixels maybe a difference in the sampler settings

Quote from fnkycoldmadeanr

There is a difference with Quad shader between dx9 and dx11 where dx9 will crop pixels out of range and dx11 will bend the pixels maybe a difference in the sampler settings

It shouldn't actually have anything to do with sampler settings. The way it checks for the pixel being within the quad is this part:

float a = (xA - texCoord.x) * (yB - texCoord.y) - (xB - texCoord.x) * (yA - texCoord.y);
float b = (xB - texCoord.x) * (yC - texCoord.y) - (xC - texCoord.x) * (yB - texCoord.y);
float c = (xC - texCoord.x) * (yD - texCoord.y) - (xD - texCoord.x) * (yC - texCoord.y);
float d = (xD - texCoord.x) * (yA - texCoord.y) - (xA - texCoord.x) * (yD - texCoord.y);
if (sign(a) == sign(b) && sign(b) == sign(c) && sign(c) == sign(d)) {

(a fairly standard algorithm to check if a point lies within an arbitrary convex quadrilateral)

I forget why I did it that way now (it was a very long time ago). It would seem simpler to just check l and m are between 0 and 1, or there's probably even a sampler setting for it, like you say.

getColor = tex2D(img, float2( l, m ));

Sorry to resurrect an old thread, but is there a way to have it so that it reads from floors that are already on the map, instead of having to load in an external .arr file?