Corrected shaders for WinUAE

[Snake79]

Originally Posted by guest.r

If you don't want to bother with it (_winuae.fx), try the masks from this post:
http://eab.abime.net/showpost.php?p=...&postcount=151
(download and unzip into plugins\masks\ ). I think Retro-Nerd is using crt-mask6 for his screenshots.

It was masks that i was missing...
Great work Thank You!

[Snake79]

Originally Posted by Retro-Nerd

Nice, but i have some sound crackles with D9D in combination with your shaders+Beam Racing. Yeah, hopefully it is fixable under D3D11.

I have the same issue, BUT even without filters/masks (sound crackles/speedup/slowdown, video accelerate/decelerate, depending of the Beamrace #number)...
Edit: I think there's something wrong with Beamracing (not filters/masks), with Standard VSync everything is working ok...

[Retro-Nerd]

It's 100%+ CPU usage in D3D9 in my case. Just checked it. D3D11 runs with app 55% usage with the same settings.

Edit: found a combination where i got this in D3D9 working, but app. 85% CPU usage is not acceptable. Looks nice though.

CRT-Trinitron-Lite-LoRes.fx




CRT-Trinitron-Lite-LoRes.fx+D50-D65-WarmerColors.fx

[Toni Wilen]

Originally Posted by guest.r

D3D11 multipass is supported, but still has some bugs, like only enlarged top left part of the image is shown (with passes before scaling (-1,...). Maybe Toni will fix it someday.

Yeah, very distant day if this is the only available information..

[guest.r]

Hi Toni, thanks for addressing it.

I tested it a little more and i found some D3D9 bugs/features too.
Used the latest beta from the beta thread.

1. If a mask is selected (standard _winuae.fx) and we try to apply a shader under the -1 pass, then the screen freezes or. becomes black. Tried it with quickstart A500 common configuration and i could reproduce it.
Log says:
D3D9 (or D3D11): Texture with SOURCETEXTURE semantic not found

Happens with D3D9 and D3D11.
Interesting is that under passes 1 or more the bug doesn't occour.

2. D3D9: The uniform variable TEXELSIZE seems to be affected by the (Y) scale multiplier (logs didn't say if the texture size is adjusted).
I noticed this with a shader that applies scanlines (under 0 pass). The scanlines look normal without the -1 pass, but get 2x denser if the -1 pass is applied with 2x multiplier etc.. Don't know if this is more a feature, but i think passes before scaling are best utilized witout being influenced by the scale multiplier (1:1 scaling). Could not check this under D3D11 yet properly, but there seems to be a change that comes with scaling too.

3. D3D11: if a shader is put under the -1 pass (or pass 1...), the effect gets applied, but the result is only showing very enlarged top left part of the texture. I could reproduce it with the quickstart A1200 configuration, changing to lores and single line mode. Strangly the default configuration (hires, double line mode) works without the "top-left" bug. Happens with my other setups even using hires.

I'll include 2 "test" shaders for the case, the "D50-D65-WarmerColors.fx" is for non 0 passes since it shows some effect - one can see when it's applied, and the CRT shader is to test the scanlines (point 2. ).

Have not tested this until recently and it's not a crittical issue i guess...

[Toni Wilen]

That wasn't very clear.. I am still not sure which test shader does what or what it shouldn't do..

And what shader is required for case 1? (Don't just say "use shader that does this and that")

[guest.r]

Sure. Even i found the bug report a bit overwhelming.

I will concentrate a bit on the "case 1" first. For non 0 (i.e. -1, 1) passes it's OK to use the "D50-D65-WarmerColors.fx" as a test shader, because it's a simple shader that works under D3D9/11 and shows an effect.

If i set the shaders/masks up like in the attached picture, it will result a black screen and the log window will get spammed (both D3D9 and D3D11).
Logs show an error:

Texture with SOURCETEXTURE semantic not found

I could use any other compatible shader under -1 or select any other mask with same results.

Thanks again Toni for giving it a thought...

[Toni Wilen]

Ok.. I am not anymore sure what shader specific masks should do (non-zero shader position). Each position also has mask slot but there was not handling for mask (which caused the not found error and in some situations it was used as main mask.. This is fixed now). Perhaps it was supposed to make select mask available for the custom shader?

But in that case zero position shader would not see the mask or both shader and winuae.fx would draw it which isn't good solution either. Should winuae ignore winuae.fx mask handling if shader takes mask parameter?

Filter GUI fixed, mask selection was not updated when shader position was changed. This got broken in 410b4.

[guest.r]

Thanks Toni, i have tested the new version and it shows very good default behavior. It makes sense to apply the mask only once since _winuae.fx application, as i recall, takes place after the main shader "event".

Could you also take a look again at the "case 3" problems?
It happens (see picture) when delegating a shader to a non-zero position with D3D11. The shader gets applied, but there seems to be a scaling/coordinate bug. It seems to be caused if scale multiplier is greater than 1.
Only enlarged top-left part of the scene is displayed.

I'm slowly moving my defaults to D3D11/beamracing and it's an annoying critter...

[Toni Wilen]

Originally Posted by guest.r

Thanks Toni, i have tested the new version and it shows very good default behavior. It makes sense to apply the mask only once since _winuae.fx application, as i recall, takes place after the main shader "event".

Yeah but do you want to optionally handle the mask rendering in your shader? If you need more flexibility etc..

Could you also take a look again at the "case 3" problems?
It happens (see picture) when delegating a shader to a non-zero position with D3D11. The shader gets applied, but there seems to be a scaling/coordinate bug. It seems to be caused if scale multiplier is greater than 1.
Only enlarged top-left part of the scene is displayed.

Could you attach config file? (windowed mode)

[guest.r]

Originally Posted by Toni Wilen

Yeah but do you want to optionally handle the mask rendering in your shader? If you need more flexibility etc..

Current behavior is very fine with me with default masks. Filtershaders can't do proper masks on their own here in general, since it's incompatible with resolution scaling and post resize options (aspects, autoscaling, integer scaling...). Shaders like _winuae.fx are a proper and good way to handle masks here (even got some calculated masks working with it ).

I will make a post about it somewhat later (with best hopes).

Could you attach config file? (windowed mode)

Yea, sure, it's my default D3D11 config.

Thanks for your time again.

[Retro-Nerd]

I noticed a problem with the multi shader option. If i put something in "1" and sideload another shader in "-1", save and reload the config, the latter always ends up at "-4". I'm pretty sure that didn't happen some month ago while i was testing the "GDAPT-Lores.fx" shader.

Hi guest.r, I remember you programmed an .fx version of CRT-Lottes for Dosbox Daum. We talked about that shader in a thread in the FS-Uae section, more than a year ago.
I really cannot remember if you made a Lottes fx version for WinUae too...
Anyway, CRT Easymode Halation looks pretty good already. :-)

[guest.r]

Originally Posted by Diduz76

Hi guest.r, I remember you programmed an .fx version of CRT-Lottes for Dosbox Daum. We talked about that shader in a thread in the FS-Uae section, more than a year ago.
I really cannot remember if you made a Lottes fx version for WinUae too...
Anyway, CRT Easymode Halation looks pretty good already. :-)

Hey, np, here is the Lottes one. Curvature is disabled by default since some setups don't work correctly with it.

[Toni Wilen]

Originally Posted by guest.r

Yea, sure, it's my default D3D11 config.

Thanks for your time again.

Hmm.. I can see the problem but I have no idea what goes wrong. All the internal parameters seem to be identical.

Have you noticed any wrong shader parameters?

[guest.r]

Originally Posted by Toni Wilen

Hmm.. I can see the problem but I have no idea what goes wrong. All the internal parameters seem to be identical.

Have you noticed any wrong shader parameters?

I have used a simple shader with no coordinates modification (test shader "D50-D65-WarmerColors.fx" above). I guess you could reproduce the bug, since i got it with a quickstart configuration on a different machine with an old nVidia card - i am currently using AMD, so is isn't only AMD related. It's definitely somehow connected with the multiplier.

But i understand these bug hunts are a bit annoying sometimes, reminds me of writing a shader for something with no error logs where the app just crashes on any error.

Main point is, D3D11 shaders work well in general, this would be more or less a bonus with anti-dithering, color altering and smoothing shaders.

So i'm happy even if this currently stands on a low priority list.

Thanks for stopping by again.

Originally Posted by guest.r

Hey, np, here is the Lottes one. Curvature is disabled by default since some setups don't work correctly with it.

Gosh, thanks! For some reason this version of CRT-Lottes isn't as impressive as the one I saw in FS-UAE. Maybe it's my fault, because I use "Automatic resize" with some 320x200 games which don't fill the whole screen. Or maybe shadowmasks are deactivated. I remember CRT-Lottes' colors to be more saturated and vivid.

[Retro-Nerd]

With the current winuae beta from yesterday (21:48 compile time) i get a black screen using every filter or single shader in D3D11. D3D9 works.

The version from the 18.02 (17:20) is still fine with D3D11 though.

[guest.r]

Originally Posted by Diduz76

Gosh, thanks! For some reason this version of CRT-Lottes isn't as impressive as the one I saw in FS-UAE. Maybe it's my fault, because I use "Automatic resize" with some 320x200 games which don't fill the whole screen. Or maybe shadowmasks are deactivated. I remember CRT-Lottes' colors to be more saturated and vivid.

Yeah, it comes without a mask, which contributes to the vivid look. It's a bit different here with WinUAE, since you can select a custom mask from the "masks" options. We're "working" on "HQ" masks here also, it would look authentic lottes then.

Originally Posted by guest.r

Yeah, it comes without a mask, which contributes to the vivid look. It's a bit different here with WinUAE, since you can select a custom mask from the "masks" options. We're "working" on "HQ" masks here also, it would look authentic lottes then.

I remember I also managed to have a more vibrant CRT-Lottes by the "brightboost" parameter, that doesn't seem to work here. I can play with WinUae display settings, and yet it's not the same thing.
Anyway, I'm nitpicking here.